What "explainable AI" actually means for a sales CRM

"AI-powered" shows up on almost every CRM's homepage now. It rarely comes with a straight answer to the question that actually matters to the person using it: why does this lead have this score, and can I trust it?

The bar we hold ourselves to

Three rules, applied to every AI feature in iSales — scoring, win probability, nudges, the conversational assistant, and call summaries:

• Every decision is logged with its inputs. Not just the output — the data that produced it, so you can check the model's work later, not just take its word for it.
• A human always has the final say. AI ranks, drafts, and suggests. Nothing it does is irreversible without someone choosing to act on it.
• It's honest when it isn't ready. Win probability doesn't guess from a handful of examples — it only turns on once an org has 3,000 closed leads to train the model against. Below that, we don't show a number dressed up as a prediction.

What this looks like in practice

Open a lead's win probability and you don't just see "91%" — you see the reasons behind it: which behaviours mattered, how recent they were, and how this lead compares to others that actually converted. Open the Root Cause Engine after a metric moves and you get the same treatment: not just "conversion dropped," but which segment — source, product, team, or stage — actually drove it.

Why this matters more in education than it sounds like it should

A counsellor who doesn't trust the AI's ranking will just ignore it and work the list top-to-bottom by gut feel anyway — at which point you've paid for a feature nobody uses. Explainability isn't a nice-to-have on top of the model; it's the only thing that makes the model worth having at all.