A new VP of Sales asked a simple question: why has our win rate dropped 15 points in six months? The dashboards said everything was fine. That was the problem.
The problem
Digging into raw opportunity records instead of aggregate metrics, I found over 400 opportunities, more than $300M in pipeline, created with a close date exactly 90 days out. To the day. Real sales cycles do not do that. Default values do.
The mechanism was incentives, not dishonesty. Compensation accelerators rewarded pipeline coverage, the dashboard measured pipeline quantity, and reps learned to feed the system what it measured. The CRM had become a record of what people needed it to say.
The approach
- Signals over self-reporting. Email and calendar integration verified that claimed conversations were actually happening. Product usage signals validated real prospect engagement.
- Evidence-weighted scoring. An automated model scored opportunities on behavioral evidence rather than stage names, producing a rep-level forecast confidence score.
- A different management conversation. Leadership stopped asking "how much pipeline do you have?" and started asking "what signals support your confidence?" That change mattered more than the model.
The outcome
The first review meeting was rough: a top performer saw his pipeline cut 60% and walked out. The data held. His real performance came from deep relationships he worked outside the CRM, and once the system started measuring reality, it gave him credit for it. He became one of its strongest advocates.
For the first time in years, Sales and Finance brought the same number to the same meeting. Full narrative, with the uncomfortable parts left in, in the essay The pipeline that wasn't real.
If a data system can be gamed, it will be gamed. The job is to make honesty cheaper than gaming.
What this means for you
If your forecast keeps surprising you in the last month of the quarter, audit the inputs before you blame the model. The tell is usually sitting in the raw records. I know where to look.