Operating question
Before you let an AI tool describe your business or take on a piece of work, decide who will correct it when something changes. This week, that question applies to a new business listing, a freelance project and the promises you ask an AI supplier to put in writing.
1. Give your business listing an owner
On November 10, OpenAI announced Businesses in ChatGPT in beta. Eligible organizations can use Business Hub to claim their profile, manage business details and see how people engage with it. The company says profile claiming is open to eligible businesses where ChatGPT is available. The separate feature that lets an assistant answer on a company’s behalf starts with eligible U.S. ecommerce and online service businesses.
Illustrative scenario. An Ottawa appliance repair shop serves some postal codes across the river but charges a different call-out fee there. Giving customers current service-area information could save a phone call. The tradeoff is another listing to maintain whenever fees or coverage change. A claimed profile also leaves open what ChatGPT may say in its own answers.
First move this week. Pick one person to compare the business name, contact details and service area across your website and existing listings. Give them a single reference to update. If your business can claim a profile, use that same reference to check the new listing.
2. Buy a handoff you can test
In research released November 10, OpenAI and Upwork compared AI use in ChatGPT with client requests on Upwork. AI-related job postings asked for slightly more tasks on average and more often paired tasks that clients rarely requested together before generative AI. The findings describe one marketplace and its postings. They leave open what a comparable project costs or saves for a Canadian business.
Illustrative scenario. A Manitoba distributor hires someone to extract order details from emailed PDFs and enter them in a shared system. A single project can connect the extraction to the review screen so staff see exactly what needs checking. Bundling the work also means one quote can hide separate jobs, including handling duplicate orders, setting permissions and writing handover instructions.
First move this week. Add one acceptance test to a draft brief. Give the contractor a fictional order with a missing price and specify who should see the exception, what they can change and what must stay out of the order system until they approve it. Ask for a price that includes testing and handover.
3. Ask where a supplier reports failures
On November 9, Canada, five other countries, AI companies including Cohere and civil-society groups including LawZero signed the Paris call joint declaration on AI safety and security. It calls for cooperation on evaluating advanced AI systems and sharing information about incidents. The declaration gives Canadian buyers a set of topics to raise with suppliers. It leaves each buyer’s contract and support arrangements to the parties involved.
Illustrative scenario. A Québec wholesaler uses AI to help prioritize customer service requests. A named incident contact and a clear way to pause the tool could help the team act if urgent requests start getting buried. Asking for evidence can slow down a purchase, and the supplier may have little to share about the particular workflow or language mix.
First move this week. Choose one tool that could delay a customer response. Ask the supplier to show where it publishes known limitations and whom your staff should notify if the tool starts misrouting requests. Write down who on your team can switch back to the manual queue while the issue is investigated.
The bigger pattern
Assign the correction before you need it. A listing needs someone to update it. A project needs someone to accept the handoff. A supplier failure needs someone who can pause the affected work and keep customers served.
When did a supplier or contractor last hand you something your team had to correct, and who owned the fix?
