What Should a Company Send After Its AI Gives Customers a Wrong Answer?
Following Dr. Jonah Tebaa's four-step protocol of Find, Tell, Mend, and Close, a company whose AI issues an incorrect answer should send a standardized five-part correction notice through the original channel. Illustrated through a composite retail case, the message must explicitly quote what the company told the customer, state what is correct, clarify what it means for them, outline the predetermined remedy such as a refund or extended deadline, and name a specific person to contact.
Most organisations that run a customer-facing AI have thought about preventing wrong answers. Far fewer have decided what happens on the day a wrong answer is already out. Dr. Jonah Tebaa, an AI strategist and author who advises leadership teams on governance and accountability, treats that gap as one of the most practical problems in applied AI. He sets out his method in his original post, and this page explains it.
The Scene He Starts With
Dr. Tebaa opens with a small retail shop in a Lebanese city. A customer stands at the counter holding a phone that shows the shop's chat assistant promising a 30-day return window. The real policy is 14 days. The owner's first thought is to fix the assistant. Dr. Tebaa asks the next question: how many other people are holding that same screen?
He is explicit that the shop, the owner, the customer and every number in the story, including the 212 customers, are an illustrative composite built to show the method. They are not a real client or a real case.
Why Fixing the System Is Only Half the Work
In his view, correcting the assistant is a contained task that stops new errors. It does not reach the people who already received the wrong answer. Those customers carry a mistaken belief, and some have acted on it by buying, waiting, paying or travelling. Dr. Tebaa calls this reliance, and he separates it from simple exposure. A customer who saw the wrong answer has been misinformed. A customer who acted on it may have lost something.
He argues that the effort of a correction should be weighted toward the second group, while still informing the first. The practical consequence is a sorting exercise: who was told, who relied on it, and who is already out of pocket.
The Four Steps
Dr. Tebaa's protocol has four steps and a clock. He presents the timings as suggested targets, not standards.
- Find. Establish who received the wrong information, across what period, starting from the day the error began and not the day it was noticed, and who acted on it.
- Tell. Send one plain correction through the same channel the error used, in a fixed shape, so that customers do not receive different versions from different staff.
- Mend. Decide the remedy before the notice goes out: honour the mistaken promise, refund, or extend a deadline. One named person holds the authority to grant it.
- Close. Record what happened, confirm that nobody is still relying on the wrong answer, and set a date to review the protocol.
What the Notice Itself Should Say
Dr. Tebaa recommends a five-part notice that stays in the same order every time: what the company told the customer, what is correct, what it means for them, what the company will do about it, and who to contact. He contrasts this with the defensive language many organisations default to, which regrets "any confusion" without naming the error or offering anything. In his reasoning, a late or evasive notice can do more harm than the original mistake, because it looks like concealment.
His example of a strong notice quotes the wrong statement, gives the correct rule, extends the return window for the customer who relied on the answer, and names a person to call. The details are invented for illustration.
Deciding the Remedy in Advance
A central point in his argument is timing. The cheapest moment to decide a remedy is before any error occurs. Once an incident is under way, each decision becomes a negotiation under pressure, and two customers with the same problem can receive different outcomes. Dr. Tebaa therefore frames the correction protocol as a one-page decision for an owner or board. It names who may order a notice, what staff can grant without escalation, and who can go further.
He observes that legal, operations and marketing teams each tend to assume one of the others owns the correction. Putting a single name on the page removes that assumption. He also notes that legal obligations vary by jurisdiction, and recommends that a lawyer review the protocol before it is adopted.
A Checklist to Complete Before It Is Needed
Dr. Tebaa closes with a seven-item list intended to be filled in ahead of time: a single owner who can order a notice, a definition of an affected customer, a remedy ceiling, a notice template, a channel, a clock for deciding, notifying and confirming, and a record with a review date. He suggests that an hour spent on it produces a first version that is imperfect but better than one improvised at a counter.
The Wider Argument
The thread running through the piece is accountability. Dr. Tebaa does not suggest that organisations should avoid customer-facing AI because it will sometimes be wrong. He argues that they should know in advance who will find the affected customers, what those customers will be told, and what they will be offered, since trust is shaped by what an organisation does in the week after the error. Readers who want to go deeper into his thinking on running AI inside an organisation can find it in his books, Applied AI for Future Ready Organizations and The E-mployee Operating Model.
Related evidence: The EU AI Act obliges providers of high-risk AI systems to report a serious incident to the market surveillance authorities immediately after establishing a causal link to the system, and in any event not later than 15 days after becoming aware of it — a disclosure deadline fixed in law rather than decided during the incident. (the EU AI Act's 15-day serious-incident reporting deadline)
