January 2026 · 8 min read
Why Most GTM Forecasts Are Fiction
If your forecast doesn't change every day, it's not a forecast. It's a report with aspirational formatting.
I sat in a quarterly business review last year where a sales leader presented a forecast that was off by 38 percent. Not from two quarters ago. From the prior quarter. The one that had already closed. The final number was in the system, and his forecast slide still showed the old figure because nobody had updated the model.
Nobody was surprised. Not the CFO, not the CRO, not the analysts. The forecast had been wrong every quarter for two years, and the operating rhythm had quietly routed around it. Finance built their own shadow numbers. The CRO used gut plus pipeline coverage as a sanity check. The board got a range so wide it communicated nothing.
The forecast existed. It didn’t function.
This is not a story about one bad sales leader. This is the default state of GTM forecasting in most mid-market and enterprise companies. The process is real. The ceremony is real. The numbers are fiction.
The structural problem with stage-based forecasting
Most CRM forecasts work the same way. A deal enters the pipeline at Stage 1, progresses through stages, and each stage carries a probability weight. Stage 3 is 40 percent likely to close. Stage 5 is 80 percent. Multiply the deal value by the probability, sum the column, and that is your forecast.
This model has one requirement that almost no organization meets: the stage definitions have to mean the same thing to every rep, in every segment, in every quarter.
They never do.
One rep moves a deal to Stage 3 after a discovery call. Another waits until they have a signed pilot agreement. A third moves deals forward on Friday afternoons because their manager told them pipeline coverage needs to be at 3x by Monday’s review. The stages are the same. The behaviors behind them are completely different.
The probability weights are historical averages that assume the population of deals in each stage is consistent over time. It is not. A Stage 3 deal in Q1 when the pipeline is healthy is a different animal than a Stage 3 deal in Q4 when reps are stuffing the funnel to hit activity targets. Same label, different reality.
So you multiply a subjective stage assignment by a historically averaged probability and call the result a forecast. Then you are surprised when it is wrong.
What a forecast needs to be
A forecast is a prediction about the future that updates as conditions change. That sentence sounds obvious, but it rules out most of what companies call forecasting.
If your forecast is a number that gets set at the beginning of the quarter and revised twice before close, that is a target with a correction process. If your forecast is a weighted sum of pipeline that only changes when a rep manually updates a stage, that is a report card for CRM hygiene, not a prediction about revenue.
A real forecast has three properties.
It moves every day. Not because someone touches it, but because the inputs change. New deals enter. Existing deals show activity or go quiet. Conversion rates shift as the quarter matures. A forecast that holds still while the pipeline moves is a forecast that is already wrong.
It separates signal from ceremony. The forecast should be driven by what is happening in the pipeline, not what a rep claims is happening. Meeting activity, engagement recency, time in stage, buyer-side behavior. These are signals. A rep updating a field to avoid a manager conversation is ceremony.
It expresses uncertainty honestly. A single number is a lie. Every forecast should be a range with a confidence interval. The width of the range tells you as much as the midpoint. A narrow range late in the quarter is a sign of conviction. A wide range early in the quarter is honest. A narrow range early in the quarter means someone is performing confidence rather than measuring it.
Pipeline velocity is the forecast
The shift that changed how I build these systems was treating pipeline not as a snapshot but as a time series.
Instead of asking “how much pipeline do we have,” we started asking “how fast is pipeline moving, and in which direction.” Daily deltas on pipeline creation, progression, and decay. A 90-day moving average on new pipeline generation. Weekly conversion rates by stage, tracked over time rather than calculated once and frozen.
The power of velocity over volume: volume tells you what exists today. Velocity tells you what is likely to exist next quarter. If pipeline generation is trending down for three consecutive weeks, your Q+1 forecast should reflect that regardless of what the current weighted pipeline says. If conversion rates from Stage 2 to Stage 3 are declining, your close forecast needs to adjust even if total pipeline looks healthy.
We built forward-looking models for Q+1 and Q+2 based on trend extrapolation. Not complex machine learning. Moving averages, conversion rate trends, and segment-level decomposition. The math is simple. The discipline of tracking it daily and letting it override the narrative is the hard part.
Because the narrative always wants to be optimistic. The pipeline review always has a reason this quarter is different. And the forecast built on narrative is the one that misses by 38 percent.
Win probability that means something
Most win probability fields in a CRM are either the stage-based default or a rep’s gut feeling entered to satisfy a required field. Neither is useful.
We moved to a model-driven probability that incorporated signals the rep does not control: days since last buyer engagement, number of stakeholders active in the deal, pace of legal and procurement activity, comparison to historical deals of similar size and segment. The model updated weekly without rep input.
This created productive friction. When the model scored a deal at 25 percent and the rep had it at 75 percent, that gap became a conversation. Not about who was right, but about what the rep knew that the model did not. Sometimes the rep had real information, a verbal commitment, a champion who was working behind the scenes. Good. Surface it, document it, and the model improves over time. Other times the rep had optimism and nothing else, and the model did its job.
The goal is not to replace rep judgment. The goal is to create a system where the forecast is the product of structured signals plus human context, not human context alone. Human context alone is how you get 38 percent misses that nobody questions until the quarter is already over.
The organizational resistance
This is the section most forecasting posts skip. The math is not the hard part. The hard part is that accurate forecasting threatens people who benefit from inaccurate forecasting.
A sales leader whose team consistently over-forecasts gets the benefit of perceived pipeline strength without accountability for the number. They look healthy in Monday’s review. The miss shows up three months later, diffused across enough time that no single decision gets blamed.
A segment leader who knows their pipeline is soft has every incentive to delay that signal. If the forecast updates daily and the trend is visible to everyone, they lose the ability to manage the narrative. The “we’ll make it up in the back half of the quarter” story does not survive a dashboard that shows velocity declining for six consecutive weeks.
Building a real forecasting system means telling these people that the number is no longer theirs to manage. It belongs to the system, and the system is transparent. That is a political decision before it is a technical one, and if you do not have executive sponsorship for it, the technical work will get quietly undermined until it stops being useful.
This is the same pattern from the first post in this series. The data part is solvable. The part where people lose control of their narrative is where the real resistance lives.
What this looks like when it works
When a forecasting system functions, the operating rhythm changes.
Monday pipeline reviews stop being performance theater where managers interrogate reps about stage progression. They become signal reviews where the team looks at what the data shows and discusses where human context confirms or contradicts the trend. The meeting gets shorter. The information gets better.
The CFO stops building shadow models. Finance gets a forecast they can plan against because it updates continuously and expresses honest uncertainty. Capacity planning, hiring, marketing spend allocation: all of these improve when the input is a real number instead of a negotiated one.
The board sees a range with a track record. After a few quarters, the model’s accuracy history speaks for itself. A forecast that lands within its confidence interval six quarters in a row is worth more than a forecast that hits the number once because someone sandbagged perfectly.
And the 38 percent miss stops happening. Not because the model is perfect, but because the model updates daily and the miss becomes visible at week three instead of quarter-end. A three-week-old problem has interventions available to it. A quarter-end miss has only explanations.
The line
If your forecast doesn’t change every day, it is not a forecast. It is a report with aspirational formatting.
If your forecast depends on reps manually updating a field, you are measuring CRM compliance, not predicting revenue.
If your forecast is a single number without a range, someone is performing confidence instead of measuring it.
The previous post argued that the real job is building decision systems, not data platforms. Forecasting is the clearest test of whether your decision system works. Because every company has a forecast. Very few of them have a system that tells the truth.