Most Sales Forecasts Are Educated Guesses. Here's What It Took to Change That
Better sales forecasting accuracy rarely comes from a better spreadsheet. It comes from changing what the forecast is built on, and who trusts it.
For most of my career, I sat in the room where the quarterly number got decided. Sales leaders brought their calls. Finance brought the plan. Someone reconciled the two in a spreadsheet late on a Thursday night, and the result went to the CEO and the board as "the forecast."
Everyone in that room knew it was a negotiation dressed up as a prediction. Reps sandbagged. Managers added a cushion. Finance applied a haircut. By the time the number reached the top, nobody could tell you what it was actually based on.
A few years ago, I led a team that set out to change that. We built an AI-driven sales forecast and put it in the hands of the field: not a pilot for a few analysts, but something reps and managers used every week to run their business. It worked. The forecast got measurably more accurate, and more importantly, people started trusting it.
The numbers belong to that company. The lessons travel, and the biggest one surprised me: the model was the easy part. Here's what actually made the difference.
Why sales forecasting accuracy breaks down
Most forecasts fail for reasons that have nothing to do with math.
They're built on opinion, not signal. A rep's commit is a judgment call shaped by incentives. If missing the forecast costs more than beating it, the forecast drifts low. Every layer of management adds its own adjustment on top.
They live in silos. Sales forecasts bookings. Finance forecasts revenue. Operations forecasts demand. Each team uses different data, different definitions and different timing, so the numbers never quite tie.
They describe the past. Most "forecasting" is really trend extension: last quarter plus a growth rate. It tells you where you've been, not what's changing underneath.
Start with the signal, not the algorithm
The first thing we did wasn't choosing a model. It was finding the data that actually predicted revenue.
We looked at what happened in the field before an order landed: activity patterns, account history, pipeline movement, seasonality, and the timing between first engagement and close. Some of the signals everyone assumed mattered turned out to be noise. Others nobody tracked turned out to be strong.
That work took longer than building the model. It also forced conversations between sales, finance and IT about what each field meant and where it came from. Those conversations were worth as much as the forecast itself.
If you're starting this journey, spend your first weeks here. A sophisticated model on top of messy, inconsistent data will produce confident, wrong answers.
Build it into the rhythm, not next to it
The fastest way to kill an AI forecast is to publish it as a separate report. People will glance at it, compare it to their own number, and keep using their own number.
We built the model's forecast into the existing operating rhythm: the weekly pipeline review, the monthly forecast call, the quarterly business review. It showed up as a reference point right next to the human commit, account by account and region by region.
That changed the conversation. Instead of "what's your number?", managers started asking "the model sees this account differently than you do. What do you know that it doesn't?" Sometimes the rep knew something real. Often the gap pointed to a deal that needed attention.
Adoption is the real project
Getting a field sales organization to rely on a forecast they didn't build is a change-management problem, not a data-science problem.
Three things mattered most:
Explain the drivers. People don't trust a black box. Showing the main factors behind each account's forecast made it something reps could argue with, and argument is the first step toward trust.
Track accuracy openly. We measured the model and the human forecast side by side every period and shared the results. Over time, the record spoke for itself.
Keep humans accountable for the number. The model informed the forecast; it didn't replace ownership. Leaders still owned the commit. They just had a better starting point.
The real test came when we finalized the forecast with the CEO and CFO for guidance. A number only holds if the GMs and VPs accountable for hitting it stand behind it. Once we showed them how the model performed at the level they cared about, the conversation shifted. It still had puts and takes on individual accounts, but at the regional level it was landing within 3%. GMs began telling the room, "We believe in the model." For the CEO and CFO, that was a huge win and a real confidence boost: a forecast the people who own the number had already bought into.
What better accuracy actually changes
A more accurate forecast isn't just a better number on a slide. It changes decisions downstream.
Finance can plan spending with more confidence. Supply and demand planning can stop over-building safety stock. Sales leaders get earlier warning when a region is drifting, while there's still time to act. And the forecast call gets shorter, because less time goes to defending numbers and more goes to the deals that matter.
The takeaway
If your forecast is mostly negotiation, don't start by shopping for an AI tool. Start by asking one question in your next forecast review: what is this number actually based on? If the honest answer is opinion layered on opinion, you've found where to begin.
This is the work we do at 108 Performance Partners: helping sales and finance teams plan and forecast as one.