When the Agent Gets It Wrong: What Good Recovery Looks Like

A bakery owner opens her inbox on a Monday morning and finds a customer asking why the quote for a wedding cake is half the usual price. The reply went out on Friday evening, drafted by the system that handles her inbox, and nobody checked it before it sent. The price was wrong — a number pulled from the wrong line of an old price list. The customer is not angry, just confused, and the owner spends her morning fixing a mistake she did not make, made by a tool she trusted.

This article was drafted with AI assistance. It is written in plain English for people who run businesses, not for people who build software, and it makes no claim about any particular product. It is about what happens when a system gets something wrong, and why the recovery matters more than the mistake.

Mistakes are certain; surprises are not

Every system makes mistakes, and that includes the human ones. A tired member of staff quotes the wrong price. A supplier sends the wrong part. An owner misreads an order. Nobody runs a business for long without getting something wrong, and nobody expects a tool to be perfect either. What people do expect — from staff, from suppliers, and from software — is that a mistake is caught before it reaches a customer, and that when one does get through, somebody notices quickly and fixes it properly.

That is the real test of any system that does work on your behalf. Not whether it ever gets anything wrong. It will. The test is what happens in the minute after it goes wrong. Does it stop and ask, or does it carry on as if nothing happened? Can you see what it did? Can you undo it? Those three questions matter more than any promise about accuracy, because accuracy is a claim about the good days, and recovery is what saves you on the bad ones.

A system that sends things without a human check is a system where a wrong number becomes a wrong promise before anyone has had a chance to look at it. A system that drafts and waits is a system where the same mistake is a draft you delete, not a customer you have to ring. The mistake is identical. The damage is not.

Three things that make a mistake small

A mistake stays small when three things are true.

The first is that it is caught early. The cheapest place to catch an error is before anything leaves the building — in the draft, in the number, in the message waiting for a yes. Every step between the mistake and the customer is a chance to catch it, and every step you remove is a chance you give up. That is why the approval step matters: it is not bureaucracy, it is the last fence before a mistake becomes a promise.

The second is that it is visible. When something goes wrong, you need to be able to see what happened: what the system read, what it decided, and what it sent. A mistake you can see is a mistake you can fix. A mistake hidden inside a system you cannot inspect is a mystery, and mysteries cost far more than mistakes, because you spend your time guessing instead of fixing. Ask any tool: can I see the record of what it did? If the answer is no, the tool is asking you to trust it blindly, and blind trust is the most expensive kind.

The third is that it is fixable. A wrong email needs a correction, sent quickly and honestly. A wrong number needs a new number, and a call to the person who received the old one. A wrong order needs cancelling before it ships. The tools that handle this well make fixing easy: the message can be recalled or corrected, the record shows exactly what went out, and the correction goes to the same person, not into a void. The tools that handle it badly make you start over — retyping the whole thing, hunting for the customer's details, hoping you remember what went wrong.

The apology matters more than the error

Here is something owners already know from experience: customers forgive mistakes far more readily than they forgive silence. A customer who gets a quick, honest correction — "we quoted the wrong price, here is the right one, and we are sorry for the confusion" — usually stays a customer. A customer who discovers the mistake themselves, days later, and has to chase you for an explanation, usually does not.

That is worth remembering when you judge any system. The question is not only how often it gets things right. It is also how gracefully it handles being wrong. Does it flag its own uncertainty, or does it state guesses as facts? Does it make it easy to say sorry quickly, or does it leave you to discover the problem from an angry customer? A system that helps you recover well is worth more than a system that claims it never needs to.

There is also a quieter benefit. When mistakes are caught, seen, and fixed properly, they stop happening the same way twice. The wrong price list gets corrected. The rule gets written down. The draft gets a second look. Every honest fix is a small improvement to the way the business runs. A mistake that is hidden, denied, or blamed on someone else teaches nobody anything, and it will happen again.

What to ask before you trust any tool

If someone offers you a system that will do work on your behalf, the brochure will tell you what it does when everything goes right. Ask instead what happens when something goes wrong.

First, what does it do when it is unsure — does it stop and ask, or does it guess and carry on? Second, can you see what it did, in plain language, so you can check it? Third, can you undo it, quickly, before a customer is affected? And fourth, when a mistake does get through, does the system help you fix it — or does it leave you to clean up alone?

A tool that answers all four clearly is worth a closer look. A tool that cannot answer them is not ready for your business, whatever its accuracy figures say. Accuracy is measured on good days. Your business is judged on the bad ones.

The short version

Every system gets things wrong sometimes, and that is not the problem. The problem is a mistake that reaches a customer before anyone sees it, that nobody can explain, and that nobody can undo. A mistake stays small when it is caught early, when it is visible, and when it is fixable — and when the business that made it says sorry quickly and honestly. Before you trust any tool with your customers, ask it what happens when it is wrong. The answer tells you more than any promise about how often it is right.

Ready-made AI agents for small businesses, built in public.