Most founders see AI integration as either a tool they buy for the team or something technical that needs an engineer to explain. Both framings miss the point. At the ฿100M scale, AI integration is mainly a business problem, not just a technology one.

The situation

A distribution business, around ฿120M, had bought every tool the market offered. The team used them. The founder kept asking why the company did not feel any different. He was measuring adoption and seeing usage, but the shape of the business had not moved an inch.

The reason was simple. All of his AI work sat in one layer, the shallowest one. He was buying speed on individual tasks and calling it transformation.

The frame: three layers

1
Task automationAI drafts, summarises, answers routine inquiries. A two-hour task becomes twenty minutes. Real speed, but the shape of the business does not change. Many companies stop here.
2
Process integrationAI changes how work flows. Review steps route automatically, week-long reporting generates overnight, knowledge trapped in heads gets captured and made operational. Gains compound. This is the untouched upside.
3
Decision supportAI surfaces what the business could not see: not what happened, but what will likely happen, and where human judgement actually matters. Rare at this scale. Hard to replicate once built.
AI integration deepens by layer. Most ฿50M–฿500M businesses stop at the first.

How I read it

The question is never which tools to buy. It is which work the tools touch, and how deeply. Layer one is already happening in most businesses by accident. The real value sits in layer two, in the handoffs between steps, where key knowledge still lives in people’s heads: the experienced buyer who knows which suppliers are reliable, the ops manager who can feel which client is about to churn.

The working

We mapped his processes against the three layers and found the pattern almost everyone has.

LayerHis effortActual upside
Task automationHighLow, already captured
Process integrationAlmost noneHigh, untouched
Decision supportNoneReal, but needs layer two first
Where the AI spend was, versus where the upside was.

He had spent eighteen months optimising the layer with the least left to give and ignored the one with the most. Not because he made a bad call, but because layer one is visible and easy to start, and layer two requires you to change how work actually moves, which is harder and quieter.

The move

Start at layer two, and only on the sequences where knowledge is currently trapped in a person. Layer three is worth wanting, but it requires layer two to be solid first. You cannot model what will happen until the record of what does happen is captured and clean.

If your team uses AI but the business does not feel different yet, you are still in layer one. The work that changes the company is in layer two.

The right question to spend a week on is not which tool. It is: which sequence of work, today, still depends on one person remembering something. That is where layer two begins.