An AI help desk for your own team

A lot of my days lately go into building internal AI help desks, the kind employees use for everyday questions. People ask them things in team chat or email the same way they'd ask a coworker. Where's that form? Who handles this kind of request?

I like starting AI inside a company this way. Staff questions are frequent, the answers already exist somewhere, and the people asking will tell you right away when an answer is off. It's a much safer place to learn than putting AI in front of customers on day one.

Start with the questions people already ask

Before building anything, I look at what people ask each other. Chat channels and shared inboxes are full of it. The same handful of questions come up over and over, and the answers usually live in one or two people's heads.

Those questions become the first scope. A desk that answers the top questions well is worth far more than one that tries to answer everything and gets a few of them wrong. For an owner, that list also shows where the business depends on one person's memory.

Answer from knowledge you own

The desk answers from a knowledge base the team writes and keeps current. It covers how the business does things and who owns which kind of request. It doesn't guess from the internet, and when the knowledge base has nothing, it says so.

Keeping that knowledge honest is most of the work. I write the manuals from the running systems themselves, and I've gone into how in write the owner's manual from the running system. When a process changes, the manual changes with it, and the desk's answers follow.

One conversation, one ticket

Each conversation becomes a single ticket, and the desk remembers the whole thread. If someone comes back an hour later with a follow-up, it knows what was already said and what was already tried. Nobody has to explain themselves twice.

It sounds small, but it's a big part of why people keep using it.

Hand-offs with context

When the desk doesn't know, or the question needs a decision, it hands the ticket to the right person. The hand-off carries the question, what the desk found and what it didn't. The person picking it up starts halfway there.

Over time those hand-offs show you exactly what's missing from the knowledge base. Each one is a chance to write down an answer the team will need again.

Trust gets earned

Early on, a person reads every reply before it goes out. That's how I'd introduce AI anywhere it speaks for the business, and I explained the reasoning in AI as an operating layer. As the replies prove themselves on a type of question, that type can move to answering on its own, while the rest stays reviewed.

If a team wants some channels to stay human, I honor that. It matters more than any feature, because people use a tool they feel in control of.

What I watch

I watch how many questions get answered without a person and how much reviewers change before a reply goes out. The edits are the best signal of all, which I wrote about in measuring whether AI drafts are actually good.

If you're thinking about AI for your own staff, I'd start here. Pick the questions your team already asks each other every day and build something that answers those well. By the time you're ready to point AI at customers, you'll already have the knowledge base and the review habits to do it properly.