Where AI earns a place in the workflow

October 28, 2026 · Chris Coulson

Coulson Solutions card listing Draft, Sort, and Flag as the jobs a model should do

AI is easy to add and easy to regret. We use it when it removes a repeatable step that a person is doing by hand, and when someone can still check the result.

The bar

A model earns a place in the system when all three are true:

The input is something the business already collects.
The output is a draft, a sort, or a flag, not an unsupervised decision.
A person can tell, quickly, whether the result is usable.

If a step needs judgment that the company is not willing to review, it stays with a person. If a step is just retyping from one box into another, it is a candidate.

What this looks like in a product

In a custom system the model is a worker behind a button or a queue, not a chat box glued to the side of the page. It might:

Draft the first version of a customer update from the job record
Group incoming requests so the right person sees them
Point at rows in a report that do not match the usual pattern

The surrounding software is still the product. It holds the record, the permissions, and the history. The model does not get a private copy of the business.

What we will not bolt on

We do not add a model to make a proposal sound current. If the workflow map from starting with the work itself does not show a repetitive step, there is nothing for AI to take. Build the record first. Add the model when the record is trustworthy.

We talk through builds like this on the Coulson Solutions YouTube channel.

Want a second set of eyes on a process you think is ready for that? Start a conversation.