Responding to a Complaint About Staff, With AI

When a customer complains about one of your staff, AI is useful for exactly three things: structuring your reply to the customer, keeping your internal notes complete and neutral, and checking your own tone before you send. It must not decide whether the complaint is true, what the employee intended, or what should happen to them. Those are judgements with employment law consequences, and a model has no standing to make them.

That line is not pedantry. A complaint about a named person can end in a grievance, a tribunal, or a reference dispute, and anything you wrote down becomes evidence, including the parts a model drafted for you.

Before you open a model: the data protection problem

The instinct is to paste the complaint in and ask what to do. Stop and consider what you would be pasting: a named employee, an allegation about their conduct, and a customer's details. That is personal data about two people, one of whom is a member of your staff, and the allegation is sensitive to their livelihood.

The practical fix takes thirty seconds:

• Replace names with roles before pasting. "The technician who attended on Tuesday" instead of the person's name. "The customer" instead of theirs.

• Remove addresses, phone numbers, account numbers and anything else that identifies either party.

• Keep the facts and the sequence, which is all the model needs to help with structure and tone.

• Check whether your AI tool trains on your inputs. If you do not know, assume it might and anonymise harder.

Anonymising also improves the output, because it stops the model inventing character detail about a person it knows nothing about. On the broader question of where AI advice stops being safe in staffing matters, see is it safe to use AI for HR advice.

Step 1: separate the three jobs

Most mishandled complaints come from doing these at once. Keep them apart and in order.

The middle row is non-negotiable. An employee hearing that a decision about their conduct was informed by an AI assessment of a customer's account is a grievance in itself, and in most jurisdictions you owe them a fair process with a human judgement in it.

Step 2: the acknowledgement, which is where AI helps most

The acknowledgement has to do something difficult: take the complaint seriously without conceding facts you have not checked, and without throwing your employee under a bus to placate a customer. That balance is genuinely hard to write when you are annoyed, which is precisely when you are writing it.

Help me write an acknowledgement of a customer complaint about a member of
my team. Facts I know: [sequence of events, roles not names, no personal data]

Requirements:
- Thank them for raising it and confirm I am looking into it.
- Do NOT accept or deny the allegation. I have not spoken to the staff member yet.
- Do NOT apologise for the conduct alleged. I may apologise for their experience
  of the situation and for the inconvenience.
- Do not criticise or speculate about the employee.
- Say what happens next and by when.
- Under 150 words, plain, calm, not corporate.

The distinction in the third bullet is the one that matters legally and is the one most templates get wrong. "I am sorry this was your experience and I am looking into it" is safe. "I am sorry my technician was rude" is a finding of fact you made before asking him.

Step 3: your notes, which will matter later

This is the unglamorous part that protects everyone, and the part AI is quietly excellent at. After each conversation, write the record immediately and have the model check it for neutrality rather than write it for you.

1. Write your own rough notes: date, time, who was present, what was said, what you agreed.

2. Ask the model to flag anything in your notes that is a conclusion rather than an observation, and anything loaded or judgemental in your wording.

3. Rewrite those lines yourself. "He was aggressive" becomes "He raised his voice, per the customer's account, which he disputes."

4. Keep the customer's account and the employee's account clearly attributed and separate. Never merge them into one narrative.

Point four is the single most valuable habit here. A merged narrative reads as though you decided whose version was true, even if you did not, and that is exactly the impression you cannot afford if this escalates.

What never to ask it

• Whether the complaint is true, or who is more credible. It cannot know, and it will produce an answer anyway.

• What disciplinary action is appropriate. That depends on your contracts, your local employment law and the employee's record. Ask an employment adviser, not a chatbot.

• To write the disciplinary letter from the complaint. Skipping the process between those two documents is the error, and the letter being well written makes it worse rather than better.

• To draft your reply to the employee before you have heard their side. Even asking shapes your thinking before the conversation that is supposed to inform it.

If a complaint turns out to be part of a pattern rather than a one-off, that is a management issue with its own causes, and often a retention issue. Our piece on using AI to reduce employee turnover covers the data side of spotting patterns. If the volume of inbound complaints is itself the problem, triage before escalation is the lever, as in automating customer support with AI.

For the rest of the small-business admin that AI can genuinely take off your hands without this much care required, start at AI for small business.

Frequently asked questions

Can I use AI to decide if a complaint about an employee is justified?

No. Beyond being unreliable, it denies the employee a fair process. In many jurisdictions a decision about conduct must involve a human weighing both accounts, and an AI-influenced finding is a straightforward grievance.

Should I tell my employee a complaint was made?

Yes, before you investigate it, and you should hear their account before forming a view. Handling a conduct complaint without telling the person is the fastest route to a procedural failure.

Is it safe to paste a complaint into ChatGPT?

Not as written. Remove names and identifying details first, for both the customer and the employee. The structure and sequence are what the model needs; the identities add nothing and create real exposure.

Can AI apologise to the customer on my behalf?

It can draft an apology for their experience, which is appropriate at the acknowledgement stage. It should not draft an apology for the alleged conduct before you have investigated, because that is a finding of fact and it is not yours to make yet.