How do you forecast when you can't trust your pipeline data?

TL;DR

A forecast built on stale deals and half-empty fields isn't a forecast. It's confident fiction. Fix visibility first — clean history, a current number you actually trust, a defensible forward view — then forecast only what has the scale and stability to hold. Below that line, the honest move is not to forecast at all.

What this really is, and why your data betrays you

Most teams treat forecasting as a math problem. It's a data problem wearing a math costume.

Here's the chain. Data entry feels like a tax, so reps route around it. The real pipeline lives in somebody's private spreadsheet, and the CRM gets updated from memory, minutes before the call. Close dates land on the last day of the quarter because that's easier than thinking. Fields get whatever was fastest to type.

Feed a model that, and it doesn't clean the mess. It formalizes it — faster, and in a nicer font.

Fix the data and the forecast lands almost on its own. Skip it and you'll hire, spend, and promise the board against a number nobody should trust.

The data, explained

You don't have to take our word that the CRM is the bottleneck. Validity's 2025 State of CRM Data Management found 76% of organizations say less than half their CRM data is accurate and complete. That's the dataset your forecast is reading from.

Our own work says the same thing from the other direction. On the Deviniti engagement, the fix that moved the needle wasn't a cleverer model — it was data quality: 100,000 records deduplicated and an enrichment-and-routing setup built to handle 500–1,200 trials a month. Clean inputs first. Forecast second.

The ceiling is real even when hygiene is good. B2B forecasting is hard in a way consumer forecasting isn't: fewer deals, more idiosyncrasy in each one, and a market that shifts under you. Which leads to the uncomfortable part.

What that data produces: the case for not forecasting

This is where we part company with most consultancies. They'll build you a forecast no matter what's underneath it. We won't — because a confident forecast on bad data is worse than none. You'll act on it.

You cannot extrapolate three quarters from five custom deals. There aren't enough observations, and one strange deal swings the whole number. Forecasting needs scale and stability. Big companies survive messy data because volume drowns the noise; a 30-person team gets wrecked by a single outlier. And when the market moves — COVID was the loud version, but a launch or a category shift does it quietly — the averages your model leans on drift, and the number flips overnight.

Below that line, a spreadsheet and three honest hypotheses beat any platform. That's not a failure. That's knowing which problem you actually have.

When to forecast, and when to fix the data first

Frequently asked questions

Can you forecast with almost no historical data?

Barely, and you shouldn't pretend to. Use a spreadsheet and hypotheses. If you need something to model against, AI can generate synthetic data while you build the real set — a stand-in, not a forecast.

What's the first thing to fix for better forecasts?

Completeness and process — filled fields, honest close dates, mapped stages. The model is the last step, not the first.

How does one outlier wreck a small team's forecast?

With low deal volume, a single freak deal isn't noise — it's a chunk of the number. Remove it, or weight it by how likely that kind of deal is to recur.

Where we come in

We start every forecasting engagement with a data-and-pipeline audit, and we'll tell you plainly if you're not ready to forecast yet. Book a call →

Related reading

• Why does your sales forecast keep missing? (the full playbook)
• What is weighted pipeline and how do you calculate it?
• Why is our revenue so lumpy and unpredictable?