The request we get most often sounds like this: «put a bot on our website». It is a reasonable request. It is also, nine times out of ten, the wrong first move — and the honest answer costs us a sale about as often as it saves the client one.

What actually breaks

An AI agent is a layer that reads company data and acts on it. If that data is unreliable, the agent does not hesitate — it acts confidently on the wrong information, and it does so at machine speed. A human sales manager looking at a lead card with a stale status will pause and check. An agent will not.

So before building anything we look for three specific failures, because they are the ones that make an agent dangerous rather than useful:

  • Leads that are not in one place. If enquiries arrive on Instagram, WhatsApp, the site form and the phone, and land in four different places, an agent cannot see the client's history. It will greet a returning customer as a stranger.
  • Statuses nobody trusts. When «in progress» means five different things depending on who set it, every automation built on that field inherits the ambiguity.
  • A next step that belongs to no one. If a deal has no owner and no scheduled action, automating the follow-up just means the wrong person gets reminded.

What the audit produces

The AI CheckUp is not a slide deck. It ends in a table — the Transformation Map — where every line is actionable: the problem stated concretely, its financial effect, the automation opportunity, the tool proposed, the integration needed, the implementation priority, and the KPI that will prove it worked.

That last column is the important one. If a layer cannot name the number it improves, we say so instead of selling it. A fair share of automation ideas do not survive that column, including some of our own.

Then the order matters

CRM foundation first, communication integration second, agents third. Not because it is tidier, but because each layer is only as honest as the one beneath it. Build in that order and the agent has clean ground to stand on. Build in the other order and you have bought a very fast way to be wrong.