“A forecast is a promise, not a guess.” — Quote Card

What a forecast-as-promise actually means

The distinction sounds like a slogan on a quote card, and in fairness that is exactly where most people first meet it. But the sentence survives outside the graphic because it names a real structural difference between two things that look identical on a spreadsheet. Both are a number in a cell. Both get read aloud on a Monday call. The difference is what happens upstream of the number and what happens downstream of a miss.

A guess is a number produced by intuition, inertia, or social pressure. It is often last quarter's actual plus a growth rate someone in finance likes. It is sometimes the number that makes the meeting end fastest. Nobody documented how it was derived, because nobody expected to be asked. When it misses, the explanation is atmospheric — the market softened, procurement slowed down, a champion left. These explanations are unfalsifiable, which is precisely why they are so popular.

A promise is a number with a chain of custody. Someone can point to the specific opportunities that compose it, the stage each one sits in, the historical conversion rate applied to that stage, and the two or three assumptions that would have to hold for the number to land. When a promise misses, the post-mortem is tractable, because you can go back and ask which link in the chain broke. Did a deal that was 80% likely fail? That is variance, and variance is expected. Did five deals you called "verbal commit" turn out to have no budget? That is a calibration failure, and calibration failures are fixable.

The practical consequence is that promise-based forecasting is not primarily about being right more often. It is about being *wrong in a way you can learn from*. A team that guesses and hits its number learned nothing. A team that promises and misses by 8% with a documented cause learned something durable. Over four quarters, the second team compounds and the first does not.

There is an adjacent version of this in operations that makes the point cleanly. When a manufacturing plant commits to a delivery date, nobody calls it a guess — it is a promise backed by capacity math, material lead times, and a named owner. Miss it and there is a root-cause review, not a shrug. Revenue forecasting is the same discipline applied to a fuzzier input, and the fuzziness is not an excuse to lower the standard. It is the reason the standard has to be explicit.

The distinction also changes who owns the number. In guess mode, the forecast belongs to whoever last touched the spreadsheet — usually the RevOps analyst who rolled it up. In promise mode, the number belongs to the rep who committed the deal, the manager who accepted the commit, and the leader who repeated it to the board. Ownership travels with the commitment. That single change explains most of the behavioral difference between the two systems.

One more nuance worth stating plainly: a promise is not a stretch goal. Organizations routinely confuse the two and then wonder why nobody believes the forecast. A stretch goal is aspirational by design and is supposed to be missed sometimes. A promise is supposed to be hit. If you publish one number and call it both, you have taught your team that promises are optional, which is a fast route back to guessing.

The step-by-step process for building a forecast you can stand behind

The mechanics matter more than the mindset, because mindset without mechanics decays within two quarters. What follows is the sequence that holds up in practice, roughly in the order a weekly cycle should run.

Start with hygiene, not with math. Before any number is computed, the underlying records have to be believable. Close dates in the past, opportunities with no activity in 30 days, deals sitting in a stage they entered four months ago — these are not forecast inputs, they are noise. A common working rule is that any opportunity whose close date has slipped three or more times gets pulled out of the committed number entirely until the rep can produce a new, defensible date with a reason attached. This step alone typically removes 10-20% of a neglected pipeline and immediately improves accuracy, not because the analysis got better but because the inputs stopped lying.

Weight by stage, using your own history. Generic probability tables shipped with a CRM are a starting point and nothing more. Pull your actual win rates by stage over the last four to six quarters, segmented by deal size if your ranges are wide, because a $15K deal and a $400K deal do not behave alike. Small deals close faster and more predictably; large ones have more stakeholders and more ways to die. A single blended rate hides both. If you have enough volume, segment by lead source too — inbound and outbound cohorts frequently differ by 15 points or more at the same stage.

Separate the categories and mean them. Commit, Best Case, and Pipeline are useful only if each has a written definition that a new rep could apply without asking. Commit should mean something close to "I will personally be surprised if this does not close" — practically, a confidence level in the 85-90% range, backed by a verbal from an authorized buyer and a mutual action plan with dates. Best Case means real but not certain. Pipeline is everything else. When Commit routinely converts below 80%, the definition has drifted and needs re-teaching, not a new spreadsheet.

Force assumptions into writing. Every committed deal above a materiality threshold — pick one, often around 5% of the quarter — should carry two or three explicit assumptions in plain language. "Legal review completes by the 12th." "The economic buyer signs before their fiscal year close." "No security review is triggered." These are the things that will break, and writing them down converts a vague miss into a specific one.

Run the inspection call as an audit, not a status update. The manager's job on the pipeline review is to look for the gap between what the record says and what the rep believes. Useful questions: who else has to say yes, what does the signature process actually look like, when did you last speak to the economic buyer, and what would have to be true for this to slip. If a rep cannot answer these for a committed deal, it is not a commit.

Roll up and reconcile top-down. Bottom-up gives you rigor; a top-down sanity check catches systematic bias. If bottom-up says $4.2M and your trailing four-quarter average is $3.1M with no change in headcount or pipeline coverage, something is inflated. Coverage ratio is the crudest useful check — most B2B teams need 3-4x pipeline against quota, and if you are carrying 2x while forecasting a beat, the math is not there.

Publish, then track the delta. The number goes out with a version and a date. Next quarter, you compare what you said in week two, week six, and week eleven against what actually happened. The shape of that convergence curve tells you more about forecasting health than any single accuracy percentage.

Costs, timelines, and what a realistic accuracy range looks like

The honest framing is that this costs time every single week and pays back over quarters, not days. Teams that expect a fast win usually abandon it in month two.

Time cost. The recurring load lands in three places. Reps spend roughly 30-60 minutes a week on record hygiene and commit justification — more in the first month while they are learning the definitions, less once it becomes habit. Front-line managers spend the most: a genuine inspection call for a book of 8-10 reps runs 60-90 minutes weekly, and it cannot be delegated to a dashboard. RevOps carries the analytical load, which is heavy up front — pulling historical conversion rates, segmenting them, and building the reporting typically consumes a few weeks of a analyst's time — and then settles into a lighter weekly roll-up and reconciliation rhythm.

Timeline to signal. Expect the first quarter to look worse before it looks better. Cleaning pipeline pulls deals out, so the forecast drops, and leadership frequently misreads this as a performance problem rather than an accuracy correction. It is worth warning them in advance, in writing. Meaningful improvement usually shows in the second full quarter, once you have one clean cycle to calibrate against. Stable, trustworthy forecasting is generally a three-to-four quarter project because you need multiple cohorts of closed deals before your own conversion rates mean anything.

What accuracy is realistic. This depends enormously on deal size, cycle length, and volume, so treat any single benchmark with suspicion. Broadly: a high-volume transactional business with short cycles and hundreds of deals per quarter can get within a few points of actual, because the law of large numbers is doing most of the work. A team running large enterprise deals with 9-12 month cycles and a handful of closes per quarter will be lumpier by nature — one deal slipping moves the whole number — and should be measured on category conversion discipline rather than a single accuracy figure. Mid-market teams typically land somewhere between. The useful target is not a specific percentage but a *trend*: is your variance shrinking quarter over quarter, and is it symmetrical rather than always in the same direction?

Directional bias is the metric nobody tracks. If you miss high three quarters running, you do not have a forecasting problem, you have an optimism problem, and it is systematic. A team that alternates between +6% and -5% is calibrated. A team that goes -12%, -9%, -14% is guessing with extra steps. Plot signed error, not absolute error, and the pattern becomes obvious within a year.

Tooling cost. You can run this entirely inside a well-configured CRM plus a spreadsheet, and plenty of good teams do. Dedicated revenue-intelligence platforms add automated activity capture, deal-risk scoring, and forecast-versus-actual snapshotting, which removes real manual work — but they are a meaningful per-seat line item and they do not fix definitions. Buying a tool before you have written down what "Commit" means produces a very expensive guess. Sequence it the other way.

The cost of not doing it. Harder to quantify but more consequential. Over-forecasting drives hiring plans, inventory, and marketing spend against revenue that does not arrive; the correction is layoffs or a spending freeze, both of which cost far more than the forecasting rigor would have. Under-forecasting is quieter and still expensive — capacity you did not add, territory you did not cover, a quarter where you left demand on the table. Both failures trace back to the same root: a number nobody was actually accountable for.

Where teams get this wrong

They confuse a promise with a stretch goal. Covered above, but it is the single most common failure and worth repeating with its consequence. When the committed forecast and the aspirational target are the same number, reps learn that commits are performative. Within two quarters, Commit means "the number my manager wants to hear," and the category is dead. Keep them separate and named differently.

They make accountability punitive. If missing a commit reliably produces a bad meeting, reps will sandbag — hold deals out of commit so they can bring them in as heroic overperformance. Sandbagging is just as destructive to accuracy as inflation, and it is harder to detect because the misses go in the flattering direction. The fix is to make the review about the *assumption that broke*, not the person. A rep who committed a deal, documented that it depended on legal clearing by the 12th, and watched legal take until the 28th did their job correctly. Treat it that way, out loud, in front of the team, and you buy honest forecasting for a year.

They inspect the number instead of the deal. A pipeline review that consists of reading percentages off a dashboard is theater. The signal lives in deal specifics — has the economic buyer been in a room with us, is there a written mutual plan, do we know how their signature process works. Managers who ask those questions get accurate forecasts. Managers who ask "so are we still good for the quarter?" get whatever answer ends the call.

They apply one probability model to everything. Enterprise and SMB motions, new logo and expansion, inbound and outbound — these have genuinely different conversion behavior, and blending them produces a number that is wrong in both directions simultaneously. Expansion revenue in particular is often far more predictable than new business and deserves its own treatment; lumping it into a single weighted pipeline throws away your most reliable input.

They forecast bookings and ignore everything downstream. A signed deal is not cash and it is not retained revenue. Teams that promise on bookings alone get blindsided when implementation slips push revenue recognition a quarter, or when churn quietly offsets new business. If the board cares about net revenue, the forecast has to include the renewal and churn side, which means CS has to be in the forecasting process, not adjacent to it. This is where forecasting stops being a sales exercise and becomes a genuinely cross-functional one.

They rebuild the model every time it misses. A miss triggers a redesign, the redesign takes six weeks, and by the time it ships there is no historical continuity to calibrate against. Resist this. Most forecasting failures are definition and hygiene failures wearing a math costume. Fix the inputs and the definitions two or three times before you touch the model.

They let the forecast live in a document nobody opens. If the committed number is not visible to the people whose behavior it is supposed to change, it is not a promise, it is a filing. Visibility does not mean a leaderboard — it means the number, its composition, and its current risk state are somewhere a rep or manager will actually encounter mid-week.

Decision framework: when rigor pays and when it is overhead

Not every forecast needs the full apparatus. Applying enterprise-grade forecasting discipline to a five-person startup selling $8K annual contracts is a way to feel productive while shipping nothing. The right level of rigor scales with what the number is being used for.

The governing question is: what decision does this forecast drive, and what does being wrong cost? If the answer is "it informs a hiring plan and a board conversation," the number needs to be a promise. If the answer is "it helps us decide which two deals to focus on this week," a directional estimate is fine and the ceremony is waste.

A workable tiering:

Low stakes, high volume, short cycles. Trust the statistics. A weighted pipeline plus a run-rate check is sufficient. The aggregate is stable enough that individual deal inspection adds little. Spend the saved manager time on coaching instead.

Medium stakes, mixed motion. Categorize deals properly and inspect the committed tier only. You get most of the accuracy benefit for a fraction of the time cost, because the committed tier is where the decisions get made.

High stakes, low volume, long cycles. Full treatment. Every material deal gets written assumptions, a named executive sponsor on both sides, and a documented close plan. Here a single deal can move the quarter, so deal-level rigor *is* forecast rigor. Scenario ranges — a downside, expected, and upside band — are more honest than a point estimate, because a point estimate implies a precision the sample size cannot support.

Brand-new business, no history. You cannot weight by conversion rates you do not have. Forecast on leading activity instead — meetings booked, proposals sent, pilots started — and be explicit that the number is a hypothesis with a review date attached. This is the one case where saying "this is a guess" is the honest and correct move, and saying so is far better than dressing a guess in promise language. Convert to promise-based forecasting once you have two or three cohorts of closed deals to calibrate against.

The same framework applies outside revenue, which is worth noting because it is where teams often see the idea click. A product launch date, a hiring plan, a capacity forecast in a services firm, a delivery commitment in logistics — each is a forecast that someone downstream will spend money against. The test is identical: is anyone allocating resources against this number? If yes, it is a promise and should be built like one. If it is genuinely just an input to a conversation, let it be a guess and say so.

Related questions

What is a good forecast accuracy target?

Less useful than a trend. High-volume transactional teams can hold tight variance; long-cycle enterprise teams are structurally lumpier. Track signed error over four quarters — shrinking and symmetrical is healthy, consistently one-directional means systematic bias, not bad luck.

How is Commit different from Best Case?

Commit means you would be genuinely surprised if it did not close — a verbal from an authorized buyer plus a dated mutual plan. Best Case is real but contingent. If Commit converts below 80% consistently, the definition has drifted and needs re-teaching.

Should the forecast be a single number or a range?

Both, for different audiences. A range with downside, expected, and upside is more honest for long-cycle or low-volume businesses. A single committed number is what leadership plans against. Publish the range internally and the commit externally, and never let the two silently merge.

Who actually owns the forecast?

The rep owns each committed deal, the manager owns accepting or rejecting those commits, and the leader owns the rolled-up number they repeat publicly. RevOps owns the process and the data, not the promise. When RevOps owns the number, nobody with deal knowledge is accountable for it.

Does this apply to non-revenue forecasts?

Yes. Hiring plans, launch dates, and capacity forecasts all have someone downstream spending against them. The test is the same: if resources are allocated based on the number, it is a promise and needs written assumptions and a named owner.

FAQ

What does "a forecast is a promise, not a guess" actually mean?

It means the number carries an implicit commitment to have done the work behind it — inspected the pipeline, applied real conversion history, and written down the assumptions that have to hold. It is not a promise that the future will cooperate. It is a promise that the analysis was rigorous and the reasoning is inspectable when it misses.

Isn't every forecast fundamentally a guess about an uncertain future?

Uncertainty is real, and treating a forecast as a promise does not pretend otherwise. The commitment is about process, not omniscience. A forecast built from documented assumptions can still miss when an assumption breaks — that is expected and fine. The difference is that you know exactly which one broke and can correct for it next cycle.

How do we stop this from turning into blame culture?

Make the review about the assumption, not the person. If a rep committed a deal, documented that it depended on legal clearing by a certain date, and legal ran long, they forecast correctly and should be told so publicly. Punish inflated commits and unwritten assumptions; never punish a well-reasoned miss.

Can a startup with no historical data do promise-based forecasting?

Partially. Without conversion history you cannot weight a pipeline meaningfully, so forecast on leading indicators — meetings, proposals, pilots — and label the number a hypothesis with a review date. Being explicit that it is a guess is far better than dressing a guess in promise language. Convert once two or three cohorts have closed.

What is the fastest single improvement to forecast accuracy?

Pipeline hygiene. Removing stale opportunities, dead deals, and anything that has slipped its close date three or more times typically strips 10-20% out of a neglected pipeline and improves accuracy immediately — not because the analysis improved, but because the inputs stopped lying. Do this before touching any model.

Where does the Quote card framing come from?

It circulates widely in sales and finance leadership as a compressed reminder rather than a citable original. The value is in the compression: it makes the accountability distinction memorable in a way a process document cannot, which is exactly what a shareable quote graphic is for.

Sources

• https://hbr.org/2010/07/why-good-projects-fail-anyway — Harvard Business Review on planning assumptions and project failure
• https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights — McKinsey research on sales growth, forecasting, and revenue operations
• https://www.pmi.org/learning/library — Project Management Institute library on estimation, risk, and forecasting practice
• https://www.bls.gov/emp/ — U.S. Bureau of Labor Statistics employment projections methodology
• https://www.federalreserve.gov/monetarypolicy/fomccalendars.htm — Federal Reserve Summary of Economic Projections, an example of published forecast ranges
• https://www.noaa.gov/ — NOAA on probabilistic forecasting and communicating uncertainty
• https://www.sec.gov/education/smallbusiness — SEC guidance relevant to forward-looking statements and disclosure
• https://sloanreview.mit.edu/ — MIT Sloan Management Review on managerial decision-making and forecasting bias

Related on PULSE

• [Forecast Bands Beat Point Estimates — Stat Card](/knowledge/gb0469)
• [Forecast Accuracy Trend](/knowledge/gb0521)
• [Forecast Commit / Best Case / Upside](/knowledge/gb0508)
• [Forecast First, Pipeline Second — Banner](/knowledge/gb0461)
• [4 Forecast Categories — Infographic](/knowledge/gb0109)
• [Forecast — Section Divider Slide](/knowledge/gb0083)