A customer asks for a service price on WhatsApp. An AI assistant replies with a promotion that expired yesterday. The issue is not limited to one inaccurate sentence. The customer may have acted on it, shared it internally, or started a payment process, while other customers may still be receiving the same answer.
When an AI assistant gives the wrong price, first contain the repeated answer and identify potentially affected conversations. Updating a knowledge file is necessary but does not resolve what happened to people who already saw the incorrect information.
Separate the information fix from customer remediation
Record the exact reply, the approved price at that time, the promotion's effective dates, the channel, and any action the customer took. Preserve the original conversation rather than editing away the error.
Distinguish an incorrect figure with no downstream action from a quote that led to a reservation, and from a case involving payment or an agreement. The appropriate commercial response is different in each case. The assistant should not decide whether a previous price is binding or promise compensation.
Contain the source of the error
Determine whether the reply came from an outdated document, conflicting promotions, or an incorrectly applied conditional discount. Pause or constrain automated price answers for the affected case until the information owner has reviewed the source. A message saying the price requires team confirmation is safer than another confident guess.
Assign an owner to investigate affected conversations and keep a record of the decision. The NIST AI Risk Management Framework recommends documented processes for responding to and communicating incidents and errors, as well as monitoring deployed systems. It is voluntary international guidance, not an Egyptian legal requirement.
Correct the customer message without inventing a new commitment
After review by an authorized employee, a suitable message could be: "We're reviewing the price information shared in our earlier conversation. That response did not reflect the approved price for this situation. Our team will confirm the applicable price, terms, and any impact on your request before we proceed."
Do not automatically declare the old price void or promise to honor it. The decision depends on the specific agreement and what happened after the message. Cases involving payment or contracts may require commercial or legal review.
Illustrative case: an expired discount
An assistant quotes a discounted package price on Tuesday, even though the promotion ended Monday. Deleting the old campaign file does not identify the customers who received the incorrect price. The team reviews conversations after the promotion ended, prioritizes cases involving commitments or payment, contacts affected customers as appropriate, and records the resolution. This is a hypothetical example, not a Mr. AI customer incident.
Test the cause before restoring automation
Test a direct price question, an Egyptian Arabic phrasing, a customer with a previous quote, a bundled-services question, and an expired promotion. Check that answers reflect the approved version and escalate cases that require judgment. Record the test results and the change date. For preventive review, see the post-launch AI QA scorecard and knowledge approval workflow.
Measure the number of affected conversations, time to containment, time to an appropriate human response, and recurrence of the same root cause. A high volume of correct replies does not cancel out one serious error.
What Mr. AI can and cannot do
Teams can use Mr. AI's approved business knowledge, conversation records and summaries, follow-up, and human handoff. Do not assume it has an automatic incident tracker that identifies every customer exposed to a wrong price. That process needs explicit ownership and testing. Start with synthetic incidents, then test reply boundaries and human intervention.
