A workflow, not a feature list
A workflow has a trigger, an input, an output, and an owner. A lead form is submitted. A support email arrives. A teammate asks for a first draft. A file needs to be labelled. Those are workflows. “Add AI” is not.
The KAN pilot picks one of those paths. The published examples are lead qualification, support triage, content drafting, and data processing. The timeline is 2–4 weeks. The published range is $2,000–$4,000.
What the model is allowed to do
Before anyone connects an inbox, write three lists: what the model may answer, what it must pass to a person, and what it must not say. A clinic’s inbox and a software company’s inbox do not share the second list.
The pilot uses Claude or OpenAI. We do not train a private model as part of that offer. If your data cannot be sent to those services, that is a constraint for the brief, and it may mean the pilot should not start.
How to tell if it worked
Count the task as it happens today: how many times a week, and how long a person spends. After the pilot, compare that with what the workflow actually did. A demo is not that comparison.
If the numbers are not better, stop. The map of the process is still useful. A retainer is for a workflow that earned it, not a default next invoice.
Where it meets the rest of KAN
Automation is easier when the form asks the right questions. That form might live on a Launch Site or inside a Web/App Build. Marketing can then treat the qualified leads differently from the raw inbox. Those are separate offers, so the pilot stays small enough to finish.
Companies in the UAE use the same shape. The language of the reply and the system it connects to are part of the scope. The shape does not change.
Written by KAN · Updated 24 October 2026