AI

Year-End Stocktake: How AI Helps Burnaby Shops Count

November 9, 2026 · 7 min read · Autana Solutions, Vancouver
Year-End Stocktake: How AI Helps Burnaby Shops Count | Autana Solutions

December is coming, and for a lot of shop owners that means the least favourite job of the year: the stocktake. Clipboards, cold storerooms, a spreadsheet full of typos. AI tools promise to make that easier. Some of that promise holds up and some of it doesn't. Here's what the evidence says, and how a small retailer in Burnaby or New Westminster could set up a count that's faster without being sloppy.

Why the year-end count still matters

The stocktake isn't optional busywork. The Canada Revenue Agency's page on inventory and cost of goods sold says you need to do an annual inventory, and that inventory is used to calculate cost of goods sold and net income on Form T2125. If your closing number is wrong, your taxable income is wrong too.

The CRA also sets out how to value it. You can use the fair market value of the entire inventory, or value individual items (or classes of items) at cost or fair market value, whichever is lower. Once you choose a method, the CRA says you have to keep using it in later years.

AI doesn't change any of those rules. What it can change is how many hours your team spends on the floor on December 31, and how many errors get into the numbers afterwards.

Where small businesses actually are with AI

Plenty of owners worry they're behind. The data says most aren't. Statistics Canada's analysis of AI use by businesses in the second quarter of 2026 found that 19.2% of businesses used AI to produce goods or deliver services in the previous 12 months, up from 6.1% in Q2 2024. Wholesale trade was at 7.9%, one of the lowest rates of any industry. And 40.0% of businesses said AI isn't relevant to them at all.

Among businesses that did use AI, StatCan found the most common uses were data analytics (36.6%), text analytics (34.5%) and virtual agents or chatbots (28.2%). Cybersecurity or privacy concerns were the top barrier at 13.4%, followed by cost at 10.6%.

That's a good guide for a stocktake. The useful work is mostly with data: reading it, reconciling it, flagging problems. It isn't robots counting shelves.

What AI is good at during a stocktake

A count is really three jobs. You record what's on the shelf, compare it with what the system thinks is there, and explain the gaps. AI is a lot more help with some of these than others.

  • Reading labels and paperwork. A vision model can read SKUs, lot numbers, expiry dates and handwritten count sheets from phone photos and turn them into spreadsheet rows. That only works when the photo is clear and the text is legible.
  • Reconciling against your POS export. Matching counted quantities against expected on hand, SKU by SKU, and listing every variance over a threshold you set takes a person hours. Software does it fast.
  • Drafting variance notes. An assistant can check your receiving log and suggest a likely reason for each big gap, like stock received on December 28 that hasn't been entered yet. A person confirms or corrects it.
  • Flagging likely write downs. Slow movers, expired lots and damaged stock can be pulled into one list so you can decide whether the lower of cost or market value applies.
  • Answering staff questions on count day. A simple chatbot that knows your count procedure saves the manager from hearing "do we count the display units?" forty times.

What it's not good at: counting piles of small things

This is the part that gets skipped in sales demos. Anthropic's own vision documentation says Claude "can give approximate counts of objects in an image but might not always be precisely accurate, especially with large numbers of small objects." The same page warns that the model can make mistakes with low quality, rotated or very small images. It also says not to use it for tasks that need perfect precision without human oversight.

Independent research says the same thing. The CountQA benchmark (Tamarapalli and colleagues, 2025) tested 15 multimodal models on over 1,500 question and answer pairs built from dense, cluttered real world images. The best model reached 42.9% accuracy, and performance got worse as the number of objects went up.

So a photo of a bin of 300 screws or a wall of phone cases isn't a count. It's an estimate. When the figure ends up on a tax form, you need an actual count.

A practical setup for a Burnaby shop

This workflow uses AI where it's strong and keeps people on the parts where it isn't.

1. Count by scanning, not by photo. Barcode scanning, with a handheld or a phone app tied to your POS, is still the reliable way to capture quantities. AI comes in after that.

2. Photograph labels and count sheets, not piles. For items without barcodes, staff count by hand, write the number on a sheet and photograph the sheet. The model reads the sheet. A person did the counting.

3. Let AI do the reconciliation. Export expected on hand from your POS, add the counted file, and have the model produce a variance list sorted by dollar impact rather than by units.

4. Recount the top variances. Send someone back to the 20 or 30 lines that matter most. That's where most errors hide, and it's a much smaller job than recounting everything.

5. Keep the evidence. Save the sheets, photos and final reconciled file with your year end records. Your accountant will want them.

What about cost? Anthropic's vision docs (checked October 2026) say images are billed in visual tokens, with each token covering a 28 by 28 pixel patch. Their worked example puts a 1000 by 1000 pixel image at about US$1.30 per thousand images on Claude Haiku 4.5, based on that model's published US$1 per million input tokens. For an illustrative example only: a shop photographing 500 count sheets at that size would pay roughly 65 US cents in image input, plus output tokens and whatever software runs the workflow. The model is rarely the expensive part. Setup, testing and human checking are.

Where this doesn't apply

It's worth being honest about fit before you spend anything.

If you have a few hundred SKUs, two staff and a free afternoon, a clipboard will probably beat any AI project on cost. Building and testing a workflow can take longer than the count itself.

If your POS data is unreliable, AI reconciliation will mostly produce noise. Clean up how you enter receiving, returns and transfers first. That pays off whether you automate or not.

If you sell small, high value items like jewellery, watch parts or electronics accessories, every unit matters and estimates aren't good enough. Count by hand or by scan, and keep AI to the paperwork.

If your photos might show customers or staff, think about privacy before you start. StatCan's survey found privacy and cybersecurity were the most common barrier to adoption, and with good reason. Take shelf photos after hours and check where your provider stores and processes data.

Finally, remember the CRA's consistency rule. AI can help you apply your valuation method. It shouldn't be choosing a new one.

The short version

AI won't count your stock for you, at least not to the standard a year end count needs. What it can do is take the paperwork off your team's hands: reading sheets, matching numbers, ranking variances and drafting notes. For a busy shop anywhere in Metro Vancouver, that can still save a long night in the storeroom.

Sources

  • Canada Revenue Agency. Inventory and cost of goods sold. Government of Canada, accessed October 2026. canada.ca
  • Statistics Canada. Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026. Catalogue 11-621-M, 2026. statcan.gc.ca
  • Anthropic. Vision. Claude Platform documentation, accessed October 2026. platform.claude.com
  • Tamarapalli, J. S., Grover, R., Pande, N., and Yerramilli, S. CountQA: How Well Do MLLMs Count in the Wild? arXiv:2508.06585, 2025. arxiv.org

If you'd like a second opinion on whether AI makes sense for your stocktake this year, or whether you're better off waiting, book a free call with Autana Solutions. We're based in Burnaby and New Westminster, and we'll tell you honestly if a clipboard is the better tool.

inventoryyear-endsmall businessBurnabycomputer visionretail

Want an AI employee for your business?

We install a 24/7 AI worker for businesses in Vancouver, Burnaby, and beyond. Book a free Discovery Call.

Book a call →

Keep reading