AI News · Edition 72 ·
The real AI win is boring: back-office work, checked by a person
A franchisee's cash-reconciliation story shows where AI helps today. Plus: chatbot answers that can be gamed, and Microsoft's Copilot rethink.
The one that matters
OpenAI says a franchise owner cut cash reconciliation time 94% with ChatGPT. Read the fine print first
What happened
OpenAI has published a case study about an independent franchisee who runs Schlotzsky's and Cinnabon outlets. OpenAI says the owner used its ChatGPT Work product to cut the time spent reconciling cash by 94% and the time spent repricing menus by 95%. It also says the owner built a workflow that could free up $100K in cash flow.
Keep two things in mind. First, this is OpenAI's own write-up, on its own site, about its own product, and no independent outlet has checked the numbers. Second, the $100K is money the workflow could free up, not money already recovered. The published summary does not explain how that figure was worked out.
What it actually means for your business
The useful part is not the percentages. It is the kind of work involved. Cash reconciliation and repricing are dull jobs that follow fixed rules and live in spreadsheets. You match what the till says against what reached the bank, flag the gaps and update a price list when costs move. Almost every small business has its own version:
- A 600-student school matching fee receipts against bank statements every month and chasing the families whose payments don't line up.
- A 30-bed clinic checking insurance or government-scheme claims against what was actually paid out.
- A workshop with 40 staff updating quotes for a few hundred parts after a supplier raises steel prices.
- A two-counter shop comparing the day's POS report with the cash in the drawer and its UPI or Interac settlements.
This is where general-purpose AI tools earn their keep today. The data is orderly, the rules can be written down in plain language, and a person can check the result in minutes. If your accountant or office manager spends a full day a week on this kind of matching, cutting that time a lot is a realistic goal. Treat 94% as the best case, not what you should expect.
What it does not mean
- It does not mean AI "does your books." Reconciliation is valuable because a responsible person signs off on it. Say a tool matches 98 of 100 transactions correctly and confidently gets the other two wrong. If nobody checks, those two become your problem.
- It does not mean your result will match. Vendors publish their best customers, not their average ones. Results like this depend on records that are already digital and fairly clean. If your receipts live in paper registers, three WhatsApp groups and someone's memory, the first job is getting the data into one place, and AI won't do that for you.
- It does not prove AI pays for itself everywhere. Consultants and software sellers will likely forward this story as proof that every business should "go AI." One self-reported case published by the vendor is an example, not evidence.
What a sensible owner should do this month
- Pick one task. Choose a single recurring reconciliation or price-update job. Time it honestly for one cycle so you know how long it takes today.
- Use a business account, not a free personal one. Business plans for the main AI chat tools have usually cost around US$25–30 per user per month. Before uploading anything, read the vendor's data terms and confirm your records won't be used to train its models.
- Protect sensitive records. Schools and clinics should hide student and patient names and ID numbers where possible. Before sharing records with any outside service, check your obligations under India's DPDP Act, or Canada's PIPEDA and your province's health-privacy rules.
- Run it side by side for a month. Keep your current method going. Have the person who normally does the job check every result the AI produces, and count the errors as carefully as the minutes saved.
- Decide on evidence. If it saves real hours and makes few enough errors that you're comfortable, keep it and try a second task. If not, you've lost one month's subscription and a few hours.
In short, this is the right kind of AI use: boring, back-office and easy to check. Just don't expect your numbers to look like the brochure.
Also worth knowing
How this briefing is put together
Every morning we read the day’s AI announcements and reporting from the companies themselves and from the technology press, then pick the handful that actually change something for a working business. The analysis is ours and it is written for owners and managers, not engineers. Every story links to its original source — read them, and disagree with us where we’ve got it wrong.