A fast-growing enterprise with an aggressive sales culture, a recently implemented CRM, and a leadership team that loved dashboards. The problem: the dashboards were beautiful fiction.
The $300M ghost
It started with a seemingly innocent question from a sales leader: "Can you help me understand why our win rate has dropped 15 points?"
I pulled the standard report. Everything looked normal. Pipeline volume was up. Average deal size was stable. The stage progression model showed opportunities moving predictably from Discovery to Closed Won.
But the win rate was down, and the leader had a gut feeling something was off.
I started digging into the underlying records. Not the aggregate metrics. The raw opportunity data. And I found something strange.
Over 400 opportunities, representing more than $300M in pipeline, had been created in the CRM with a close date exactly 90 days from creation. To the day. Not 89. Not 91. Exactly 90.
That is not how real sales cycles work. That is how a default value works.
The incentive architecture
What I found was a textbook example of metrics driving behavior. The compensation plan rewarded quota attainment with accelerators, and the pipeline dashboard leadership reviewed weekly showed coverage ratio: how much pipeline each rep had relative to quota.
Reps learned quickly that the system measured pipeline quantity, not pipeline quality. So they gamed the input. Opportunities with optimistic close dates and inflated values, moved through stages on schedule. The dashboard stayed green. The forecast stayed "on track." The deals were nowhere near ready to close.
This had been going on for a long time. The CRM was not a record of customer conversations. It was a record of what reps needed the system to say to keep their managers happy.
Rebuilding the signal
The technical fix was conceptually simple and organizationally hard: replace self-reported pipeline with signal-based pipeline.
- Email and calendar integration verified that reps were actually talking to the people they claimed to be talking to.
- Product usage signals validated that prospects were actually engaging with the platform.
- An automated scoring model weighted opportunities on behavioral evidence, not stage names.
- A rep-level forecast confidence score became part of the management conversation.
The harder part was changing the questions leadership asked. Not "how much pipeline do you have?" but "what is the confidence level of your pipeline, and what signals support it?"
The confrontation
The first time the signal-based view was presented to sales leadership, the room pushed back hard. Some pipelines shrank dramatically overnight, and the instinct was to blame the system: "These people always hit their numbers. The model must be wrong."
I said very little in those meetings. The data was the data. And it revealed something more interesting than gaming: some of the strongest performers really did hit their numbers, just through a completely different mechanism than the CRM claimed. Their real work, deep relationships worked largely outside the system, was invisible to the old dashboard and finally visible to the new one.
Some of the loudest skeptics became the strongest advocates, because once the system measured reality, it gave people credit for work they were already doing. Managers could see the real patterns. The company could learn from them.
What I learned
The most dangerous dashboards are the ones that look right. The ones that confirm what leadership already believes. The ones that stay green long after reality has turned red.
If your data system can be gamed, it will be gamed. Not because your people are dishonest, but because your incentive structure rewards the wrong inputs.
The job of a data leader is not to build better reports. It is to build systems that make gaming more expensive than honesty: metrics that are hard to fake, signals that are hard to manufacture, and consequences that flow directly from evidence.
The pipeline was never real. The question was whether anyone wanted to know.
The condensed version of this story is in the case study The pipeline that wasn't there. The full build details are in the gated version, shared on request.