Posts By :

Nadeem Shaikh

Sage green and peach graphic: a reorder checklist (cover, trend, lead time, order) beside the Runway inventory table, GrtLabs logo
Deciding What to Reorder on BigCommerce: A Simple Method
Deciding What to Reorder on BigCommerce: A Simple Method 1024 538 Nadeem Shaikh

To decide what to reorder on BigCommerce, compare each product’s days of cover with how long your supplier takes to deliver, check whether sales are rising or falling over 30, 60 and 90 days, and then size the order to cover lead time plus the period you want to stock for. The data points you to the products that need attention; you make the final call based on cash, space and plans.

This guide gives you a four-step method you can run in about 20 minutes a week.

Why a simple method helps

Guessing works until it doesn’t. Order too little and a best seller runs out, which can cost sales. Order too much and cash sits on the shelf. A repeatable method keeps you out of both traps without needing a complex system.

The four-step method

Step 1: Check days of cover

Days of cover = stock on hand ÷ average daily sales (from the last 30 days). It tells you roughly how many days your stock will last. Sort products from lowest to highest. The top of the list is where to start. (Our low stock guide covers how to set a threshold.) (More detail in our days of cover guide.)

Step 2: Compare with lead time

Lead time is the number of days from placing a purchase order to having stock ready to sell. If a product’s days of cover is close to or below its lead time, new stock may not arrive before it runs out.

Add a buffer (safety stock) for delays or busy weeks. A simple rule many small stores use:

Look closely when days of cover < lead time + buffer

Step 3: Check the trend

Compare units sold over 30, 60 and 90 days (see how to build an inventory report with those columns):

  • Speeding up: the last 30 days is more than a third of the 90-day total. You may want a bit more.
  • Steady: roughly a third each month.
  • Slowing down: the last 30 days is less than a third of the 90-day total. You may want a bit less, or wait.

Also think about anything the numbers cannot see: an upcoming promotion, a holiday, or a product you plan to discontinue.

Step 4: Size the order

A plain-terms way to estimate quantity:

Order quantity ≈ (average daily sales × days you want to cover) − stock on hand

“Days you want to cover” is usually lead time plus how long you want the new stock to last, plus your buffer. Round to your supplier’s case or minimum order size.

A worked example (made-up numbers)

These numbers are an example only.

Your mug supplier takes 10 days to deliver. You want a 5-day buffer and you order roughly once a month (30 days).

Product On hand Sold 30d Sold 90d Avg daily Days of cover
Ceramic mug 20 60 140 2.0 10
Wool scarf 45 30 60 1.0 45

Mug:

  1. Days of cover: 10.
  2. Lead time + buffer: 10 + 5 = 15. Cover is below that, so it needs attention.
  3. Trend: 60 of 140 sold in the last 30 days, more than a third. Sales are picking up.
  4. Size: days to cover = 10 (lead) + 30 (cycle) + 5 (buffer) = 45. Average daily 2.0 × 45 = 90. Minus 20 on hand = about 70 units. If the mug comes in cases of 24, you might choose 72.

Scarf:

  1. Days of cover: 45.
  2. Well above 15, so no rush.
  3. Trend: 30 of 60 in the last 30 days, also rising, but stock is healthy.
  4. You could leave it this week and check again next week.

These are your decisions to make. If cash is tight, you might order 48 mugs now and more later. If a big sale is coming, you might order more. The math is a starting point, not a rule.

Common mistakes to avoid

  • Ignoring stockout days. If a product was out of stock for part of the month, its sales look lower than real demand.
  • Treating no sales as no need. A blank days of cover (no sales in 30 days) means not enough data, not endless stock. Check 60 and 90-day sales.
  • Forgetting minimums. Supplier minimums and case sizes often decide the final number.
  • Checking too rarely. A weekly look catches problems while there is still time to act.

Getting the numbers without a spreadsheet

You can run this method from exported orders and products in a spreadsheet. If you would rather have the numbers ready, Runway is a free, read-only BigCommerce app from GrtLabs. It shows units sold in the last 30, 60 and 90 days, stock on hand and days of cover for each product, with the products that run out soonest at the top. You can filter to low cover or out of stock, open a product in your BigCommerce admin, and download the list as a CSV to work out quantities with your own lead times. It is on the BigCommerce App Marketplace.

To be clear about what it does not do: Runway does not tell you how much to order, create purchase orders, email suppliers or change your stock. It shows the data; you decide and place the order yourself. If you need a tool shaped around your own buying process, our team builds custom BigCommerce apps and integrations.

Quick recap

  1. Sort by days of cover.
  2. Compare with lead time plus a buffer.
  3. Check the 30/60/90-day trend.
  4. Size the order, then adjust for cash, case sizes and plans.

FAQ

How do I know when to reorder a product?

Compare its days of cover with your supplier lead time plus a buffer. If cover is close to or below that number, the product needs attention now.

How much should I reorder?

A simple estimate is average daily sales times the number of days you want to cover, minus stock on hand. Then adjust for case sizes, minimums, cash and upcoming events.

What is lead time?

Lead time is the number of days between placing an order with your supplier and having that stock ready to sell.

What is safety stock?

Safety stock is extra stock, often expressed as a few extra days of sales, kept to cover supplier delays or sudden jumps in demand.

Does Runway create purchase orders?

No. Runway is a free, read-only BigCommerce app. It shows units sold, stock on hand and days of cover, and exports a CSV. You decide what to order and place the purchase order yourself.

Forest green graphic: QuickBooks in ChatGPT and Claude, a draft invoice review checklist ending in Approve, then send, GrtLabs logo
QuickBooks With ChatGPT and Claude: What the Connector Does and What to Check
QuickBooks With ChatGPT and Claude: What the Connector Does and What to Check 1024 538 Nadeem Shaikh

Yes, you can now run QuickBooks Online from ChatGPT or Claude. Intuit’s official QuickBooks connector lets you create estimates and invoices from your live customer and product lists, send them, send reminders on overdue invoices and share payment links, all from a chat. Payments collected through QuickBooks Payments are recorded in your books. The assistant drafts; you review and approve before anything goes to a customer. It is a real time saver for everyday billing, but it does not replace an accounts payable process, approval rules across a team, or connections to your other systems.

This guide is for small business owners, bookkeepers and accounts receivable staff who already use QuickBooks Online. It covers what the connector does, how to set it up, what to check before you send anything, and where a custom AI workflow still makes sense. Feature details below come from Intuit’s own pages, linked in each section. Intuit says the capabilities are expanding, so check its help pages for the current list.

What the QuickBooks connector in ChatGPT and Claude can do

According to Intuit’s What’s new with QuickBooks Online page, you can connect QuickBooks to ChatGPT or Claude to create estimates, email invoices, send automatic reminders and share payment links, with payments recorded in your books automatically. Intuit lists it for paid QuickBooks Online plans, for eligible merchants with QuickBooks Payments turned on.

TaskHow it works in chat
Create an invoiceDescribe it in plain words; the assistant pulls the customer and items from QuickBooks and shows a draft
Edit before sendingAsk for changes in the same chat, such as a new line item or different payment terms
Send and get paidAfter you approve, the invoice goes out with an online payment link and syncs to QuickBooks
EstimatesBuild a quote the same way, then convert it to an invoice when the customer accepts
Follow-upsAsk which invoices are overdue and send a reminder for a specific one
Customers and productsAdd a new customer or product or service from the chat

The key point: an invoice created in the chat is a real QuickBooks invoice, not text you copy over later. That is why the review step matters.

How to connect QuickBooks to Claude or ChatGPT

Intuit’s step-by-step guide describes a short setup for each assistant:

  1. Claude: open Claude, go to Connectors, find Intuit QuickBooks, click Connect, sign in with your Intuit account and approve the connection.
  2. ChatGPT: open ChatGPT, go to Apps, find Intuit QuickBooks, click Connect, sign in with your Intuit account and approve the connection.
  3. Pick the right company if your Intuit login has more than one QuickBooks company.
  4. Grant only what you need where the connection screen lets you choose permissions.
  5. Try it on a test customer first before you send anything to a real one.

Intuit notes that if you need to reconnect, you should disconnect fully, wait a moment, then connect again. Connecting the AI platform may also depend on your plan with that platform, so check your ChatGPT or Claude account settings.

What to check before anything is sent

Intuit says the assistant shows you a draft first and nothing goes out without your approval. Use that step properly. Before you approve an invoice or estimate, check:

  • Customer: the right record, especially when two customers have similar names.
  • Items and quantities: the right product or service from your list, not a new one the assistant created by mistake.
  • Amounts, tax and discounts: match what you agreed with the customer.
  • Terms and due date: your standard terms unless you meant to change them.
  • Email address: the billing contact, not an old or personal address.
  • Memo and description: nothing internal or confidential that the customer should not see.

Give the assistant everything it needs so it guesses nothing: the exact customer name as it appears in QuickBooks, each item, quantity, rate and the terms. Vague requests produce drafts you have to fix.

Who should use it, and how

  • Owner-operators who invoice a handful of customers a week and want it done from the phone or between jobs.
  • Bookkeepers who want quick answers such as which invoices are overdue, then send reminders after a check.
  • Small AR teams who agree on a simple rule: the assistant drafts, a named person approves and sends.

Set a few house rules before the whole team starts using it: who may send invoices from chat, which customers or amounts always need a second look, and how corrections are handled. If a sent invoice is wrong, fix or void it in QuickBooks as you normally would.

Risks to keep in mind

  • Wrong record, real invoice. Because drafts are real QuickBooks records, a mistake that gets approved goes to a customer.
  • Shared access. Anyone who can use the connected chat account can act in your books within the permissions you granted. Do not connect QuickBooks to a shared or personal login.
  • Instructions hidden in content. If you paste emails or documents into the same chat, they can contain text that tries to steer the assistant. Read our guide to prompt injection protection.
  • No team approval chain. The connector is built for one person approving in a chat, not multi-step approval by amount or department.
  • Regions and plans. Availability depends on your QuickBooks plan, your country and QuickBooks Payments eligibility. Check Intuit’s help pages for your account.

Where a custom AI workflow still fits

The connector handles invoices you send. Many finance teams have bigger problems around it:

NeedWhy the connector alone is not enough
Vendor bills arriving by email and PDFThat is accounts payable intake: reading, coding and approving bills. See automate invoice processing with AI.
Matching payments to invoices across bank and processor dataSee AI invoice reconciliation for how matching and exception review work.
Invoicing from a CRM, job system or project toolA workflow can create invoices when a job closes, using rules you set, instead of someone typing a request.
Approval rules by amount, customer or departmentCustom workflows can route drafts to the right approver and log who approved what.
High volume or batch billingRules-based automation is steadier than one chat request at a time.

Our comparison of AI agents vs RPA explains when a rules-based process beats a conversational one. If you are picking a first finance project, start with the one that costs your team the most hours each month.

How GrtLabs helps finance teams use AI with QuickBooks

GrtLabs Inc is a Houston-based AI and software development company, founded in 2019. Through our AI Factory we build AI agents and automations that connect accounting tools like QuickBooks to the rest of your business: bill intake, approval routing, reconciliation exception queues and invoicing triggered from your job or CRM data. Every workflow has human approval steps where you want them and a log of what it did.

Most projects start with AI Discovery, a short engagement that maps your current billing and AP process and picks the first workflow worth automating. Houston businesses can also see our AI development company in Houston page.

Using QuickBooks and still buried in billing work? Talk to our Houston team about an AI workflow built around your books, your approvals and your systems.

Frequently asked questions

Can ChatGPT or Claude create invoices in QuickBooks?

Yes. With Intuit’s official QuickBooks connector, you can create, edit and send invoices and estimates from a chat using your live QuickBooks customers and items. Intuit says the assistant shows a draft and nothing is sent until you approve it.

How do I connect QuickBooks to Claude?

In Claude, open Connectors, find Intuit QuickBooks, click Connect, sign in with your Intuit account and approve the connection. In ChatGPT, the same steps start from Apps.

Does the QuickBooks connector record payments?

Intuit says payments collected through QuickBooks Payments are recorded in your books automatically. Payment links need an eligible QuickBooks Payments account.

Is my QuickBooks data used to train AI models?

Intuit states that QuickBooks data accessed through the connector uses a secure, authenticated connection and is not used to train AI models. Review Intuit’s privacy policy and your AI platform’s settings for full details.

Can the connector process vendor bills?

The connector is aimed at invoicing, estimates, reminders and payment links. Reading and approving vendor bills from email and PDFs is accounts payable work, which usually needs a separate document processing and approval workflow.

When do I need a custom AI workflow instead?

When invoices should be created automatically from a CRM or job system, when drafts need approval by amount or department, when volume is high, or when you need matching and reconciliation across several data sources.

Deep plum and tangerine graphic: AI receptionist answering a call, booking an appointment and handing an emergency to the team, GrtLabs logo
AI Receptionist for Small Business: What It Answers and How Handoff Works
AI Receptionist for Small Business: What It Answers and How Handoff Works 1024 538 Nadeem Shaikh

An AI receptionist for a small business answers your phone or chat when nobody is free, handles the routine questions (hours, location, services, how to book), takes the caller’s details, books or requests an appointment in your calendar, and passes anything urgent, unclear or sensitive to a real person. It works well when it has a short, accurate script of facts about your business and a clear handoff rule. It goes wrong when it is left to guess.

This guide is for owners and office managers at home service companies, clinics, law and accounting firms and other small businesses that miss calls during busy hours or after closing. It covers what an AI receptionist can answer, how booking and handoff to a person should work, what drives the cost to run one, and the mistakes to avoid before you switch it on.

What an AI receptionist actually does

An AI receptionist is an AI agent connected to your phone line, website chat or text messages. It listens or reads, works out what the person wants, answers from information you gave it, and takes a small number of actions in your tools, such as checking open time slots or creating a lead in your CRM. Unlike an old phone menu, the caller just says what they need in their own words.

Usually a good fitUsually hand to a person
Business hours, address, parking and service areaEmergencies and anything safety-related
Which services you offer and how to request a quoteComplaints and upset customers
Booking, moving or cancelling a standard appointmentPricing exceptions, discounts and disputes
Taking a message with name, number and reason for the callMedical, legal or account-specific questions
Collecting job details before a callback (address, issue, photos by text)Anything the script does not clearly answer

The difference between this and a website chatbot is mostly the actions. A chatbot answers questions; an agent can also book, update and hand off. Our AI chatbot vs AI agent guide explains where that line sits.

How booking works

Booking is where an AI receptionist saves the most time and causes the most trouble if it is set up loosely. A safe setup looks like this:

  1. Connect to your real calendar or scheduling tool, not a copy, so it only offers slots that are actually open.
  2. Set booking rules in writing: which services it can book, how long each one takes, lead time, buffer between jobs, and which service areas or locations each team covers.
  3. Confirm details back to the caller (date, time, address, service) before it books.
  4. Decide what it may change. Many businesses let it book and move standard appointments but send cancellations of large jobs to a person.
  5. Log every booking with the call summary in your CRM or job system, so your team sees what was promised.

If your scheduling tool has no way to connect, start with “request an appointment”: the AI collects the details and your team confirms. It is slower but much safer than letting it guess at availability.

How handoff to a person works

The handoff rule matters more than the voice or the script. Write down exactly when the AI stops and a person takes over:

  • Always transfer or alert immediately for emergencies, safety issues and callers who ask for a human.
  • Take a message and notify for anything it cannot answer from your information, with the caller’s name, number and a short summary.
  • Tell the caller what happens next, for example “Someone from the office will call you back today,” and only promise what your team really does.
  • Route by topic so billing questions reach billing and service problems reach the right manager.
  • Review missed handoffs weekly. A quick look at call summaries shows where the AI should have passed the call on.

Callers forgive talking to an AI. They do not forgive being stuck in a loop with no way to reach a person.

What it costs to run

We will not quote a number here, because the price depends on your setup. These are the things that drive it:

  • Channels: phone calls cost more to run than chat or text, because voice adds speech recognition, speech generation and telephony charges, usually billed per minute.
  • Call volume and length: most running costs scale with how many conversations it handles and how long they last.
  • Integrations: connecting to your calendar, CRM, job management or practice software is where most of the build effort goes.
  • Off-the-shelf or custom: a ready-made AI receptionist service is quicker to start; a custom build fits unusual booking rules, several locations or your own systems.
  • Upkeep: someone has to keep hours, services and handoff rules current and review calls.

For a fuller breakdown of build and running costs for AI agents, see AI agent development cost: what drives the price.

Where AI receptionists go wrong

  • Making things up. Without a clear script of facts, it may invent a price, a policy or an opening. Tell it to say “I’ll have someone confirm that” instead of guessing.
  • Double bookings from a calendar that is not connected live or rules that are not written down.
  • No easy way to a person, which frustrates callers and loses jobs.
  • Outdated information after hours, services or staff change.
  • Taking odd instructions from callers. A caller can try to talk it into a discount or into revealing information. Limit what it can do and read our guide to prompt injection protection.
  • Collecting sensitive details it does not need, such as full card numbers or medical history. Keep the intake to what the callback needs.

How to launch one safely

  1. Write a one-page fact sheet: hours, services, service area, booking rules, and who handles what.
  2. Write the handoff rules from the section above.
  3. Start with one channel (often after-hours calls or website chat) and one or two call types.
  4. Test it with 30 or so real call reasons your office already gets, including the hard ones.
  5. Tell callers they are speaking with an AI assistant, and follow the rules for recording and disclosure where you operate.
  6. Review summaries every week for the first month, then fix the fact sheet and rules.

If you are choosing a first AI project in general, our AI implementation guide walks through picking and piloting one use case.

How GrtLabs builds AI receptionists and front-desk agents

GrtLabs Inc is a Houston-based AI and software development company, founded in 2019. Through our AI Factory we build AI agents and AI assistants that answer from your own information, take actions in the tools you already use through automations, and hand off to your team with a clear summary. Every agent has guardrails, human approval steps where you want them, and a log of what it did.

Most projects start with AI Discovery, a short engagement that scopes the first use case, the systems to connect and the handoff rules before you commit to a build. Houston businesses can also see our AI development company in Houston page.

Missing calls during busy hours? Talk to our Houston team about an AI receptionist built around your calendar, your services and your handoff rules.

Frequently asked questions

What is an AI receptionist?

It is an AI agent connected to your phone line, website chat or text messages that answers routine questions, takes caller details, books or requests appointments and hands urgent or unclear calls to a person.

Can an AI receptionist book appointments?

Yes, when it is connected to your real calendar or scheduling tool and has written booking rules for services, durations, buffers and service areas. If your tool cannot connect, it can collect the details as an appointment request for your team to confirm.

How does an AI receptionist hand a call to a person?

You set handoff rules: transfer or alert immediately for emergencies and callers who ask for a human, and take a message with a summary for anything it cannot answer. It should tell the caller what happens next.

How much does an AI receptionist cost for a small business?

It depends on the channels (voice usually costs more to run than chat or text), call volume and length, the systems it connects to, and whether you use a ready-made service or a custom build. Upkeep and call reviews also take staff time.

Will customers know they are talking to an AI?

They should. Tell callers they are speaking with an AI assistant, offer an easy way to reach a person, and follow the recording and disclosure rules where you operate.

Is an AI receptionist right for my business?

It fits best when you miss calls during busy hours or after closing, most calls are routine questions or bookings, and you have a clear process for callbacks. If most calls need judgment, start with message-taking and after-hours coverage.

Sunflower yellow graphic: Ask HR bot built safely, chat answering from the current handbook and handing exceptions to HR, GrtLabs logo
How to Build an Ask HR Bot From Your Policy Documents Safely
How to Build an Ask HR Bot From Your Policy Documents Safely 1024 538 Nadeem Shaikh

To build an Ask HR bot from your policy documents safely, give it only the current, approved versions of your policies, make it answer from those documents with a link to the exact section, and have it hand anything personal, sensitive or unclear to a person on your HR team. The hard part is not the AI. It is keeping the handbook clean: one current version of each policy, a clear owner, an effective date, and a note on which employees or locations it covers.

This guide is for HR managers who want employees to get quick, correct answers to policy questions without opening a ticket. It covers how to prepare your policy documents, how to stop the bot from quoting an outdated version, what it should refuse, and how to test it before launch. Access control and single sign-on are covered in our earlier guide to an internal AI assistant for HR and IT.

What an Ask HR bot should and should not do

An Ask HR bot is a chat assistant that answers employee questions from your own HR documents: the employee handbook, leave and benefits summaries, expense and travel policies, holiday calendars and onboarding guides. It works by retrieval-augmented generation (RAG): it searches your documents for the relevant passage, then writes a short answer based only on that passage.

Good fitHand to a person
“How many holidays do we get this year?”“Can I get an exception to the leave policy?”
“Where is the expense form and what is the deadline?”Anything about a specific employee’s pay, discipline or performance
“What does the remote work policy say about equipment?”Complaints, harassment, safety or discrimination reports
“Who do I contact about benefits enrollment?”Medical, accommodation or legal questions
“What is the dress code for client visits?”Anything the policy does not clearly answer

The rule of thumb: if the answer is written in a current policy and applies to everyone it covers, the bot can answer it. If it needs judgment or touches one person’s situation, it routes to HR.

Step 1: Get your policy documents ready

Most wrong answers from an Ask HR bot trace back to the documents, not the model. Before you connect anything, clean up the source.

  1. Make one list of every policy. Handbook chapters, standalone policies, benefits summaries, FAQs and old announcement emails that people still treat as policy.
  2. Pick one current version of each. Archive drafts, old PDFs and copies saved in personal folders. If two documents disagree, the bot will too.
  3. Add an owner and an effective date. Put them at the top of each document so the bot can show them and you know who approves changes.
  4. Mark who it applies to. If a policy differs by state, office, country, or for hourly versus salaried staff, say so in the document, in plain words.
  5. Split long documents by topic. Clear headings such as “Bereavement leave” help the search step find the right passage and let the bot link straight to it.
  6. Convert scans to real text. Scanned PDFs need text extraction first (see how document AI reads a form), or the bot simply cannot see them.

Step 2: Keep outdated versions out of answers

The biggest risk with an HR policy chatbot is a confident answer from last year’s handbook. Build these controls in from the start:

  • Only index approved documents. Connect the bot to one approved folder or library, not to everyone’s shared drive.
  • Remove old versions from the index the moment a policy is replaced, not just from the folder view.
  • Show the source and date with every answer, for example “From the PTO Policy, effective January 1.” Employees can then check it themselves.
  • Re-sync on a schedule and after every policy change, with a named person who confirms the update went through.
  • Review what was asked. A monthly look at questions and unanswered topics shows which documents are missing or unclear.

Step 3: Set clear rules for answers and refusals

Write the bot’s instructions the way you would brief a new HR coordinator:

  • Answer only from the retrieved policy text, and say “I can’t find that in our policies” when it is not there. No guessing.
  • Quote or link the exact section it used.
  • Ask which location or employee group applies when a policy differs, instead of picking one.
  • Never give legal, medical, tax or immigration advice.
  • Do not collect personal details in the chat. For anything about one person’s case, point to the right HR contact or open a ticket.
  • Keep a respectful, neutral tone, and route complaints and safety reports straight to a person.

Many teams also turn off general web knowledge for this bot entirely. An Ask HR bot that sounds helpful but answers from the internet instead of your handbook is worse than no bot.

Step 4: Test it with real questions before launch

Collect 30 to 50 real questions HR already gets by email or chat, with the correct answer and the policy section for each. Include the awkward ones:

  • Questions where the answer differs by location or employee type.
  • Questions about a policy that changed recently.
  • Questions the policy does not answer at all (the bot should say so).
  • Personal or sensitive questions (the bot should hand off, not answer).
  • Questions phrased the way employees actually write them, with typos and slang.

Have the policy owner score each answer as correct, wrong or should-have-refused. Fix the documents or instructions, run the set again, and repeat until wrong answers on policy facts are gone. Keep the question set and re-run it after every handbook update.

Step 5: Launch small and keep a person in charge

  • Start with one team or office and a few policy areas, such as leave, holidays and expenses.
  • Add a visible “Talk to HR” button in every answer.
  • Log questions and answers with access limited to HR, and decide how long logs are kept.
  • Give one person in HR ownership of the bot’s documents, instructions and test set.
  • Expand to more policies only after the first set has been accurate for a few weeks.

Who can use the bot, and which documents each group can see, is an access-control question. Our HR and IT assistant guide walks through single sign-on, permissions and logging in detail.

Common mistakes with an Ask HR bot

  • Pointing it at a whole shared drive full of drafts and old versions.
  • No effective dates, so nobody can tell which answer is current.
  • Letting it answer when the policy is silent, instead of saying it does not know.
  • Testing only with easy questions written by the project team.
  • No owner after launch, so the bot drifts out of date with the handbook.

How GrtLabs builds Ask HR bots

GrtLabs Inc is a Houston-based AI and software development company, founded in 2019. Through our AI Factory we build internal AI assistants and RAG systems that answer from your own documents with source links, permissions and HR handoff built in. We can deploy inside your own AWS or Azure account so employee data and policies stay in your environment.

Cost depends on how many documents and systems are involved and how it connects to your HR tools, which is why most projects start with AI Discovery: a short engagement that scopes the first use case before you commit to a build. For the budget side, see what drives AI development cost, and for picking a first pilot, our AI implementation guide.

Want employees to get policy answers without waiting on HR? Talk to our Houston team about an Ask HR bot built on your own handbook.

Frequently asked questions

How do I build an internal Ask HR bot using policy documents safely?

Start with one current, approved version of each policy, with an owner and effective date. Connect the bot only to that approved library, make it answer from the retrieved text with a link to the source, refuse when the policy is silent, and hand personal or sensitive questions to HR. Test it with real employee questions before launch.

What is an Ask HR bot?

It is a chat assistant that answers employee questions from your own HR documents, such as the employee handbook and leave, benefits and expense policies. It searches those documents and writes a short answer with a link to the policy section it used.

How do you stop an HR policy chatbot from using outdated policies?

Index only an approved folder, remove old versions from the index as soon as a policy is replaced, show the policy name and effective date with every answer, and re-sync after each policy change with a named person confirming it.

What questions should an Ask HR bot not answer?

Questions about one employee’s pay, discipline or performance, complaints and safety reports, medical, accommodation or legal questions, requests for exceptions, and anything the written policy does not clearly cover. Those go to a person on the HR team.

Can an employee handbook chatbot handle policies that differ by location?

Yes, if the documents say who each policy applies to. The bot should ask which location or employee group applies when a policy differs, rather than guessing.

Does an Ask HR bot replace the HR team?

No. It takes routine policy lookups off HR’s plate and points people to the right document. Judgment calls, sensitive cases and policy decisions stay with your HR staff.

Coral and mint graphic: AI invoice reconciliation, matching invoices to payments, bank lines and statements, GrtLabs logo
AI Invoice Reconciliation: Matching Invoices to Payments
AI Invoice Reconciliation: Matching Invoices to Payments 1024 538 Nadeem Shaikh

Yes, AI can match invoice details with actual payment records. AI invoice reconciliation reads each invoice, payment, bank line and vendor statement, pairs them up even when amounts, dates and references don’t line up exactly, and sends only the leftovers to a person. Your accounting system or ERP stays the system of record: the AI suggests matches and flags exceptions, and your team approves what gets posted.

This guide covers what AI invoice reconciliation actually matches, why simple rules miss so many items, how the matching works step by step, and where you should keep a person in the loop. If you still need to get invoices captured in the first place, start with our guide to automating invoice processing without a new ERP, which covers extraction and 2-way and 3-way matching against purchase orders.

What AI invoice reconciliation matches

Reconciliation is the step after an invoice is approved: proving that what you were billed, what you paid and what the vendor thinks you owe all agree. There are usually four pairings to check.

MatchWhat it answersTypical mess
Invoice to paymentWas this invoice paid, once, for the right amount?One payment covers several invoices, or one invoice is paid in parts
Payment to bank lineDid the money actually leave the account?Bank descriptions are short and truncated, and dates shift by a few days
Vendor statement to your AP ledgerDoes the vendor agree with what you owe?Credits, short-pays and invoices you never received
Invoice to invoiceIs this a duplicate?The same bill sent twice, rescanned, or re-sent with a new number

Purchase-order matching (invoice against PO and goods receipt) comes before all of this and is covered in the Nov 3 invoice post. AI invoice reconciliation picks up after the payment run.

Why rule-based matching leaves so much for people

Most accounting tools already match on exact rules: same amount, same reference, same date. That works for clean items. The rest lands in a spreadsheet for someone to work through by hand, usually because of things like:

  • Partial and combined payments. One transfer pays five invoices, minus a credit memo.
  • Different references. The vendor’s invoice number, your internal number and the bank memo are three different strings.
  • Fees and small differences. Bank or card fees, rounding, early-payment discounts and currency conversion change the amount.
  • Timing. A payment clears days after it was sent, so the dates don’t match.
  • Messy vendor names. “ACME SUPPLY CO” on the bank line, “Acme Supply Company LLC” in your vendor list.

AI helps exactly here. A language model can read a short bank memo and a remittance email the way a person would, and compare names, amounts, dates and references together instead of requiring one exact field to match. It is good at proposing the likely match and explaining why, which is what a reviewer needs.

How AI invoice reconciliation works, step by step

  1. Pull the records. Read open and paid invoices from your accounting system or ERP, payment runs, bank feed or statement exports, and vendor statements. Remittance emails and PDFs are read with document intelligence (see how IDP reads a form).
  2. Clean and normalize. Standardize vendor names, dates, currencies and reference formats so records can be compared.
  3. Match the easy ones with rules. Exact amount and reference matches are confirmed by plain rules first. They are cheap, fast and easy to audit.
  4. Let AI propose the rest. For the leftovers, the AI suggests matches: one-to-many, many-to-one, amount differences explained by a fee or discount, and likely duplicates. Each suggestion comes with a confidence level and a plain reason, such as “same vendor, amount equals invoices A plus B minus credit C.”
  5. Send exceptions to a person. Low-confidence suggestions, differences above your tolerance, and possible duplicates go to a review queue. The reviewer accepts, edits or rejects.
  6. Post through your normal controls. Approved matches are written back through the ERP’s API, an import file or, for older systems, an automation bot. Nothing is posted without the rules you set.
  7. Learn from corrections. Reviewer decisions become examples and rules, so the same vendor quirk is handled next time.

The split matters: rules for the certain matches, AI for judgment calls, and people for anything that changes the books in an unusual way. For how this differs from scripted bots, see AI agents vs RPA.

Where to keep a person in the loop

  • Write-offs and adjustments above an amount you choose.
  • Any suspected duplicate payment before money is recovered or a payment is stopped.
  • New vendors, or a change to a vendor’s bank details.
  • Month-end close sign-off.
  • Anything the AI marks as low confidence.

Every suggestion, approval and change should be logged with who approved it and when. That record is what makes the process auditable and what your accountant will ask for.

What you need before you start

  • Read access to your accounting system or ERP (API or regular exports). On QuickBooks Online, see what the QuickBooks connector in ChatGPT and Claude does.
  • A bank feed or statement export in a consistent format.
  • A few months of real invoices, payments and statements, including the messy ones.
  • Clear tolerance rules: what difference is acceptable, and who approves what.
  • One owner in finance who reviews exceptions and decides when rules change.

Start with one entity, one bank account or one group of high-volume vendors. Measure how many items still need a person and how long close takes, then expand. Our AI implementation guide covers picking a first pilot, and what drives AI development cost explains the budget side.

Common mistakes

  • Letting AI post entries directly with no tolerance limits or approval step.
  • Skipping the rules layer and sending every item through a model, which is slower and harder to audit.
  • Testing only on clean sample data instead of last quarter’s real exceptions.
  • Replacing the ERP when it only needed a matching layer on top.
  • No owner for the exception queue, so it piles up and the team goes back to spreadsheets.

How GrtLabs builds AI invoice reconciliation

GrtLabs Inc is a Houston-based AI and software development company, founded in 2019. Through our AI Factory we build document intelligence, AI automation and agentic AI that work alongside the accounting systems you already run, with exception queues, approval steps and logging built in. We can deploy inside your own AWS or Azure account so financial data stays in your environment.

Cost depends on the use case, the data involved and the systems we need to connect, which is why most projects start with AI Discovery: a short engagement that ranks use cases and recommends a proof of concept before you commit to a build.

Spending too many hours matching invoices to payments? Talk to our Houston team about AI invoice reconciliation for your AP process.

Frequently asked questions

Can AI match invoice details with actual payment records?

Yes. AI can compare invoices with payments and bank lines using vendor name, amount, date and reference together, including partial payments, combined payments, fees and small differences. It proposes matches with a reason and a confidence level, and a person reviews the uncertain ones.

What is AI invoice reconciliation?

It is using AI to check that invoices, payments, bank transactions and vendor statements agree, and to flag anything that does not, such as missing payments, duplicates or amount differences. It sits on top of your accounting system or ERP instead of replacing it.

How is AI invoice reconciliation different from 3-way matching?

3-way matching checks an invoice against the purchase order and goods receipt before payment. Reconciliation happens after payment and checks the invoice against what was actually paid, what cleared the bank and what the vendor statement shows.

Does AI invoice reconciliation work with our existing ERP?

Usually, yes. It reads data through the ERP’s API or regular exports and writes approved matches back the same way. For older systems without an API, approved matches can be posted through import files or an automation bot.

Can AI find duplicate invoices and duplicate payments?

It can flag likely duplicates, such as the same amount and vendor with a slightly different invoice number or a rescanned copy. A person should confirm each one before any payment is stopped or money is recovered.

Will AI post journal entries without approval?

Only if you allow it. A safe setup confirms exact matches by rule, sends AI-suggested matches above your tolerance to a review queue, and logs every approval so the process stays auditable.

Emerald and blush graphic: AI agent development cost, five price drivers plus monthly running costs, GrtLabs logo
AI Agent Development Cost: What Drives the Price
AI Agent Development Cost: What Drives the Price 1024 538 Nadeem Shaikh

AI agent development cost is set by two things: the route you choose (configure a tool you already pay for, hire a freelancer, hire a development company, or build in-house) and how much the agent has to touch. One workflow that reads data costs far less than an agent that writes to several systems, handles sensitive records and needs approval steps. Running costs after launch (model usage, hosting, monitoring and upkeep) are a separate line you should budget for from day one.

This guide does not quote a magic number, because any honest number depends on your process. Instead it shows what moves the price up or down, what keeps costing money after launch, and what to bring to a vendor so the quote you get is real. Several development shops published their own cost guides in late October, for example KUMO’s guide to hiring an AI agent development company; any price band in a guide like that is that vendor’s own number, not a market rate.

The four routes to an AI agent

Before you compare quotes, decide which route you are on. Each one moves cost to a different place.

RouteWhat you pay forBest whenWatch out for
Configure a platform you already pay forLicenses or usage credits, plus your team’s setup timeThe job fits the platform’s built-in connectors and rulesLimits on custom logic, data access and where data is processed
FreelancerHours or a fixed project feeA small, well-defined agent with one or two integrationsYou own project management, testing, security review and support after launch
AI development companyA scoped engagement: discovery, build, deployment, handoffThe agent crosses several systems, needs approvals and must run in productionVague scopes; make sure you get the code, logs and cloud accounts
In-house hire or staff augmentationSalaries or contractor time, plus hiring and ramp-upAI is core to your product and you need people on it every weekTime to hire, and a single person becoming the only one who understands the agent

Most small and mid-size teams should test the first route before paying anyone to build. Our guide to off-the-shelf AI vs custom AI walks through that decision. If you need extra hands rather than a full project, staff augmentation is a fourth way to add engineers without a full-time hire.

What drives AI agent development cost

Two agents with the same name (“support agent,” “order agent”) can differ a lot in effort. These are the drivers that matter most for AI agent development cost:

  1. Number of systems it connects to. Each API, database or document store is an integration to build, test and keep working. One CRM is simpler than a CRM, an ERP and a shared drive.
  2. Read-only vs taking actions. An agent that looks things up is cheaper than one that updates records, sends messages or moves money. Actions need permissions, approval steps and rollback plans.
  3. Data access and quality. Clean, reachable data is cheap. Data spread across inboxes, PDFs and old systems needs extraction and cleanup first.
  4. Accuracy bar and testing. A draft that a person always reviews needs less testing than a step that runs on its own. Building an evaluation set of real examples is part of the cost.
  5. Human review screens. If people approve or correct the agent’s work, someone has to build the review queue and the rules that send items there.
  6. Security, identity and hosting. Single sign-on, per-user permissions, running inside your own AWS or Azure account and protecting against prompt injection all add design and build time. See how to give AI agents real access safely.
  7. Logging and audit needs. Regulated or sensitive work needs a durable record of every step the agent took and every approval a person gave.
  8. One agent or several. Multiple agents handing work to each other cost more to design, test and monitor than one agent with a short tool list.

Notice what is not on the list: the model itself is rarely the biggest build cost. The work is in the integrations, the guardrails and the testing. If an agent is not the right shape at all, a simpler tool may do the job; compare AI chatbot vs AI agent and AI agents vs RPA before you scope.

Running costs after launch

The build is a one-time line in your AI agent development cost. These keep coming every month:

  • Model usage. Model APIs bill by tokens, the pieces of text the model reads and writes. Providers such as Anthropic and Amazon Bedrock list input and output tokens at separate prices, so long prompts, big documents and chatty multi-step agents cost more per run.
  • Discounts you can design for. Both pricing pages offer lower rates for work that does not need an instant answer (Anthropic: “Save 50% with batch processing”; Bedrock: batch inference “at a 50% lower price compared to on-demand” for select models) and cheaper reads of cached prompts. A smaller model can also be enough for narrow steps; see small language models.
  • Hosting and storage. The servers, queues, databases and search indexes the agent runs on.
  • Monitoring and evaluation. Dashboards, alerts and re-running your test set after every change. Our LLMOps guide covers what this looks like.
  • Upkeep. Models get retired and APIs change. For example, Anthropic set a retirement date for Claude Sonnet 4.5 on its API; see what to check before a model retires. Someone has to re-test and switch.
  • The owner’s time. A person on your team reviews exceptions and decides what to improve next. That time is real cost.

Ask every vendor for an estimate of monthly running costs at your expected volume, and set a budget alert in your cloud account before the first production week.

Why the pilot is the cheap part

A demo that works on a handful of examples is not a production agent. Moving to production usually adds:

  • A test set of real cases that runs every time the prompt, model or tools change
  • A cost ceiling and alerts so one bad loop does not run up the bill
  • Monitoring that tells a person when the agent fails, before customers notice
  • A fallback when the model or provider is down, and a way to roll back a bad change
  • Handover documentation and training for the people who will run it

Budget for that production layer up front, or the second invoice will surprise you. The NIST AI Risk Management Framework is intended for voluntary use and is a useful checklist for what “production ready” should mean for your risk level.

How to get a reliable AI agent quote

Vendors can only price what you describe, so the clearest way to pin down AI agent development cost is a clear brief. Bring this to the first call:

  1. One workflow in one paragraph: what starts the work, the steps a person does today, the systems touched, and what “done” looks like.
  2. Real samples: a set of actual tickets, documents or records, including messy ones.
  3. A systems list: every app the agent must read from or write to, and whether each has an API.
  4. Approval rules: which actions the agent may take alone and which need a person.
  5. Where it must run: a vendor’s cloud, or inside your own AWS or Azure account. Our AWS Bedrock vs Azure OpenAI guide helps here.
  6. Volume you know today: how many runs a day or month, from your own records.

Then ask for line items: discovery, build, integrations, testing, deployment, training and monthly running costs, listed separately. A single lump sum hides where the money goes. Our checklist on choosing an AI development company has the questions to ask about code ownership, logs and support.

When not to build a custom agent

  • The task already fits a tool you pay for, with its built-in connectors.
  • No one on your team will own the agent and review its exceptions.
  • The process changes every week, so there is nothing stable to automate yet.
  • A fixed rule or script would do the job; you do not need judgment at each step.
  • You cannot get access to the data or systems the agent needs.

In those cases, spend first on fixing ownership, access or the process itself. Our AI implementation guide covers picking a first use case that is ready.

How GrtLabs scopes AI agent development

GrtLabs Inc is an AI and software development company based in Houston, Texas, founded in 2019. Under our AI Factory we build agentic AI, AI automation, AI assistants and the cloud setup they run on.

Our answer to “how much does it cost?” is the same one on our Houston AI development page: it depends on the use case, the data involved and the systems we need to connect. That is why we start with AI Discovery, a focused 2–3 week engagement that ends with use cases ranked by value and feasibility, clear build or no-build recommendations, and a recommended proof of concept. The roadmap is yours whether or not you build with us.

  • We start small, prove the result on your own data, then expand.
  • Agent design covers goals, tools, permissions and escalation rules, with human-in-the-loop approval steps and logging.
  • On AWS / Azure AI we deploy inside your cloud account, right-size services, set up monitoring and budgets, and review usage with you.
  • We hand off with documentation and training for the people who will run it.

Want a real estimate for your agent? Talk to our Houston team or start with AI Discovery.

Frequently asked questions

How much does AI agent development cost?

It depends on the route you choose and how much the agent has to touch: the number of systems it connects to, whether it only reads data or also takes actions, the quality of your data, your accuracy bar, and your security and logging needs. Get a scoped quote with line items rather than relying on a general price band.

What makes an AI agent more expensive to build?

More integrations, write access to business systems, messy or hard-to-reach data, a high accuracy bar, human review screens, single sign-on and per-user permissions, hosting inside your own cloud, audit logging, and multiple agents handing work to each other.

What are the running costs of an AI agent after launch?

Model usage billed by input and output tokens, hosting and storage, monitoring and evaluation, upkeep when models are retired or APIs change, and the time of the person who owns the agent. Batch processing, prompt caching and smaller models can lower model usage.

Is it cheaper to use an off-the-shelf AI tool?

Often, yes, when the job fits the tool’s built-in connectors and rules. Custom development makes sense when the process crosses several systems, needs your own permissions and approvals, or must run inside your own cloud account.

How do I get an accurate AI agent quote?

Describe one workflow in a paragraph, share real samples, list every system the agent must touch, state which actions need approval, say where it must run, and share the volume you see today. Ask for discovery, build, testing, deployment, training and monthly running costs as separate line items.

How does GrtLabs price AI agent work?

Cost depends on the use case, the data involved and the systems we need to connect. We usually start with AI Discovery, a 2 to 3 week engagement that ranks use cases and recommends a proof of concept, so you can see what is worth building before committing to a larger project.

Amber and charcoal graphic: a low-stock gauge beside the Runway inventory table, GrtLabs logo
Finding Low Stock Products in BigCommerce: A Simple Guide
Finding Low Stock Products in BigCommerce: A Simple Guide 1024 538 Nadeem Shaikh

The most reliable way to find low stock products in BigCommerce is to look at days of cover (how long stock will last at recent sales pace), not just the quantity left. Pick a threshold that fits your supplier lead time, such as 14 days, and check which products fall under it. Out of stock products and products with no recent sales are worth looking at separately.

Below we explain why quantity alone can fool you, how to set a sensible threshold, and a quick weekly routine.

Why “low stock” is not just a small number

Imagine two products, each with 10 units left. One sells five a day. The other sells one a month. The first runs out in two days. The second will last most of a year. Both show “10 in stock”, but only one is really low.

That is why a fixed quantity rule (for example, “anything under 10 is low”) often flags the wrong products. It misses fast sellers with a decent count and worries you about slow sellers with a small count.

Use days of cover instead

Days of cover = stock on hand ÷ average daily sales (usually from the last 30 days). It turns every product into the same unit: days. A product with 7 days of cover is closer to running out than one with 40, no matter what the raw counts are. Our days of cover guide walks through the formula in detail.

Setting your low stock threshold

Your threshold should be at least as long as it takes to get more stock, plus a buffer.

  • Lead time: how many days from placing an order with your supplier until the stock is ready to sell.
  • Safety stock / buffer: extra days to cover delays or a sudden jump in sales.

Threshold ≈ lead time + buffer

A common starting point is about two weeks. Runway, for example, marks products with under 14 days of cover as low cover. If your supplier takes three weeks, you would likely want a higher number for your own planning.

A worked example (made-up numbers)

Example numbers only, not real store data.

Your supplier usually delivers in 9 days and you like a 5-day buffer, so you watch anything under 14 days of cover.

Product On hand Sold last 30 days Days of cover Under 14 days?
Ceramic mug 20 60 10 Yes
Travel cup 6 9 20 No
Wool scarf 45 30 45 No
Linen tote 0 12 0 Out of stock
Gift card holder 12 0 blank No recent sales

What stands out:

  • The mug has 10 days of cover, below your 14-day line. It is worth a look today.
  • The travel cup has only 6 units, which looks scary, but at its slow pace that is 20 days. It is fine for now.
  • The tote is already out of stock, so it is losing sales.
  • The gift card holder has not sold in 30 days. It is not low; it may be a different problem.

You decide what to do with each one. The view just makes it obvious where to look. Once you know where to look, our guide to deciding what to reorder on BigCommerce covers the next step.

Three groups to check every week

1. Low cover

Products under your threshold. These are the ones most likely to run out before new stock arrives.

2. Out of stock

Products at zero. Check whether they are on order, discontinued, or simply forgotten.

3. No recent sales

Products that have not sold in 30 days. They are not low, but they tie up cash and shelf space. Check 60 and 90-day sales before deciding anything.

A view is not an alert

It helps to be clear about the difference:

  • A low stock view is a list you open and read. You choose when to check it.
  • An alert is something that messages you automatically.

Runway is a view. It does not send emails or messages. Many small store owners find a set weekly check (for example, Monday morning with coffee) works well, because it fits naturally into planning orders.

How to find low stock products

With a spreadsheet

  1. Export the last 30 days of orders and your product list with stock.
  2. Total units sold per SKU.
  3. Calculate days of cover: =IF(sold_30=0,"",on_hand/(sold_30/30)).
  4. Filter to days of cover under your threshold, then separately to stock of zero.

With a free app

Runway is a free, read-only BigCommerce app from GrtLabs. It shows units sold in the last 30, 60 and 90 days, stock on hand and days of cover for each product, with the products that run out soonest at the top. You can filter to low cover (under 14 days), out of stock, or no recent sales, open any product in your BigCommerce admin, and download the list as a CSV. Get it on the BigCommerce App Marketplace.

Runway only reads orders, products and inventory. It does not change your stock, create purchase orders or email suppliers. If you need a custom workflow on top of your store data, our team builds BigCommerce apps and integrations.

Quick recap

  • Judge “low” by days of cover, not raw quantity.
  • Threshold ≈ supplier lead time + buffer; 14 days is a common starting point.
  • Check low cover, out of stock and no recent sales as three separate lists.
  • A weekly habit works well with a view-based tool.

FAQ

How do I find low stock products in BigCommerce?

Compare stock on hand with recent sales. Calculate days of cover for each product and look at the ones under your threshold, plus anything at zero stock.

What counts as low stock?

It depends on how long your supplier takes to deliver. A common rule of thumb is any product whose days of cover is shorter than your lead time plus a small buffer.

Why not just use a minimum quantity?

A fixed quantity ignores sales speed. Ten units can last two days or a year depending on how fast the product sells.

Does Runway send low stock emails?

No. Runway is a read-only view. You open it to see low cover and out of stock products; it does not send emails or messages.

What does Runway count as low cover?

Runway marks products with under 14 days of cover as low cover, where days of cover is stock on hand divided by average daily sales over the last 30 days.

Terracotta and olive graphic: a BigCommerce inventory report with 30, 60 and 90 day sales columns next to the Runway table, GrtLabs logo
How to Build a BigCommerce Inventory Report (Units Sold Over 30, 60 and 90 Days)
How to Build a BigCommerce Inventory Report (Units Sold Over 30, 60 and 90 Days) 1024 538 Nadeem Shaikh

A useful BigCommerce inventory report puts units sold per product over the last 30, 60 and 90 days right next to current stock on hand. You can build it yourself by combining an order export and a product export in a spreadsheet, or use a free app that shows it in one table. Seeing sales and stock together is what makes the report useful.

This guide covers what to include, how to build it step by step, and when an app saves you time.

What should an inventory report show?

Many store owners start with a stock list: product name and quantity. That tells you what you have, but not whether it is enough. A report you can act on answers two questions at once:

  1. How much do I have? (stock on hand)
  2. How fast is it selling? (units sold over recent periods)

From those two, you can work out a third: days of cover, or how long current stock will last. We explain that in detail in our days of cover guide.

Why 30, 60 and 90 days?

  • 30 days shows what is happening right now.
  • 60 days smooths out a slow or busy week.
  • 90 days shows the bigger trend and helps you spot products that are speeding up or slowing down.

If a product sold 30 units in the last 30 days but only 45 over 90 days, it is picking up speed. If it sold 10 in 30 days and 90 over 90 days, it is slowing down.

The two routes

Spreadsheet route App route
Cost Free Free (Runway) or paid tools
Time per refresh Manual, every time Opens ready to read
Flexibility Total What the app shows
Good for Small catalogs, one-off checks Regular weekly checks

Building the report with exports and a spreadsheet

Step 1: Export your orders

In your BigCommerce admin, go to Orders and export orders covering at least the last 90 days. Make sure the export includes product line items (product name or SKU, and quantity).

Step 2: Export your products

Go to Products and export your catalog with current stock levels. This gives you stock on hand for each product or SKU.

Step 3: Total units sold by period

In your spreadsheet, add a column that marks each order line by age (days since the order date). Then use a pivot table or SUMIFS to total quantity per SKU for:

  • orders in the last 30 days
  • orders in the last 60 days
  • orders in the last 90 days

Example formula for 30 days: =SUMIFS(qty, sku, A2, order_date, ">="&TODAY()-30)

Step 4: Join stock on hand

Use XLOOKUP or VLOOKUP to pull stock on hand from the product export into the same sheet, matched by SKU.

Step 5: Add days of cover (optional but helpful)

=IF(sold_30=0,"",on_hand/(sold_30/30))

Leaving it blank when nothing sold avoids a misleading “infinite” value.

Step 6: Sort

Sort by days of cover, lowest first, so the products closest to running out are at the top.

A worked example (made-up numbers)

These numbers are an example only.

Product On hand Sold 30d Sold 60d Sold 90d Days of cover
Ceramic mug 20 60 100 140 10
Wool scarf 45 30 50 60 45
Linen tote 8 12 30 48 20
Gift card holder 12 0 2 5 blank

What this tells you:

  • The mug is selling faster lately (60 in the last 30 days versus 140 over 90) and has only 10 days of cover.
  • The scarf has plenty of stock for now.
  • The tote is slowing down: 12 in the last month versus an average of 16 a month over 90 days.
  • The gift card holder has barely moved, so its cover is blank.

None of this tells you what to do. It gives you the facts so you can decide, based on your supplier lead times, cash and plans. For a step-by-step way to make that call, see how to decide what to reorder on BigCommerce.

Common problems with DIY reports

  • It goes stale fast. The sheet is only as fresh as your last export.
  • Variants and SKUs get messy. Make sure you match on the same identifier in both exports.
  • Big catalogs are slow. Thousands of order lines can make formulas sluggish.
  • Cancelled or refunded orders. Decide whether to include them and stay consistent.

Using an app instead

If you check stock every week, rebuilding the sheet gets old. Runway is a free, read-only BigCommerce app from GrtLabs that shows this report in one table. For each product you see units sold in the last 30, 60 and 90 days, stock on hand, and days of cover. Products that run out soonest are at the top, you can filter to low cover, out of stock or no recent sales, open a product in your BigCommerce admin, and download everything as a CSV. It is listed on the BigCommerce App Marketplace.

Runway only reads orders, products and inventory. It does not create purchase orders, email suppliers or change your stock. If you need purchase orders or supplier workflows, a full inventory system is a better fit. If you need something built around your own process, our team builds custom BigCommerce apps and integrations.

Quick recap

  • Put sales and stock side by side, not just stock.
  • Use 30, 60 and 90-day windows to see the trend.
  • Add days of cover to compare products on one scale.
  • Spreadsheets work; an app saves time if you check often.

FAQ

Does BigCommerce have an inventory report?

You can export orders and products from your BigCommerce admin and combine them in a spreadsheet. Apps such as Runway show units sold and stock on hand together in one table.

How do I see sales by product in BigCommerce?

Export orders with line items, then total quantity per product or SKU in a spreadsheet for the periods you care about, such as 30, 60 and 90 days.

Why compare 30, 60 and 90-day sales?

Comparing periods shows whether a product is speeding up, steady or slowing down, which a single number cannot show.

Can I export my inventory report to CSV?

Yes. A spreadsheet can be saved as CSV, and Runway lets you download its product list as a CSV.

Is Runway free?

Yes. Runway is a free, read-only BigCommerce app from GrtLabs with no upfront or monthly fee.

Mulberry and gold graphic: AI implementation in Houston rollout plan with four steps and GrtLabs logo
AI Implementation in Houston: Where to Start
AI Implementation in Houston: Where to Start 1024 538 Nadeem Shaikh

AI implementation in Houston for a small or mid-size business looks like this: pick one measurable use case, run a short pilot on your real data, prove the result, then expand. It is not a company-wide AI rollout on day one.

Houston teams already have real operations: documents to process, orders to route, questions to answer, and systems that need to talk to each other. That is the starting point for practical AI development in Houston, not a blank-slate transformation program. Searches for AI implementation Houston often land on service pages; this post is the how-to behind them. Other Houston IT firms have published their own take on where to start and what to avoid; see for example Braintek’s AI implementation guide for Houston small businesses. GrtLabs’ plan covers the first use case, the pilot, data and access, and who runs it after launch.

When you are ready for AI implementation (and when you are not)

An AI readiness assessment puts the hard questions on the table before you spend on a build: goals, data, systems, and people. You are closer to ready when:

  • You can name one process that hurts every week (re-keying invoices, triaging tickets, answering the same policy questions).
  • Someone on your team owns that process today and will own the result after launch.
  • You can get to the data or systems that process needs, even if access takes a ticket.
  • Leadership will accept a small pilot with a clear stop/go decision instead of a vague “AI initiative.”

You are not ready if the only brief is “add AI somewhere,” if no one will review exceptions, or if the data lives in personal inboxes and shared drives with no owner. In those cases, fix ownership and access first. Discovery still helps, but the first output may be a no-build call.

A useful readiness signal in Houston is proximity: can leadership and process owners meet for workshops in person, in the same time zone, without waiting weeks for travel? Face-to-face discovery often surfaces constraints that a remote kickoff misses, such as who actually controls a shared drive, which vendor portal still requires a fax, or which approval still lives in email.

Pick the first use case for AI implementation in Houston

The first use case decides whether the project sticks. Rank candidates with three filters:

FilterWhat good looks likeWhat to skip
MeasurableYou can count time saved, error rate, or throughput before and afterVague goals like “be more innovative”
Data availableReal samples exist (invoices, tickets, documents, CRM records)Ideas that need data you do not have yet
Clear ownerAn ops, finance, support, or sales lead will run it day to dayProjects that only “IT” or “the AI vendor” would own

Good first bets for Houston SMBs often sit in operations, support, sales ops, and back-office work. Our industry roundup on AI for Houston businesses covers energy, healthcare, logistics, and related use cases without pretending every company needs the same build.

Decide early whether you need a chatbot, an assistant, or an agent. A chatbot answers. An agent plans, calls tools, and finishes multi-step work with guardrails. Our guide to AI chatbot vs AI agent and our overview of what agentic AI is help you pick the right shape before you write a scope.

Also decide build vs buy. Some problems fit an off-the-shelf tool. Others need a custom workflow wired to your ERP, CRM, or warehouse system. See off-the-shelf AI vs custom AI.

Write the first use case in one paragraph before anyone talks models: the trigger (what starts the work), the steps a person does today, the systems touched, the definition of done, and the metric. If you cannot fill that paragraph, the use case is not ready. If you can, you already have the outline of the pilot acceptance test.

A practical pilot plan (discovery first)

AI Discovery at GrtLabs is a focused 2–3 week engagement. We work with your leadership and process owners to understand goals, data and systems, identify where AI can create real value, and decide what is worth building first, and what is not.

What you get from discovery:

  • A discovery workshop with key stakeholders
  • An AI readiness review of your data, systems and processes
  • A prioritized list of use cases ranked by value and feasibility
  • Clear build / no-build recommendations
  • A roadmap with a recommended proof of concept to start with

The roadmap is yours whether or not you build with us. That matters: a good pilot plan should leave you with a decision, not a dependency.

How an engagement typically moves after discovery, from our Houston AI development page:

  1. Discovery. Map process, data and systems; agree on one first use case.
  2. Proof of concept. Build the recommended use case in short iterations with your team, testing against your real documents and data.
  3. Production on AWS or Azure. Deploy inside your cloud account with access controls you own, plus monitoring, logging and cost budgets.
  4. Handoff and support. Document the solution and train your team to run it.

For agentic work, we start small, prove the result on your own data, then expand. Each stage has a clear output, your team is involved throughout, and nothing goes to production without documentation and training for the people who will run it. See Agentic AI. Typical agent deliverables include the agent design (goals, tools, permissions, escalation rules), integrations with your APIs and document stores, human-in-the-loop review steps, logging and monitoring, and handover documentation.

For repetitive document, order, and back-office flows, AI Automation maps the process, connects the systems you already use, and ships automations your team can run every day. Example shapes include extracting data from invoices or purchase orders into your ERP, routing orders and syncing inventory, generating routine reports, and validating records with mismatches flagged for review. When an automation hits an exception, the item should route to a person with the context they need, not disappear into a silent failure.

Keep the pilot narrow on purpose. One workflow, a fixed sample set, a written success bar, and a calendar date for the stop/go meeting. Expanding to a second use case is a reward for a clear win, not a way to rescue a fuzzy first attempt.

Data, access and security before you scale

Most stalled AI projects fail on access, not on models. Before the pilot expands:

  • Permissions. The system should respect existing roles. Internal Q&A over your documents should cite sources and stay inside what each user is allowed to see.
  • Identity. Prefer your company identity (SSO) and access controls you already trust, not a separate login that no one audits.
  • Where models run. For production, many Houston teams want the work inside their own cloud account. Our AWS and Azure AI service covers architecture, model hosting, retrieval, pipelines, identity and access controls, and cost monitoring.
  • Logging. Keep a trail of what the system did, what a person approved, and what changed. Our post on the AI agent audit trail covers what to record.
  • Ongoing risk. The NIST AI Risk Management Framework is intended for voluntary use and is a useful checklist for trustworthiness after launch, not only at kickoff.

Do not wait until “phase 2” to ask where customer or employee data will live. Ask in discovery. Also decide what must never leave your account, what can be processed by a managed model API under your cloud contract, and who can approve a new tool connection. Those answers belong in the same document as the first use case.

Cost monitoring belongs here too. Cloud AI spend is easy to ignore in a pilot and hard to unwind later. Set a budget owner and a simple alert before the first production week, even if the dollar amount is small.

Who runs it after launch

If the only owner is “the AI guy” or the vendor, the pilot dies when that person is busy. Name an ops owner before go-live:

  • Reviews exceptions and low-confidence items
  • Knows when to escalate to IT or the vendor
  • Watches simple metrics (volume handled, error rate, time to complete)
  • Owns the backlog of the next two use cases

On automations, we hand off with documentation and training so your team can run them. On agents, deliverables include human-in-the-loop review steps, logging, monitoring, and handover documentation. Your ops lead should be able to explain, in one sentence, what the system is allowed to do without asking a person.

Schedule a 30-day and 90-day review on the calendar at launch. At 30 days, check volume, exception rate, and whether the owner still has time. At 90 days, decide whether to expand, redesign, or stop. AI implementation Houston projects that skip those reviews often keep a weak pilot alive until someone quietly turns it off.

Common mistakes in AI implementation Houston teams make

  • Boiling the ocean. Ten use cases, no pilot, no owner.
  • Demo data only. Real Houston invoices, tickets and scans break pipelines that looked fine in a slide deck.
  • No exception path. Every automation and agent hits edge cases. Design the human handoff on day one.
  • Chatbot when you needed an agent (or the reverse). Answers are not the same as finishing work across systems.
  • Skipping readiness. Goals, data and ownership unresolved until halfway through a build.
  • No stop/go. A pilot without a written success bar becomes an endless experiment.

How GrtLabs helps with AI implementation in Houston

GrtLabs Inc is an AI lab and software development company based in Houston, Texas, founded in 2019. Under our AI Factory, we design, build and deploy practical AI systems for teams that already have real operations.

Typical path:

  • AI Discovery (2–3 weeks) to rank use cases and pick the first proof of concept
  • AI Automation or Agentic AI for the build, with guardrails and human review where it matters
  • AWS / Azure AI when the work needs secure cloud hosting, RAG, and access controls inside your account

Discovery sessions and stakeholder workshops can happen face to face in Houston, in your time zone. Learn more on our AI development company in Houston page.

Ready to pick the first use case and run a real pilot? Talk to our Houston team or start with AI Discovery.

Frequently asked questions

What does AI implementation in Houston look like for an SMB?

Pick one measurable use case, run a short discovery and pilot on your real data, prove the result, then expand. Name an ops owner before launch. Do not start with a company-wide AI program.

How long should an AI discovery or pilot take?

GrtLabs AI Discovery is typically a focused 2 to 3 week engagement. It ends with ranked use cases, build or no-build recommendations, and a recommended proof of concept. The roadmap is yours whether or not you build with us.

How do I choose the first AI use case?

Require three things: a metric you can improve, data or systems you can access, and a clear day-to-day owner. Skip vague innovation goals and projects that only a vendor would maintain.

Do we need to replace our ERP or CRM to implement AI?

Usually no. Good automations and agents connect to the tools you already use, with exception paths to a person when something does not match. Replacing a core system is a different project.

Where should our AI models and data run?

Many teams deploy inside their own AWS or Azure account so identity, access controls, monitoring and cost budgets stay under company control. Decide this in discovery, not after a prototype is live.

Who should own AI after launch?

An operations, finance, support or sales lead who already owns the process. They review exceptions, watch simple metrics and decide the next use case. IT and vendors support; they should not be the only owners.

Coral graphic: days of cover explained, with boxes, an hourglass and the Runway inventory screen
How to Calculate Days of Cover for BigCommerce Products
How to Calculate Days of Cover for BigCommerce Products 1024 538 Nadeem Shaikh

Days of cover tells you how many days your current stock will last if sales keep their recent pace. To work it out, divide the units you have on hand by your average daily sales, usually from the last 30 days. A product with 20 units on hand that sold 60 units in 30 days has 10 days of cover.

That one number makes it much easier to see which products could run out first. In this guide we walk through the formula, a worked example, the tricky edge cases, a spreadsheet method you can use today, and a free app that does the math for you.

What is days of cover?

Days of cover (also called days of inventory cover or days of stock) is a simple estimate: “At the speed this product has been selling, how long will what I have last?”

It is not a forecast and it does not know about next month’s holiday rush. It looks back at recent sales and forward at your current stock. That makes it easy to understand and easy to check, which is exactly why many small store owners like it.

Why it beats looking at stock counts alone

A stock count of 50 sounds healthy. But 50 units of a product that sells 10 a day lasts five days, while 50 units of a product that sells one a week lasts almost a year. Days of cover puts every product on the same scale (days), so you can compare a fast seller and a slow seller side by side.

The days of cover formula

Here it is in plain terms:

Days of cover = stock on hand ÷ average daily sales

And average daily sales is:

Average daily sales = units sold in the period ÷ number of days in the period

Most people use a 30-day period. It is recent enough to reflect what is happening now, but long enough that one busy afternoon does not throw the number off.

A worked example (made-up numbers)

The numbers below are an example only, not real store data.

Say you sell three products:

Product On hand Units sold (last 30 days) Avg daily sales Days of cover
Ceramic mug 20 60 2.0 10
Wool scarf 45 30 1.0 45
Gift card holder 12 0 0 blank

Step by step for the mug:

  1. Units sold in 30 days: 60
  2. Average daily sales: 60 ÷ 30 = 2 per day
  3. Days of cover: 20 ÷ 2 = 10 days

The scarf has more stock relative to its sales, so it has 45 days of cover. The mug is the one to look at first.

The edge case: no recent sales

Look at the gift card holder. It sold nothing in 30 days, so average daily sales is zero, and dividing by zero does not give a real answer. Some tools show “infinite” days of cover, which can be misleading. It makes a product that nobody is buying look like it has endless healthy stock.

A cleaner approach is to leave days of cover blank when there are no recent sales. A blank is honest: “There is not enough recent sales data to estimate this.” It also nudges you to ask a different question about that product, such as whether it needs a price change or a better product page.

Other things to keep in mind

  • New products: If a product launched 10 days ago, a 30-day average understates its pace. Treat the number with care.
  • Stockouts: If a product was out of stock for part of the month, it could not sell during those days, so the average looks lower than real demand.
  • Seasonality: A 30-day window before a big holiday will not reflect the holiday itself. Compare with 60 and 90-day sales to get a feel for the trend.

How to calculate days of cover in a spreadsheet

You can do this without any app:

  1. Export your orders for the last 30 days from your BigCommerce admin.
  2. In a spreadsheet, total the units sold for each product or SKU (a pivot table works well).
  3. Add a column for current stock on hand from your product export.
  4. Add a column for average daily sales: =units_sold/30.
  5. Add a column for days of cover: =IF(units_sold=0,"",on_hand/(units_sold/30)).
  6. Sort by days of cover, lowest first.

This works fine, but you have to repeat it every time you want fresh numbers, and it gets slow with a big catalog.

What is a “good” number of days of cover?

There is no single right answer, because it depends on how long your supplier takes to deliver (lead time) and how much buffer you like to keep (safety stock).

A simple way to think about it:

  • If your supplier usually takes 10 days to deliver, a product with 8 days of cover may run out before new stock arrives.
  • If you want a buffer of about a week on top of that, you might start paying attention once a product drops below roughly 17 days (10 days lead time plus 7 days buffer).

These are example numbers. Your own lead times and comfort level decide the threshold. The data shows you where each product stands; you make the call. When you are ready to turn these numbers into an order, see our simple method for deciding what to reorder on BigCommerce.

A free way to see days of cover for every product

If you would rather not rebuild the spreadsheet every week, Runway is a free, read-only BigCommerce app from GrtLabs that shows this for you. For each product it shows units sold in the last 30, 60 and 90 days, stock on hand, and days of cover, calculated the same way as above (on-hand stock divided by average daily sales from the last 30 days). Products that run out soonest sit at the top, products with no recent sales show a blank instead of an endless number, and you can download the list as a CSV. You can find it on the BigCommerce App Marketplace.

Runway only reads orders, products and inventory. It does not create purchase orders, email suppliers or change your stock. It gives you the numbers; you decide what to order.

If you need something more custom for your store, our team also builds BigCommerce apps and integrations.

Quick recap

  • Days of cover = stock on hand ÷ average daily sales.
  • Use the last 30 days for a recent, steady average.
  • Leave it blank when there were no sales, rather than calling it infinite.
  • Compare against your supplier lead time plus a buffer.
  • Check 60 and 90-day sales to spot trends.

FAQ

What is days of cover in inventory?

Days of cover is an estimate of how many days your current stock will last at your recent sales pace. It is calculated by dividing stock on hand by average daily sales.

How do I calculate days of cover?

Divide units sold in a period (often 30 days) by the number of days to get average daily sales. Then divide stock on hand by that number. For example, 20 on hand ÷ (60 sold ÷ 30 days) = 10 days of cover.

What if a product has no sales in the last 30 days?

Average daily sales is zero, so the formula cannot give a real answer. It is clearer to leave days of cover blank than to show an infinite number.

What is a good days of cover number?

It depends on your supplier lead time and the buffer you want to keep. Many store owners compare days of cover with lead time plus a few extra days of safety stock.

Does BigCommerce show days of cover?

You can calculate it from your order and product exports in a spreadsheet. Runway, a free read-only BigCommerce app from GrtLabs, shows days of cover for each product along with units sold over 30, 60 and 90 days.