AI is good at pipeline discipline, not at selling

We have been asked variations of the same question every month for the last year: should we let AI run our sales process? The honest answer is no. AI is bad at the parts of selling that matter — building trust with a person on a call, reading what they did not say, knowing when to push and when to wait.

But AI is excellent at the parts of pipeline discipline that nobody enjoys doing. Looking at 47 open deals on a Monday morning and identifying which 9 have not been touched in 14 days. Spotting that the close date has slipped twice on a particular deal. Noticing that a verbal-commit deal still has no pricing engagement on file. These are the tasks that get skipped by a tired VP Sales on the third week of a hard quarter.

Pipeline Pulse — our weekly automation — does these tasks every Monday at 6am. It runs eight deterministic at-risk rules over the open pipeline, then asks Gemini 2.0 Flash to write a short paragraph for each at-risk deal explaining what the rule found. The rules are the work. The model is the writing.

We are deliberate about this distinction. We do not let the model decide which deals are at risk — that decision is made by code we wrote, that we can audit, that any partner can defend in a board meeting. We let the model do the writing because the writing is the thing that makes the at-risk list readable on a Monday morning.