The Forecast Built on Opinions: Commit, Best Guess, and What Makes a Number Reviewable
A forecast is a claim about a month that has not happened. The difference between a commit and a best guess, why a number with no stated basis is worse than no number, and what makes a forecast reviewable after the month closes.
A forecast is a claim about a month that has not happened. It is written in the present tense, in a room where nobody can check it, about an outcome that will not be known for weeks. Every element of that sentence is a weakness, and the weakness is not fixed by confidence.
The reason forecasts go wrong is usually not that the person making them is careless. It is that a number gets produced, circulated, and discussed without anybody ever stating what the number is a claim about. A forecast with no stated basis is a preference written in a font that makes it look like information, and it is more dangerous than no forecast at all, because a number gets acted on.
The claim in one line
A forecast is only reviewable if you can reconstruct, after the month closes, what the person making it thought they were asserting. A commit and a best guess are different kinds of claim, not different amounts of confidence, and a figure that does not say which kind it is cannot be examined when the month turns out differently.
Commit and best guess are not the same size, they are not the same kind of thing
The distinction is usually described as a probability attached to a deal, and that framing hides the only part that matters. What separates a commit from a best guess is not how likely it is. It is whether a named person is accountable for the gap if it does not happen.
A commit is a statement that somebody will be asked about the shortfall personally, and can point at the specific thing that was supposed to make it happen. A best guess is a statement that the deal is real and progressing, that it may or may not land inside the month, and that nobody will be held to a figure. Both are honest positions. They are just not interchangeable, and a total that mixes them without saying so is a number whose error cannot be interpreted.
The constructed forecast below is a worked example. Every deal and figure is invented, and the arithmetic is shown underneath so it can be checked by hand.
One month's forecast and what actually invoiced (worked example, illustrative — the deals are invented; arithmetic shown)
| Deal | Forecast value | Basis stated at the time | Invoiced in the month | Gap on that deal |
|---|---|---|---|---|
| A | 100,000 | Commit | 100,000 | 0 |
| B | 75,000 | Best guess | 75,000 | 0 |
| C | 60,000 | Best guess | 0 | 60,000 |
| D | 40,000 | Best guess | 0 | 40,000 |
| Total | 275,000 | 100,000 commit, 175,000 best guess | 175,000 | 100,000 |
Checking the arithmetic.
- Forecast total: 100,000 + 75,000 + 60,000 + 40,000 = 275,000.
- Committed portion: 100,000 (deal A alone).
- Best-guess portion: 75,000 + 60,000 + 40,000 = 175,000.
- Check that the two portions add back to the total: 100,000 + 175,000 = 275,000. Matches.
- Actually invoiced: 100,000 + 75,000 = 175,000.
- Total gap: 275,000 - 175,000 = 100,000.
- Gap on the committed portion: 100,000 - 100,000 = 0.
- Gap on the best-guess portion: 175,000 - 75,000 = 100,000.
- Check: 0 + 100,000 = 100,000, which matches the total gap. Matches.
The committed portion was right. The best-guess portion missed by 100,000, which is not a failure — two of four genuine deals simply did not land inside that month. The point of the example is not that the forecast was wrong. It is that the same 100,000 gap means two completely different things depending on what was claimed.
The same gap, read two ways
Take the identical month and relabel the two best-guess deals as commits. The forecast is now 275,000 with no distinction inside it, and the actual invoices are still 175,000, and the gap is still 100,000. The arithmetic has not changed at all. What has changed is the meaning available to the reader.
Under the first labelling, the business learns that its best guesses are unreliable, which is information: it can look at why C and D did not close and learn something about its cycle length. Under the second labelling, the business learns that somebody committed to 275,000 and delivered 175,000, which is a broken promise, and the obvious response is to stop trusting forecasts entirely.
Now relabel everything as a best guess. The gap is still 100,000, and now nobody can say the team missed, because a best guess was never a claim about the month. The number floats free of accountability, and the review becomes a conversation about whether people are being too negative — which is not a conversation that improves a forecast.
This is why the label, not the arithmetic, is the load-bearing part of a forecast. The arithmetic is the same in all three versions.
Why a number with no stated basis is worse than no number
An empty forecast has an obvious defect, and everyone who sees it knows the plan is unsupported. That visibility is protective: it prompts questions, it keeps the room honest, and it puts the burden on whoever wants the number to produce it.
An unsupported number does the opposite. It fills the gap in a plan, it looks like the answer to a question nobody asked to have answered carefully, and it moves the conversation to arguing about whether the number is high or low — which is impossible to settle when nobody can say what the number is measuring. The conversation that would have improved things never happens, because the number made the meeting feel complete.
The specific danger is that unsupported numbers compound. They are carried into next month as a base, adjusted, and carried again. Three months later the business is forecasting against a figure whose only justification is a figure someone typed in a cell in January. Nobody can audit it, because there is nothing to audit it against. And because it feels like a number, it is treated as a fact in every argument that follows — including the argument about whether to try harder or hire somebody.
What makes a forecast reviewable after the month closes
A forecast becomes reviewable when it carries enough information that somebody who was not in the room can reconstruct the claim and check it. That takes four things, and none of them is a number.
The four things a forecast has to carry (illustrative)
A stated basis for each figure
Not a sentiment, not a gut feel described as a judgement. The thing that is supposed to make the deal happen, written so that a reader can go and check whether it happened. A dated next action with a named owner, a verbal agreement with a date attached, a signed document. If the basis cannot be checked, it is an opinion wearing a label, and the label should say so.
A named author and a date
One person who stands behind the figure, and the date it was written. A forecast circulated without an author has nobody to ask, which means it becomes a shared belief within about two conversations — at which point it stops being anybody's claim and becomes the room's mood.
A stated horizon
Which month, and by what measure. Invoiced by the last day of the month, cash in the bank by the fifth, signed by the tenth. Three reasonable businesses produce three different answers to the same question, and without a stated horizon two of them are comparing incompatible numbers and calling it a miss.
A basis for change
What was revised during the month, and why. A forecast that is only ever seen once at the start of the month cannot be held to anything, because nobody can see whether the number moved when the underlying evidence moved. If the figure is going to be updated, the update needs the same four things as the original.
One more thing to state explicitly.
If the measure is cash collected rather than work invoiced, say that — and remember that an invoice being settled is itself a projection of allocations, not a stored flag. In NoxOrigin an invoice and a payment are two different records. Nothing on the invoice says paid. What says an invoice is settled is the set of allocations pointing at it, and an unallocated credit sitting against a customer is not a settlement, it is an unanswered question about which invoice it was for. A forecast built on collections therefore has a second layer of estimation on top of the first, and that layer deserves to be named rather than assumed away.
The worked example in article two in this set, on the clean pipeline, covers the ageing and the counted records behind a figure. The post at /blog/invoice-and-payment-are-not-the-same-record covers the allocation layer in detail if you need it.
The arithmetic trap: a forecast and an invoice are not the same figure
A small but reliable way for a forecast to look wrong is to compare a deal's commercial value against the invoice that later gets raised for it, without accounting for tax. The 18% GST figure in the example below is used only to keep the arithmetic checkable by hand. It is not tax advice, it is not a statement about what rate applies to any real transaction, and the correct rate and treatment for your business is a question for your own chartered accountant.
One deal, forecast against the invoice raised (worked example, illustrative — the deal is invented; the 18% figure is used only so the arithmetic is checkable; arithmetic shown)
| Step | Amount | What it is |
|---|---|---|
| Forecast commercial value | 100,000 | The figure the sales forecast carried for this deal. |
| Invoice raised, total | 118,000 | The tax invoice for the same commercial value. |
| Tax component | 18,000 | 118,000 - 100,000. |
| Taxable value on the invoice | 100,000 | 118,000 / 1.18. |
Checking the arithmetic.
- Taxable value: 118,000 / 1.18 = 100,000.
- Check forward: 1.18 x 100,000 = 118,000, which matches the invoice total.
- Tax component: 118,000 - 100,000 = 18,000.
- Check the tax: 100,000 x 0.18 = 18,000. Matches.
The forecast said 100,000 and the invoice says 118,000. Compared directly, that looks like an 18,000 overperformance, or worse, a forecasting error in the seller's favour. Neither is true. The two figures measure different things: one is the commercial value of a deal, the other is a tax document. A forecast review that does not name its measure will eventually trip over this, and the resulting argument is about tax with people who do not know tax, which is a bad use of a sales meeting.
What to check before you circulate a forecast (illustrative)
- Every deal is labelled as a commit or a best guess, and no total is circulated without both figures
- Each committed deal has a basis that somebody could go and check, written on the record
- The horizon is stated: which month, and measured by invoice date, cash date, or something else
- One named person authored it, and the date is on it
- The measure is unambiguous about tax, so a forecast value is not compared against a tax-inclusive invoice
- If the measure is collections, the allocations behind each settled invoice are identifiable
- Revisions during the month are recorded with a reason, so the forecast can be read as a series of claims rather than one frozen guess
Where this stops, honestly
There is no CRM-native forecasting model in NoxCRM. Nothing in the system weights deals by stage, assigns a probability, or produces a predicted month-end outcome, and no pipeline screen will show you a number that looks like a forecast without somebody having typed it. We consider that the correct position for this article's subject, but it should be said plainly rather than left as a pleasant surprise: if you are evaluating a tool because you want a number to appear,
NoxCRM does keep the records a forecast review needs — opportunities with owners, stages, dated activities, stage history, and the quote and invoice chain that follows a won deal — so a forecast you build by hand has something real behind it rather than a spreadsheet of intentions. What it does not do is judge your commitments, and it does not file any statutory return of any kind. Anything about the tax treatment of a real invoice belongs with your chartered accountant, not with a sales forecast and certainly not with us.
The review rhythm and the labelling rules are part of setup in the dedicated NoxCRM deployment, which is scoped per business rather than sold as a separate product. If you want them built into a working routine rather than run manually from a sheet, that is the conversation to have. The arithmetic in this article needs nothing but a sheet: /templates/quotation-template and the sales pipeline planner at /tools/sales-pipeline-planner both work before any system is in place, and a forecast format you can run by hand is a forecast format you will actually keep using.
What we have not measured
We have not measured forecast accuracy for any business, we do not know what a good hit rate looks like, and we are not going to give you a target to chase. A target for a forecast is a manufactured constant, and chasing it produces exactly the behaviour this article is about: numbers chosen to look right rather than to describe something.
What you can measure without inventing anything is your own labelling discipline. Take last month's forecast. If you cannot identify, for each deal, whether it was a commit or a guess, then last month's forecast cannot be reviewed — and no amount of accuracy would have helped, because there would be nothing to attribute the result to.
Frequently asked questions
What hit rate should we be aiming for?
We do not have one, and neither should you take one from a vendor. A hit rate is a measurement of how often a particular person's forecasts came true, and it is meaningless without knowing what they were claiming in the first place and how long their sales cycle runs. A monthly target for a forecast is worse than useless, because the fastest way to hit one is to make the forecast safe. The measure worth watching is narrower: can you look back at a month and say what each figure was claiming.
Is it bad to commit to deals you are not certain about?
Committing to something you do not believe is the thing that makes a forecast worthless, and it is usually done for a reason that is easy to name: a commit feels like a promise about the person making it rather than a claim about the market. A best guess that is labelled as a best guess is a perfectly respectable thing to put in a forecast. Most of the trouble people have with forecasting comes from wanting their numbers to be commitments, because commitments feel more like leadership.
Should the forecast be about invoices or about cash?
Both are legitimate, and you must pick one for a review and state it, because they answer different questions. Invoicing measures work done. Cash measures whether anybody paid. The gap between them is a collections problem, and a forecast that quietly switches between the two will sometimes look accurate and sometimes look disastrous while describing the same business. NoxOrigin does not store paid as a field on an invoice: an invoice and a payment are separate records, and settled status comes from the allocations pointing at the invoice.
Does NoxCRM predict the month for me?
No. There is no CRM-native forecasting model, no stage weighting, and no probability assigned to a deal. NoxCRM keeps the underlying records — opportunities, owners, stages, dated activities, and history — and the judgement stays with the person who owns the number, where this article argues it belongs. If you want a tool to output a predicted month-end figure without anyone having reasoned about it, NoxCRM will not do that.
How often should the forecast be reviewed?
Often enough that nobody is defending a number they formed weeks ago, and not so often that reviewing it becomes the meeting. The structure matters more than the frequency: each revision should carry the same four things as the original — basis, author, date, and horizon — so that the month can be read at the end as a sequence of claims that moved when the evidence moved, rather than as one frozen opinion that happened to be right or wrong.