We’ve sat in enough kickoff calls to know how this conversation usually opens. “We just need someone to plug in some AI.” Three weeks later we’re sitting with the team, looking at a workflow nobody owns, a CRM nobody trusts, and a folder of PDFs that have been waiting for somebody to do something for nine days.
The model isn’t the problem. The model is the easy bit. The hard bit — the bit that decides whether anything gets used — is everything around it. Where does the work come in. Who decides what matters. What’s the source of truth. Who has to sign off. What happens when the system is wrong.
The pattern we keep seeing
Most rollouts don’t fail because the AI is bad. They fail because the workflow underneath is informal, undocumented, and held together by three people’s memory. You can’t automate what nobody’s written down.
“We don’t have a process problem. We have a ‘three-people-remember-it’ problem.” — a finance lead, week one of an audit
What we do about it
We treat the audit as a real piece of work. We map the workflow on paper. We interview the people who do it. We find the part of it that’s actually hurting — not the part that sounds futuristic. Only then do we scope a build.
- If the process isn’t clear, the system won’t be either.
- If the team won’t use it in week four, it doesn’t matter how good week one was.
- If we can’t see the ROI, we won’t scope a build. We’ll tell you straight.
The receipt
Every system we ship comes with a runtime log. You see what it’s doing, when, and what it caught. That’s not a feature — that’s the whole bargain. Trust comes from receipts, not promises.