The chatbot is the easy part

Every SMB I talk to opens with the same ask: a chatbot. On the website, in Slack, or on top of a knowledge base nobody has cleaned up in three years. The chatbot is the easy part, and it is rarely where the value is.

A chatbot is the thinnest layer of the system. You can stand one up in an afternoon with an API key and a prompt that fits in a Slack message. The work is everything underneath it.

Underneath is what data it can see, what it is allowed to write back, which workflow it interrupts and which one it replaces, who owns the prompt after the founder stops editing it, and what happens when an answer is wrong with a customer on the other end. A better model solves none of that.

I think of it as surface versus substrate. The surface is the chatbot, the form, the widget, the thing your team sees in a demo. The substrate is the workflow change underneath. Surface is cheap and fast. Substrate is slow and political, and it is where the value lives.

The teams I have seen win with AI started with one workflow they were tired of: patient intake, sales order generation, vendor security questionnaires. They mapped it end to end and only then picked a surface. Sometimes that was a chatbot. More often it was a form, a script, an email watcher, or a tool nobody outside the team sees.

If you are an SMB founder looking at AI, the question to ask is which workflow your people complain about every week. Start there and the surface follows.