Why are sales forecasts so often wrong?

Sales forecasts are often wrong because they are built on data entered by salespeople, with probabilities nobody has measured and close dates that keep slipping. The formula is rarely the problem. The problem is that deals stay in the pipeline too long, that each stage gets a fixed probability nobody has checked, and that the value in the CRM differs from what is eventually signed and invoiced. On top of that comes behaviour: a forecast that also serves as a target or a performance measure gets steered by the people who produce it.

How do you recognise a forecast that is structurally wrong?

A forecast that misses now and then is normal. A forecast that keeps missing in the same direction has a cause. Most B2B companies recognise one of these patterns:

• The quarter opens with a forecast that hits the target. In week six it is revised. In week twelve it turns out ten or twenty percent lower.
• A large share of the "commit" deals does not close in the quarter, but in the next one, or never.
• The sales forecast and the finance revenue figure drift apart, and nobody can say exactly why.
• Some salespeople are always too high, others always too low, and everyone knows it, but the forecast is never adjusted for it.

If you see one of these patterns, the error is in the process, not in the outcome of a single quarter.

Why is your pipeline forecast wrong?

1. Deals stay in the pipeline too long

Marking a deal as lost means admitting defeat. Leaving it open costs nothing, at least not in the short term. So deals stay open. They get a new close date, a note saying "customer will come back to us", and they keep counting in the pipeline.

The result is a pipeline that is larger than the real chance of revenue. If your weighted pipeline is the basis of your forecast, the forecast grows with every deal that is not closed out. The article how do you spot an unreliable pipeline gives the signs that help you find those deals.

2. Close dates slip

A salesperson sets a deal to "closes end of March". At the end of March it moves to the end of April. Then to the end of May. In every forecast, the deal sits in the current period.

A deal that has already been pushed back three times has a different probability from one that has never moved. Most forecasts treat them the same. A simple measure: count how often the close date has been moved and lower the probability with each move, or take deals out of the current period's forecast after a set number of moves.

3. Stage probabilities are chosen, not measured

In many CRMs, every stage has a default probability. "Qualification" 10 percent, "demo given" 30 percent, "quote sent" 50 percent, "negotiation" 75 percent. Someone set those percentages at some point, often when the CRM was implemented, and they are rarely checked.

When you look at your own history, the real percentages are almost always different. Sometimes lower, sometimes higher, and sometimes so different per product or per salesperson that a single percentage per stage makes no sense. A forecast using unmeasured percentages has a built-in error that returns every period. How to calibrate them is covered in forecasting from CRM data.

4. Deal values are too high

A salesperson enters a deal at the value of the first quote. Then comes negotiation: a discount, a smaller scope, longer payment terms. The signed value is lower, but the CRM is not always updated. Or it is updated with the total contract value over three years, while the forecast works per quarter.

You only see that gap when you put won deals next to actual invoicing. The article CRM forecast vs actual revenue shows how to measure it.

5. The forecast doubles as a performance review

If salespeople are held to their forecast, or if a low forecast leads to awkward conversations, they start steering the forecast. Some hold deals back to have a pleasant surprise later. Others put deals in the forecast too early to relieve the pressure. Both behaviours are rational, and both make the forecast unusable.

This is not a character problem, it is a design problem. As long as the forecast is at once a prediction, a target and a performance measure, it will be good at none of the three.

6. Revenue after the deal is forgotten

A sales forecast usually stops at the won deal. But a won deal is not yet revenue. It has to become an order or contract, be delivered, invoiced and paid. If something goes wrong there, the sales forecast closes and the revenue does not. The article why pipeline is not revenue goes into this in more depth.

For companies with many multi-year contracts or projects, there is the added factor that revenue is spread over a longer period. A EUR 360,000 deal signed in March may be worth EUR 10,000 to March's revenue.

7. The CRM is incomplete

Not every deal is in the CRM. Some salespeople only enter deals when they are almost won. Others keep their own list in Excel. Recurring revenue from existing customers is often not in the CRM at all. A CRM-based forecast then sees part of reality and extrapolates it to the whole.

How do you fix a sales forecast?

A better forecast does not start with a different model, but with removing these causes.

1. Separate forecast, target and performance review. The forecast is what you expect. The target is what you want. The review is about what was achieved, not about how well someone predicted.
2. Measure the win rate per stage on your own history. Take the deals from the past twelve to twenty-four months and count, per stage, how many were eventually won.
3. Count close-date slips. Lower the probability, or take deals out of the current period, after a set number of slips.
4. Clear out stalled deals. A deal with no activity for longer than your usual sales cycle moves to a separate status or is marked as lost.
5. Compare deal value with signed and invoiced value. Measure the average gap and correct the forecast for it.
6. Add the layers sales cannot see. Running contracts and recurring revenue belong in the revenue forecast, taken from contract administration and the accounts.
7. Measure your forecast afterwards. Record each month what you predicted and compare it with what happened. Discuss the deviation, not the number.

Worked example

Worked example: suppose a B2B software company uses default stage probabilities. "Quote sent" is set at 50 percent. In the current quarter there are EUR 800,000 of deals in that stage, so the forecast counts EUR 400,000.

An analysis of the past two years shows that 30 percent of deals in this stage are eventually won, and that the signed value is on average 10 percent below the CRM value. The corrected expectation is then EUR 800,000 times 0.3 times 0.9, or EUR 216,000. For this one stage, the forecast was EUR 184,000 too high, every quarter again.

The figures in this example were chosen to show the mechanism. Your own percentages can only be found in your own history.

Checklist for your next forecast meeting

• Do you know the measured win rate for each stage over the past two years?
• How many deals in the forecast have a close date that has already moved two or more times?
• How many deals have had no activity for longer than your average sales cycle?
• How large was the gap between forecast and actual revenue over the past four quarters, and in which direction?
• Are running contracts and recurring revenue in the forecast, or only new deals?

For a broader understanding of forecasting, from contracted revenue to scenarios, the full overview is in what is revenue forecasting.

Frequently asked questions

Is a forecast that comes in too high always the fault of sales?

No. Often the error lies in how the forecast is set up: unmeasured probabilities, no correction for slipping close dates, no distinction between deal value and revenue. Sales provides the data, but the process determines what happens to it.

How far can a forecast be off?

There is no general standard. More important than the size of the deviation is its direction. A forecast that is alternately too high and too low is honestly uncertain. A forecast that is always too high has an error you can find and correct.

Does a different CRM help?

Rarely. The causes lie in how it is used and in the missing connection with the rest of the business, not in the package. A new CRM with the same habits gives the same forecast.

Can AI improve a sales forecast?

Yes, mainly because a model bases each deal's probability on history instead of a default percentage, and takes signals such as activity and slipped close dates into account. It does not solve the problem of data that is itself incomplete or too optimistic.