When your forecast no longer holds up in front of the board, that is a good time to bring in a fractional chief revenue officer.
A correct total can hide a different quarter
You can hit your sales forecast exactly and still wonder whether you can trust the next one.
Imagine you're forecasting £200,000 in bookings for the quarter ending 30 June. On 1 April, you've committed three named deals. Each goes into the forecast at its full expected value. You expect these particular deals to close, rather than a percentage of a larger pool.
| Deal | In the 1 April commit | Booked by 30 June | What happened |
|---|---|---|---|
| A | £100,000 | £100,000 | Closed as expected |
| B | £60,000 | £0 | Slipped beyond the quarter |
| C | £40,000 | £0 | Lost |
| D | £0 | £100,000 | Closed, but was outside the original commit |
| Total | £200,000 | £200,000 | No difference in the total |
The headline looks great. You forecast £200,000 and booked £200,000. The absolute difference is £0. Divide that by the £200,000 you actually booked and your absolute percentage error is 0%.
Now look at the deals you were counting on. Of the £200,000 committed on 1 April, £100,000 booked that quarter. Half arrived; a different deal made up the rest.
You can account for every pound: £200,000 minus £60,000 slipped, minus £40,000 lost, plus £100,000 added equals £200,000.
B might still close later. Winning D might have been excellent work. Neither changes what you expected on 1 April.
Does it matter? That depends on what you did with the forecast. If you kept particular people free to deliver B, D might need a different team. If you used it to plan cash, bookings were only part of the answer: payment terms and collection dates matter too.
The total tells you how much booked. The original deal list tells you what happened to the work you were expecting. I'd want to see both before trusting the next forecast more.
What did you need the forecast for?
If you're forecasting across a large pool of similar deals, you don't need to pick every winner.
You might expect a certain proportion to close. Some will, some won't. That can still give you a useful total, even when you can't say exactly which customers will buy.
That only works if the pool holds enough qualified deals, which is what your pipeline coverage ratio should show.
Hyndman and Athanasopoulos explain the trade-off in Forecasting: Principles and Practice. Forecasting each part separately gives you more detail, but those smaller parts can be harder to predict. More detail doesn't automatically make the forecast better.11 Rob J. Hyndman and George Athanasopoulos, Forecasting: Principles and Practice, third edition, “Single level approaches”. Bottom-up forecasting's information and noise trade-off.https://otexts.com/fpp3/single-level.html
So there's a fair objection here: if you only needed the total, and your method keeps getting it right, why worry about which deals made it up?
I wouldn't call that a bad forecast. A portfolio forecast predicts the result across a pool of deals. It can do its job even when the winners change. The table above isn't an argument against probability weighting, statistical models or forecasting a total.
What matters is what you were told you could rely on.
Expecting a certain amount from a pool of deals is one promise. Expecting three named customers to sign by June is another. You can't judge them as though they're the same thing.
In the example, the total was right and only half the originally committed value arrived. Both are useful facts. You don't need to blend them into one more impressive-looking percentage.
The difference matters when a particular deal drives a particular plan. If you have a small implementation team, knowing which customer will need them can matter a lot. If you're selling thousands of small transactions, the overall volume might be enough.
Start with a plain question: what were you relying on this number to do? Set a bookings target? Plan a customer's delivery? Work out when cash would arrive? One forecast can help with several decisions, but it needs the right information for each.
The usual sales forecast categories and meeting routine still help. First, though, everyone needs to mean the same thing when they talk about the number.
Save the forecast before the quarter changes
If you want to learn from a miss, keep a copy of what you predicted before you knew the answer.
Your CRM shows what the team thinks now. It might no longer show what they thought when you agreed to hire someone, reserved a delivery slot or committed to the quarter's number.
A deal moves to next month. Its value changes. It drops out of commit. Those can all be sensible updates. But if they replace the old view, you've lost the record you need to check the earlier forecast.
This is a basic forecasting principle: test a prediction against results you didn't have when you made it. Hyndman and Athanasopoulos distinguish that from checking how well a method fits information it already had.22 Hyndman and Athanasopoulos, “Evaluating point forecast accuracy”. Evaluation using observations not used to construct the forecast; definitions of forecast-error measures.https://otexts.com/fpp3/accuracy.html Their guidance on testing forecasts over time keeps that order intact.33 Hyndman and Athanasopoulos, “Time series cross-validation”. Preserving the order of information when evaluating predictions. The CRM snapshot application is the argument made here.https://otexts.com/fpp3/tscv.html
For your team, the practical step is simple: save the old snapshot as well as the updated view.
Good CRM hygiene keeps the updated view accurate, and the snapshot keeps the old one.
Write down when you made the forecast, which period it covers and what you're measuring. Bookings, recognised revenue and cash received are different things. Keep the total, the deals included, each deal's expected contribution and the evidence you had then.
Compare forecasts made at the same point in the period. On the final Friday of a quarter, you know much more than you did on day one. That late estimate might help with a board update. It doesn't tell you whether the earlier number was safe to plan around.
Be equally clear about percentages. Our example divides the error by actual bookings. Dividing by forecast bookings would be a different calculation. If actual bookings are zero, you can't calculate that first percentage; show the difference in pounds instead.
If you don't have an old snapshot, say so. A board pack or dated export might give you part of the picture. Today's deal history can help explain a change, but don't pass it off as the untouched forecast from three months ago.
Ask why the buyer will be ready by that date
A close date is more useful when someone can explain what needs to happen before it.
You can have a good relationship, a demo that went well and an enthusiastic contact without knowing how the purchase gets approved. Those are encouraging signs. They don't tell you whether approval, contracts and signature can all happen this quarter.
A public sales discussion puts the temptation plainly: "It feels great to hype up your pipeline in the team meeting."44 Public sales discussion, “It pays to be paranoid”. The quoted sentence is commentary on pipeline optimism; this article does not rely on the post's earnings anecdote.https://www.reddit.com/r/sales/comments/1dqn1xy/it_pays_to_be_paranoid/ It's easy to feel good about an update and still know very little about the buying process.
Start with the deal whose delay would cause you the most trouble. Ask what has to happen between now and the proposed close date. Who owns each step? What's been confirmed? What are you still assuming?
If procurement hasn't started, why do you expect it to finish this month? If your contact expects approval, has the person who can approve it actually looked at the purchase? A date the buyer has worked through deserves a different explanation from a date the seller hopes for.
You're not asking anyone to predict the future perfectly. A buyer can change their mind after giving you good reasons to expect a sale. Procurement can uncover something nobody could reasonably have seen coming. A slipped deal doesn't automatically mean someone made a bad call.
Ask whether the evidence supported the date at the time.
Perhaps nobody knew who had to approve the purchase. That gives you something to check earlier next time. Or perhaps you knew the approval process and it changed unexpectedly. That might be uncertainty you need to leave room for in your plans.
Either way, you learn more than you would by asking the seller whether they're really confident. This also gives your sales-stage exit criteria a practical purpose: show what you learned about the deal before moving it forward.
I'd start with the missing fact that could have changed the call. Ten extra CRM fields won't necessarily help as much as finding the one approval nobody checked.
Find out why the deals moved
Before changing your forecasting method, work out why the last forecast changed.
Put the original deal list beside what happened. Separate deals that moved beyond the quarter from deals you lost. Add the ones you hadn't included and note any changes in value. Keep the explanation next to each deal so it doesn't disappear inside a new total.
Salesforce's Pipeline Inspection documentation shows how to track deals moved out of a quarter and look into them.55 Salesforce Trailhead, “Understand Pipeline Health with Metrics and Charts”. Documentation of moved-out opportunity tracking, illustrated with fictional training characters.https://trailhead.salesforce.com/content/learn/modules/sell-smarter-with-pipeline-inspection/understand-pipeline-health-with-metrics-and-charts You can start doing the same comparison with a dated export and a spreadsheet.
The reason matters. An approval you assumed existed is a different problem from a buyer unexpectedly losing their budget. A seller holding a deal for next month's compensation is different again.
One seller asks: "Curious what everyone's thoughts are on pushing out sure thing deals by a few days/weeks to benefit next month/quarter?"66 Public practitioner question, “Sandbagging deals?”. The question illustrates a possible timing incentive; it is not a prevalence finding or proof that its author delayed a deal.https://www.reddit.com/r/sales/comments/1jz9a8l/sandbagging_deals/ That question doesn't tell us how common the behaviour is. It does give us another possibility to check: someone may have a reason to prefer a later date.
If the incentive is the problem, a more detailed update won't change it. If nobody understands the buying process, a tougher forecast meeting won't fill in the missing information.
Look at what you knew then before deciding why a deal moved. The fact that it eventually slipped doesn't prove the original date was unreasonable.
Then ask what you did because you expected it to close. A £60,000 slip and a £60,000 loss leave the same hole in this quarter's bookings. But they mean different things for next quarter's workload and the sales work still needed.
A useful pipeline review should give someone a question they can go and answer. Why are our forecasts always wrong? is hard to act on. Which dates depended on an approval we'd never checked? gives you somewhere to start.
Change one thing, then see what happens
One quarter can show you where to look. It can't prove that the whole forecasting system is broken.
Our £200,000 example shows how different deal movements can cancel each other out. It doesn't tell us how often that happens, whether the team expected it or whether another method would have done better. You need more than one result to answer those questions.
Researchers study forecasts over time too. The January 2025 Stanford abstract of Rationalizing Firm Forecasts describes ten survey waves over five years, covering more than 6,000 firms.77 Nicholas Bloom, Mihai A. Codreanu and Robert A. Fletcher, Rationalizing Firm Forecasts, January 2025 working-paper abstract, Stanford Institute for Economic Policy Research. Study-design figures are quoted from this version; no claim about persistent intervention effects is made.https://siepr.stanford.edu/publications/working-paper/rationalizing-firm-forecasts That describes the study's scope; it isn't a benchmark for your sales team.
Pick one explanation you can check. Say you find that deals keep slipping when nobody knows the approval process. On the next forecast, make that uncertainty visible. Try requiring the team to establish the approval steps before relying on the proposed date.
Save that forecast. When the results come in, compare it with forecasts made equally far ahead, using the same measure. Check both the total error and what happened to the deals you committed. Note whether you put fewer deals into commit, too.
Otherwise, you can make commit look better simply by becoming too cautious. If surprise wins still supply much of the total, closing a larger share of a smaller list may not mean you understand the business any better.
A better result next quarter would be encouraging. It wouldn't prove that your new rule caused it. The mix of deals, their timing and the market can change as well.
The aim is to have a forecast you can explain. What did you expect? Why did that seem reasonable? What changed? Those answers give you more reason to trust the next number than the fact that the last one happened to land.
Explaining every forecast this way, quarter after quarter, is the groundwork for predictable revenue.