Workflow re-architecture, not chatbots

A chatbot is the answer when the workflow is already healthy. Most operations don't have that problem.

The signal that an ops team needs a chatbot looks like: "people keep asking the same thing and we have a clear knowledge base." That's a real but narrow problem. Surface a search. Done.

The signal that an ops team needs workflow re-architecture looks like: "we have a queue that grows faster than we can clear it, the routing is wrong half the time, and three different people keep handling the same ticket." That's a different problem and a chatbot won't touch it.

Most engagements we see fall in the second category. The team has been told to "add AI" and has interpreted that as "add a chat interface." They're trying to fix a loop problem with an interface treatment.

The loop is the system that decides what work happens, who owns it, when it's done, and what counts as wrong. AI moves the needle on a loop when it can do the judgment that someone is currently doing twenty times a day, encode the rule that's currently in someone's head, and route the edge cases somewhere a human catches them before they hit a customer.

A chatbot does almost none of that. It's a front door. The loop runs behind the front door. Putting an AI front door on a broken loop makes the loop slightly faster at producing wrong answers.

So when we get asked to "build a chatbot," the first conversation is usually about whether that's actually the work. Sometimes it is. Usually it isn't.