Your AI made a promise. Who owns it?

A customer asks the chat window on your website whether their order can arrive before the weekend. The assistant says yes.

Nobody at your company said that. Your business did.

AI now answers customers, drafts emails, books appointments and takes small actions on its own. Each of those can carry a promise: a date, a price, a fix, a callback. The promise is real to the customer. The question is who inside your business knows it was made, and who makes sure it’s kept.

Most companies haven’t decided

Large companies are finding this out first. In a 2026 survey of 750 senior technology leaders, only 7.2% of organisations had a named person formally accountable for how their AI agents behave. Grant Thornton surveyed nearly 1,000 US business leaders the same year and found that 78% lacked full confidence they could pass an independent AI governance audit within 90 days. They call it the AI proof gap.

Small businesses feel it differently. 62% of small business owners use AI to run their business, and 86% say it matters that customers can still reach a human. Owners already sense the risk. They just haven’t drawn the line yet.

A promise is a promise, whoever makes it

You don’t need a new tool to fix this. Use the same discipline you’d want for any commitment your business makes:

1. Find where AI speaks for you. Website chat, email drafts, booking tools, quoting, anything that replies to a customer without a person reading it first.
2. Decide what it’s allowed to promise. Dates, prices, refunds, exceptions. Write down which ones AI can confirm and which ones it must hand to a person.
3. Record what it promised. If the assistant told a customer “Friday”, that needs to land somewhere your team will see it, not stay buried in a chat log.
4. Give every promise one owner. Not “the team” and not “the AI”. One named person who is responsible until it’s done.
5. Close it with proof. A promise is finished when there’s evidence it happened and the customer has been told, whatever the task list says.

This is the ERMODA Check applied to a new kind of speaker: what was expected, who needed to agree, who owns it, what counts as done, and whether the person relying on it knows.

For an owner-run business

Start with the one place AI talks to customers. Read a week of its conversations and highlight every sentence that commits you to something. That list is your first set of rules.

For a larger firm

Put a name next to every AI agent that can act or reply externally, the same way you would for a vendor or a new hire. Then ask your teams to show you, with evidence, three promises an agent made last month and how each one ended.

AI will make more promises on your behalf every year. The businesses that do well with it will be able to show what it promised, and prove it was kept.