With an audit, almost always. It is a fixed scope piece of work that ends in a written roadmap you own outright. That way you find out what is actually worth building before committing to a build budget, and we find out whether we are genuinely useful to you. Build work is quoted separately once we both know what we are looking at.

We use enterprise model endpoints with zero retention, so nothing you share is used to train a public model. We work to the minimum data needed for the job, we will sign an NDA and a data processing agreement before anything sensitive moves, and if you are in a regulated sector we design around your constraints rather than asking you to relax them.

Anything with a reasonable API, and a fair amount that has not. CRMs, accounting and ERP systems, ticketing, email and messaging platforms, document stores, internal databases and spreadsheets. We are deliberately not tied to one vendor, so the recommendation is based on your situation rather than a partnership we are trying to service.

Then we say so. A good share of what looks like an AI problem is a process problem, a data quality problem, or a report nobody reads. Those are cheaper and faster to fix without a model involved, and an audit that tells you to fix the process first has still paid for itself. We would rather lose the build than sell you something that quietly fails in six months.