What ninety days of history tells you before you automate anything
Before an agent touches a deal, the graph should be able to describe your motion back to you. If it cannot, do not proceed.
The most common deployment mistake is enabling automation on day one. It feels like speed. It is actually skipping the only step that tells you whether the rest will work.
The read-only quarter
Connect everything. Let the graph map ninety days backward. Change nothing. Then sit down and read what it believes about how you sell.
- Which stages actually predict close, versus which ones are administrative theatre.
- Where deals concentrate before they stall, and whether that differs by segment.
- Which reps have a motion worth propagating, and which have a territory worth fixing.
Reading the map honestly
You are looking for the places the map disagrees with your intuition. Those are either a data problem or a belief problem, and both are worth an afternoon before either becomes an automated action running against your pipeline.
If the system's picture of your motion surprises you, that is the most valuable output it will ever produce. Do not automate past it.
Then turn on one play
One. The narrowest, most reversible play you have, usually a nudge on deals that have gone quiet past their own baseline. Run it for two weeks. Measure acceptance, not just outcomes; if reps are declining most suggestions, the model is wrong in a way the outcome data will take a quarter to reveal.
Teams that follow this sequence are typically running four or five autonomous plays by the end of the quarter, and, more to the point, they can explain every one of them to a sceptical CRO.
