Planning & Forecasting
Why do we keep missing our own forecast?
Simplify24 September 20269 min read
In short
A forecast that is wrong at random is a hard problem. A forecast that is wrong in the same direction every quarter is an easy one, because a consistent bias can simply be measured and corrected. Track forecast against actual by driver for two quarters, find where the bias sits, and adjust for it. Most of the causes are about who produces the number and what happens to them if it is wrong, not about the technique.
Eight quarters, and seven of them came in below forecast. The method is not obviously wrong, the people are competent, and each individual miss had a specific and believable explanation: a deal slipped, a customer delayed, a launch moved.
That pattern is the tell. Genuine uncertainty produces misses in both directions. A forecast that is wrong the same way for seven quarters is not suffering from uncertainty, it is carrying a bias, and a bias is a far easier problem than uncertainty because it can be measured and subtracted.
The reasons are almost never technical. They are about who produces the number, what happens to them if it is wrong, and what nobody wants to say in the meeting.
The forecast is a negotiation, not an estimate
In most companies the revenue forecast is produced by the people responsible for delivering it, and is used to judge them. That single arrangement creates the bias.
It pulls in two directions at once, and different people resolve it differently. A salesperson whose quota derives from the forecast has a reason to keep it low. A sales leader presenting to a founder has a reason to keep it high. A founder presenting to a board has a reason to keep it high. Nobody in that chain is lying, and every one of them is answering a slightly different question from the one asked.
The fix is structural rather than motivational. Separate the commitment from the expectation, and be explicit about which is which.
- The commitment is what the sales team signs up to deliver. It should be ambitious and it belongs in compensation and target-setting.
- The expectation is what finance plans the cost base against. It should be the honest statistical view, weighted by what actually happened historically, and it should be lower.
- The two being different numbers is correct and it needs saying out loud, because a team that believes the two are the same will produce one number that does neither job.
If the same number is used to motivate the team and to plan the spending, it will be wrong for one of those purposes. Usually both.
Six causes, in the order they usually apply
1. Confidence weighting
Deals are weighted by how likely somebody feels they are. Those percentages are optimistic in a consistent direction and nobody ever goes back to check them against outcomes.
The correction is to weight by historical stage conversion instead: if deals at Proposal sent have closed 28% of the time across two years, a deal at that stage is worth 28% of its value regardless of how it feels. It will be wrong about every individual deal and roughly right across twenty of them.
2. Timing, which slips one way
Deals slip later far more often than they pull forward. Almost nothing closes early. So even a forecast with perfect conversion rates lands revenue in the wrong month.
Measure how much later your closed deals actually landed than first forecast, and add that to every expected close date. If the average was three weeks, add three weeks. It feels pessimistic and it is simply what happened.
3. Stage definitions that mean nothing
Weighting by stage only works if a stage means the same thing to everyone. Where a stage is defined by what your team did rather than what the customer did, deals sit two stages further along than they belong, and the weights become decorative.
Proposal sent is an action you took. Customer has acknowledged the proposal and asked for a revision is an observable customer behaviour. Only the second is checkable by someone who was not on the call.
4. Nobody moves a deal backwards
Deals advance through the pipeline and almost never retreat, even when the buyer has gone quiet for six weeks. So the pipeline fills with deals at late stages that are not actually late-stage, and the weighted total is inflated by construction.
The fix is a hygiene rule rather than an analytical one: any deal with no customer contact in thirty days moves back a stage or comes out. Applied automatically, without a discussion each time.
5. The plan was never a forecast
Frequently the number being missed is the annual plan, built in March from a growth rate that the board found acceptable, and it was never an estimate of anything.
Missing it is then not a forecasting failure at all. It is the gap between an ambition and a reality, and treating it as a forecasting problem sends everyone looking in the wrong place. The budget and the forecast should be two different numbers reported side by side, precisely so this distinction stays visible.
6. The cost side is never the miss
Worth noticing: almost every conversation about missed forecasts is about revenue. Costs miss too, and they miss in the opposite direction, which is to say they come in higher.
Hiring that happens later than planned makes costs look good and is usually bad news. One-off costs that were not in the plan. Suppliers who raised prices. A cost forecast that is consistently under is the same bias problem wearing different clothes, and it partly offsets the revenue miss, which is how a company can miss revenue by 12% and profit by only 6% and conclude that things are fine.
Measure the bias before fixing anything
This is the whole method, and it takes an hour a month.
- 01For each of the last six to eight periods, record what was forecast and what happened. Total, and by driver: new business, expansion, churn, and the main cost lines.
- 02Calculate the variance each period, signed rather than absolute, so the direction is visible.
- 03Take the average. If it is close to zero and the individual periods swing both ways, you have uncertainty and the answer is to forecast ranges rather than points.
- 04If the average is meaningfully negative, you have a bias, and you now know its size.
- 05Then find which driver carries it. It is usually one: new business timing, or a particular segment, or one salesperson's deals.
A company that discovers its forecasts have averaged 14% high for two years has learned something immediately actionable, which is more than any refinement of method would have produced.
What to change in the meeting
Most of the durable improvement comes from how the forecast is discussed rather than how it is calculated.
- Report forecast against actual every month, with the variance, in the same format. Two quarters of that row changes behaviour more than any analytical change.
- Ask for the reason a deal moved, and record it. After twenty of those you will know whether you have a conversion problem or a timing problem, and they need different responses.
- Make it acceptable to lower a forecast. In companies where reducing a number is treated as a failure, nobody reduces one, and the correction arrives all at once at quarter end.
- Separate the discussion of the commitment from the discussion of the expectation, so nobody has to defend two positions at once.
- And review the stage conversion rates every six months, because they drift as the business changes.
Forecast a range, and say what decides it
Once the bias is corrected there is still genuine uncertainty left, and a single number is a poor way to express it.
A range is better, and it is only useful if it comes with the two or three specific things that decide where in it you land. Revenue between ₹58 and ₹72 lakh, with the difference being the two enterprise deals in procurement, tells a reader considerably more than ₹65 lakh does, and it is honest about what is actually known.
It also changes how a miss gets discussed. A company that forecast a point estimate and came in 12% below spends the meeting explaining a failure. A company that forecast a range and landed at the bottom of it spends the meeting on the two deals that decided it, which is the conversation worth having.
For planning, use the bottom of the range rather than the middle. The cost base should be affordable in the outcome you named as plausible, not only in the one you consider likely.
The objection is that ranges look indecisive to a board. In practice the opposite happens: a range with named dependencies reads as somebody who understands their own business, and a point estimate that is missed repeatedly reads as somebody who does not.
When the forecast is fine and the business is not
One situation worth separating, because the response is entirely different.
If the forecast is missed because deals that were genuinely at a late stage are not closing at all, that is not a forecasting problem. Something has changed in the market, the product or the competition, and the forecast is accurately describing a business that has got harder.
The tell is in the conversion rates rather than the timing: deals are not slipping, they are dying. A company that responds to that by improving its forecasting has tuned the instrument while ignoring what the instrument is telling it.
So the first question when a forecast is missed should always be whether the miss is timing or conversion. Timing is a forecasting problem. Conversion is a business problem, and it is the more important of the two.
Who should own the number
One structural change does more than any analytical one: the forecast used for planning should not be produced by the people whose performance is measured against it.
That does not mean sales is excluded. Sales owns the pipeline, the stage assessment and the commitment, and nobody else can supply those. What finance owns is the conversion from that pipeline into an expectation, using historical rates rather than judgement about individual deals.
The separation matters because it removes the conflict rather than asking anyone to overcome it. A sales leader is not being asked to produce a conservative number and defend an ambitious one in the same meeting, which is an unreasonable thing to ask of anybody.
In a company too small for that split, the founder can do it by being explicit about which hat is on: this is the number we are going after, and this separate number is what we are spending against. Written down as two lines rather than held as one number with a mental adjustment.
What to do this quarter
- 01Build the forecast against actual record for the last six periods. An hour, and it tells you whether you have bias or uncertainty.
- 02If it is bias, size it and apply it as a correction while you work on the cause.
- 03Rebuild your stage conversion rates from the last two years of closed deals, and use those as weights instead of confidence.
- 04Add the average slippage to every expected close date.
- 05Write stage definitions in terms of observable customer behaviour, and apply the thirty-day hygiene rule.
- 06And start reporting forecast against actual every month, permanently.
The free sales forecast template on this site does the weighting and the accuracy tracking, and the piece on budget against forecast covers why the two numbers should be reported separately in the first place. Neither requires anyone to become better at predicting the future, which is fortunate, because nobody is going to.
About Simplify
Simplify is a finance clarity and investment readiness practice working with founders across India, built on six years inside startups. We write about the questions founders bring before a decision, not after it.
Planning happening after the problem rather than before it?