How do you design a 2027 forecast that distinguishes commit vs best case vs pipeline?

The Thursday your commit number stops meaning anything

Picture a $40M ARR B2B software company entering week eleven of Q3 2027. The CRO walks into the board prep call with a commit of $9.4M against a $10.2M plan. Three weeks later the quarter lands at $8.1M. Nothing dramatic happened — no logo churned, no competitor undercut them. Four deals that had been sitting in commit for six straight weeks simply moved to next quarter, each with a different reason and none of them new information.

That is the failure that a tiered forecast is supposed to prevent, and the reason it did not is almost never the math. It is that "commit" had become a social category rather than an evidentiary one. When you interview the AEs afterward, you find each of them defining the word differently. One meant "I am confident." One meant "my manager already put it in the roll-up and I did not want to be the person who pulled it." One meant "the champion told me it is happening" without ever having asked whether procurement had a queue, a legal reviewer, or a signature authority threshold that the deal size tripped.

The design fix starts by refusing to define tiers by feeling. A tier is a container with an entrance exam. If a deal cannot produce the artifacts the tier demands, it does not sit there, regardless of how strongly the rep believes. That inversion — from "how sure are you?" to "what do you have?" — is the whole architecture. Everything downstream, including the AI overlay, the variance band, and the governance charter, exists to keep that inversion from eroding under quarter-end pressure.

It also helps to name what you are optimizing for. You are not trying to maximize the commit number; you are trying to minimize the *distance between the commit number and the actual*. Those are different objectives and they pull in opposite directions in the last two weeks of a quarter. A forecast design that does not make that tension explicit will lose it every time, because the person who inflates commit gets rewarded in the moment and punished only in the retrospective — and retrospectives are quiet.

One more framing note that matters more than it sounds: the same discipline applies whether you are forecasting new logo bookings, expansion, renewals, or services revenue. Renewal forecasts are usually the sloppiest, because auto-renew clauses create an illusion of certainty; a renewal with an auto-renew clause and a silent, disengaged buyer is a Best Case deal wearing a Commit costume. If your CS org forecasts renewals on a different taxonomy than sales uses for new business, your consolidated number is an apples-and-oranges sum, and the CFO will eventually discover it.

What each tier actually has to prove

Commit means the buying side, not the selling side, has done something irreversible or nearly so. The evidence set is concrete: a mutual close plan with dates the buyer agreed to in writing, redlines resolved or down to non-blocking items, procurement engaged with a requisition or PO path identified, and a named signatory who knows the document is coming. If your security review, vendor onboarding portal, or MSA negotiation is still open, that is not commit — those steps have their own queue times and you do not control them.

Best Case means there is exactly one identified thing standing between the deal and close, and you can name it, name who owns it, and name the date it resolves. "Pricing approval from their VP Finance, owned by our champion, expected the 18th" is Best Case. "Feels good, they love us" is not Best Case; that is Pipeline with enthusiasm. The discipline of forcing a single named unblock event is what separates a usable upside number from a wish list, and it is the tier where most organizations bleed accuracy.

Pipeline means active, two-way motion inside the period with qualification complete — a validated pain, a confirmed budget path, an identified economic buyer, and a next step on the calendar that the buyer accepted. Deals here have real risk: competitor in a bake-off, champion recently changed roles, evaluation started late relative to your median cycle. They belong in the forecast conversation because some of them will land, but they do not belong in any number you speak out loud to a board.

Total Pipeline is everything open. It is a coverage instrument, not a forecast instrument, and conflating the two is how coverage ratios get quoted as if they were predictions. Use it to answer "do we have enough at-bats for next quarter," never "what will we close this quarter."

The subtle rule that makes all four tiers hold: evidence decays. A buyer conversation from eleven days ago is not the same asset as one from two days ago. Build a staleness clock into the tier criteria — if a commit deal has had no buyer-side touch in ten business days, it auto-flags for review and the burden of proof shifts back to the rep. Without decay, tiers ratchet upward and never come down, because the artifact that qualified the deal in week two is still technically on file in week twelve.

How the mechanism actually runs, week to week

The operating rhythm is what converts definitions into a forecast. A three-touch weekly loop works for most mid-market and enterprise teams, tightening in the final month of the quarter.

The rep-level call comes first. Each AE walks their commit and best-case deals against the criteria — not the story, the criteria. A useful forcing device is a three-question screen: have you spoken with the economic buyer inside the last seven business days, is the funding path confirmed rather than assumed, and is there a buyer-accepted next step with a date. Fail two of three and the deal drops a tier automatically. The automatic part matters; if the downgrade requires an argument, the argument will be avoided.

The manager roll-up comes second, and this is where the model overlay earns its keep. Whatever forecasting tool you run — Clari, BoostUp, Salesforce's native scoring, or a homegrown model on your own closed-won history — it produces a probability independent of the rep's judgment. You are not looking for the model to be right. You are looking for *disagreement*, because disagreement is a search light. When a rep calls 90% and the model says 45%, one of two things is true: the rep knows something the model cannot see (a relationship, a competitor stumble, an internal mandate), or the rep is pattern-matching to hope. Both outcomes are useful, and both must be written down.

The executive review comes third and should be boring by the time it happens. If the first two touches did their job, the exec review is a read-out of a number plus a band, plus the two or three deals whose movement would change the story.

Reason codes are the connective tissue nobody wants to build and everybody wishes they had. Every tier movement — up or down — requires a code from a controlled list: procurement queue extended, legal capacity, champion departed, budget reallocated, competitor re-entered, security review opened, signatory changed, buyer priority shift. Free text is where accountability goes to die. A closed vocabulary lets you answer, at quarter end, the only question that improves next quarter: *which of these reasons keeps happening, and is it a deal problem or a design problem?* If "procurement queue extended" is your top code three quarters running, your qualification criteria need a procurement-timeline field, and your commit definition needs to require it — that is a process fix, not a coaching fix.

Numbers worth calibrating against

Treat the following as calibration targets you validate against your own history, not as external facts. The most valuable benchmark you will ever have is your own trailing four quarters, segmented by deal size and segment.

Start with tier close rates. A healthy commit tier converts in the low-to-mid 90s. If yours converts at 80%, the definition is too loose — deals are entering on rep confidence rather than buyer evidence. If it converts at 98-100%, the definition is too strict, and you are almost certainly sandbagging: real deals that qualify are being parked in Best Case so the team can beat the number. Both are definitional failures, and both are correctable in a quarter by adjusting the entrance exam rather than by exhorting people to be more accurate. Best Case typically lands somewhere in the 60s to low 70s; Pipeline in the high 20s to low 40s. When your Best Case rate drifts toward 40%, the "one named blocker" rule has stopped being enforced.

Variance band is the second calibration. A mature team publishes a committed number with a plus-or-minus band in the mid-single digits and lives inside it most quarters. Two failure signals bracket it: a band so tight it is obviously performative — you are claiming a precision your close-rate distribution does not support — and a band so wide it is not a forecast, just a range. Derive the band empirically. Take your trailing eight quarters of commit-to-actual variance, look at the spread, and set the band to cover roughly the middle of that distribution. Then narrow it as the quarter progresses: wide in month one, tighter in month two, tightest in the final three weeks when most of the uncertainty has resolved.

Big-deal distortion deserves its own number. Large deals close at materially lower rates than mid-size deals in almost every organization, and they close later than the rep predicts more often than they close earlier. If your average deal is $60K and you have a $600K deal in commit, that deal should not carry the same probability weight as ten $60K deals, because its outcome is a single coin flip with correlated dependencies — one procurement office, one legal team, one budget cycle. Compute close rates by deal-size decile and apply the deal's own cohort rate rather than the blended rate. Teams that do this discover their forecast was quietly being driven by three deals wearing the credibility of thirty.

Segment your calibration further where volume allows: new logo versus expansion, inbound versus outbound sourced, and by sales cycle stage entry date. Expansion deals into engaged accounts often close at rates that would look absurd for new logo, and if you blend them, you understate expansion and overstate new business simultaneously. The two errors partially cancel at the total level, which is precisely what makes them dangerous — the aggregate looks fine while both components are wrong.

Finally, calibrate the *slip*, not just the win. Track, per tier, what fraction of deals close in the committed period versus one period later versus never. A commit tier where 92% close is excellent. A commit tier where 78% close in-period, 16% close one quarter later, and 6% die is a different animal than one where 78% close and 22% die — the first is a timing calibration problem, the second is a qualification problem, and they need opposite interventions.

Trade-offs: how much structure is too much

There is a real cost curve here, and pretending otherwise is how RevOps teams build forecast processes their sales org quietly routes around. Every criterion you add to a tier is a field someone must populate, a conversation someone must have, and a moment of friction at the exact time reps are trying to sell. The question is not "what is the most rigorous design," it is "what is the most rigorous design this org will actually run in week twelve."

Three archetypes cover most of the choice space. The lightweight design — three tiers, criteria expressed as a short checklist, weekly manager review, no model overlay — costs almost nothing and gets you most of the way if your deal count is high and your deal sizes are homogeneous. A high-velocity SMB motion closing 200 deals a quarter genuinely does not need probabilistic overlays; the law of large numbers is doing the work, and a stage-weighted roll-up will be accurate enough. The heavy design — four tiers plus sub-tiers, per-cohort probability weights, model reconciliation with mandatory variance reviews, a signed governance charter, monthly audits — is right when a handful of deals determine the quarter and a miss is a board event.

The middle design is where most teams should land: strict evidence-based tiers, a model overlay used only as a disagreement detector rather than as a number, reason codes, and a variance band. It captures the majority of the accuracy gain at a fraction of the operational load.

The alternatives worth knowing about, since they show up in vendor conversations and board questions: pure stage-weighted forecasting, which multiplies every open deal by its stage's historical close rate and sums — cheap, unbiased, and useless for telling you *which* deals to work; pure model forecasting, which is often the most accurate single number at the total level but produces no coaching signal and no accountability, since nobody on the team owns a model output; and scenario forecasting, where you publish three full scenarios rather than a number and a band, which finance organizations sometimes prefer because it maps to how they build plans. These are not mutually exclusive with tiers. The strongest setups run stage-weighted or model numbers *alongside* the tier roll-up specifically so the gap between them becomes a discussion item. When your tier roll-up says $9.4M and your stage-weighted number says $7.9M every single quarter, the constant delta is a measurement of your team's optimism bias, and you can subtract it.

A related trade-off that surfaces downstream: how tightly you couple the forecast to compensation and capacity planning. If commit accuracy feeds a manager's bonus, you will get accurate commits and shrinking commit tiers — people optimize for what is measured. If it feeds nothing, you get inflation. The workable middle is to measure and publish accuracy without paying on it directly, while making it a standing item in manager performance reviews. And on the capacity side, remember that Marketing and Finance consume this data too: demand-gen targets are usually derived from a pipeline-coverage ratio, so if your Pipeline tier is inflated, you will under-invest in top-of-funnel while believing you are covered. The tier hygiene problem is not contained to sales.

Where these designs break, and the specific counter

The first breakage is definitional drift. Tier criteria written in January are interpreted loosely by June and abandoned by October. Counter: put the criteria in the CRM as required fields with validation, not in a wiki page. If commit requires a mutual close plan date and a procurement contact, make the tier field un-settable to Commit without both. Governance that lives in a document loses to governance that lives in a form.

The second is the exec-review theater problem, where the weekly meeting becomes a recitation of the same deals with the same narratives. Counter: change the agenda from "walk the top deals" to "walk the *movements*." Only deals that changed tier, changed close date, or changed amount get airtime, plus anything the model and the rep disagree on. This typically cuts the meeting in half and doubles its information content.

The third is the sandbagging equilibrium, which is quieter and more corrosive than inflation. A team that has been punished for missing commit learns to under-commit, and the forecast becomes reliably low. The tell is a commit close rate above 96% combined with consistent overachievement against the committed number. Counter: measure absolute variance, not signed variance. Beating your commit by 18% is a forecasting failure, not a win, because Finance made hiring and spend decisions against a number that was wrong.

The fourth is the model-as-authority trap. Once an org buys a forecasting tool, there is a strong pull to let the score be the answer. But models trained on your history encode your history, including its biases; a model trained during a period when your team systematically under-qualified enterprise deals will keep discounting enterprise deals after you have fixed qualification. Counter: retrain on a rolling window, and treat every large rep-versus-model gap as a two-way audit — sometimes the rep is wrong, and sometimes the model is stale.

The fifth is the missing feedback loop. Reason codes get captured and never read. Counter: a standing quarterly review that ranks reason codes by frequency and by dollars, with an explicit decision for the top three: is this a coaching item, a qualification-criteria item, or a product/process item outside sales entirely? If "security review opened late" is the top code, the fix is a pre-emptive security packet in the sales kit, not a conversation about rep rigor.

The sixth is scope leakage across revenue types. New business, expansion, renewals, and services all get summed into one committed number, but only new business runs on the tier discipline. Counter: apply the same evidence-gate logic to every revenue stream, with stream-appropriate artifacts. A renewal's Commit evidence is a confirmed budget line and an engaged sponsor, not an auto-renew clause. A services forecast's Commit evidence is a signed SOW and a staffed start date.

The seventh, and the one that quietly determines whether any of this survives, is ownership ambiguity. If nobody has standing authority to downgrade a deal over a sales leader's objection, the tiers will inflate at exactly the moments accuracy matters most. The clean structure gives RevOps stewardship of the definitions and the right to challenge any assignment failing documented criteria, gives the sales leader ownership of the final number and the right to override with written justification and a reason code, and gives Finance an audit right over a sample of deals each month. Overrides are fine — unlogged overrides are not. The audit trail is what makes the override survivable, because next quarter you can look at whether the overrides were right.

Related questions

How do you distinguish a slipped deal from a lost deal in the forecast?

A slip closes in a later period with the same buyer and scope; a loss ends the opportunity. Track them with separate reason codes and separate rates, because slips indicate timing calibration problems while losses indicate qualification or competitive problems, and the interventions are opposite.

Should the pipeline tier feed pipeline-coverage targets?

No. Use Total Pipeline for coverage ratios and the Pipeline tier for current-period forecasting. Mixing them means your coverage number silently inherits current-period optimism, causing under-investment in demand generation while the dashboard shows adequate coverage.

How does the design change for a high-velocity SMB motion?

Fewer tiers, lighter criteria, more reliance on aggregate close rates. With 150+ deals a quarter and low variance in size, stage-weighted math is accurate enough. Reserve evidence-gate rigor for any deal above roughly three times your median ACV.

Who should present the number to the board — RevOps or the sales leader?

The sales leader owns the number publicly; RevOps owns the method and the data behind it. Splitting it this way keeps accountability with the person who can influence the outcome while keeping the measurement independent of that person's incentives.

How long does it take to see accuracy improve after redesigning tiers?

Expect roughly two full quarters. The first quarter surfaces how badly the old definitions were drifting; the second is the first clean read, because you need one complete cycle of deals that entered tiers under the new criteria rather than being grandfathered in.

FAQ

What is the practical difference between Commit and Best Case?

Commit requires buyer-side proof — a mutually agreed close plan, resolved redlines, an identified procurement path, and a known signatory. Best Case requires exactly one named blocker with a named owner and an expected resolution date. The distinction is evidentiary, not emotional: if you cannot name what is missing, the deal is not Best Case, it is Pipeline.

How do you stop reps from inflating commit at quarter end?

Make downgrades automatic rather than negotiated. Tie tier eligibility to CRM-validated fields, apply a staleness clock so evidence expires, and measure absolute variance rather than signed variance so both inflation and sandbagging register as errors. Inflation is a design problem before it is a behavior problem.

Do you need an AI forecasting tool to run this?

No. The tiers, evidence gates, reason codes, and variance band deliver most of the accuracy gain on their own. A model overlay adds value primarily as a disagreement detector — it flags deals where rep judgment and historical patterns diverge sharply. If you cannot afford a tool, a simple close-rate table by deal-size decile and stage gets you a usable baseline.

What variance band should we publish to Finance?

Derive it from your own trailing eight quarters of commit-to-actual variance rather than adopting an external figure. Set the band to cover the bulk of that historical spread, then narrow it as the quarter progresses — widest in month one, tightest in the final weeks. A band you consistently blow through is a band set by wishful thinking.

Can a deal skip Best Case and go straight to Commit?

Yes, and it should be uncommon. It happens legitimately when a fast-moving buyer completes procurement and legal in a compressed window. Require the same evidence set regardless of path. If skips are frequent, your Best Case criteria are too demanding and deals are being held out of a tier they qualify for.

How does this apply to renewals and expansion rather than new logo?

The same evidence logic applies with different artifacts. Renewal Commit evidence is a confirmed budget line and an engaged sponsor, not an auto-renew clause — auto-renew with a silent buyer is Best Case at most. Expansion Commit evidence is a funded use case and an identified approver. Forecasting all revenue streams on one taxonomy is what makes the consolidated number trustworthy.

Sources

• https://hbr.org/2010/07/stop-losing-sales-to-customer-indecision
• https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/the-multiplier-effect-how-b2b-winners-grow
• https://www.gartner.com/en/sales/topics/sales-forecasting
• https://www.salesforce.com/sales/analytics/sales-forecasting/
• https://hbr.org/2017/06/how-to-improve-your-sales-forecasting
• https://www.bain.com/insights/topics/sales-and-marketing/
• https://sloanreview.mit.edu/article/the-perils-of-bad-forecasting/
• https://www.cfo.com/
• https://www.forrester.com/blogs/category/sales-operations/

Related on PULSE

• [How do you build a tracking system for deal slippage that distinguishes between forecast inaccuracy, AE optimism, and structural process problems?](/knowledge/q9520)
• [How should a 2027 sales org design situational case studies for AE interviews?](/knowledge/q12558)
• [Why do 2027 B2B buyers trust peer reviews over AI-generated case studies when evaluating consolidated vendors?](/knowledge/q16354)
• [For a founder with sales experience vs a non-sales founder building a sales org for the first time, does the case for deal-closing-first still hold, or do they need different sequencing?](/knowledge/q9556)