Your Stage-Probability Table Is Fiction. Here's What Actually Predicts Whether a Deal Closes.

Somewhere in your CRM, there is a row in a settings table that says "Proposal Sent = 60%." Nobody alive remembers who typed that. It was probably copied from a Salesforce demo environment, adjusted slightly to look considered, and has been silently attached to every deal your team has ever run since. It is not a forecast. It is a default. And if you are running a weekly pipeline call off it, you are reading horoscopes and calling it revenue operations.

The number feels precise because it is a number

That is the entire trick. Sixty percent looks like analysis. It sits in a field next to a deal value and your brain multiplies them automatically. You have a weighted pipeline. You have a forecast. Except you do not, because the 60 was not derived from anything. Not from your historical close rates at that stage. Not segmented by deal size, vertical, or rep. Not adjusted for how long the deal has been sitting there. It was typed in by someone who has probably left the company, and it has compounded in false confidence ever since.

This is not an edge case. Pull your own data. Take every deal your team closed in the last 12 months and look at what probability the CRM assigned at the point they entered "Proposal Sent" or equivalent. Then look at the deals that died at the same stage. If your stage-probability mapping is doing any real work, closed deals should have been assigned meaningfully higher probabilities than lost ones at that moment. In most CRMs, they were not. The stage was the same. The probability was the same. The outcome was different. The number predicted nothing.

The rep-variance test exposes it instantly

Here is the fastest way to demonstrate that your stage probabilities are fiction: take five reps who all have deals in "Negotiation" at £80,000. Your CRM says each of those deals is 80% likely to close. Now ask yourself the following questions about each one.

Has the economic buyer spoken to anyone on your team in the last two weeks, or has contact been exclusively with a champion who stopped returning calls? Is there a written mutual close plan with dates the buyer proposed, or one your rep invented in the CRM? Has pricing actually been discussed, or has your rep marked it "Negotiation" because they sent a deck with numbers on it? Is legal involved on their side, or is that still theoretical?

The answers will not be uniform. One of those "80% deals" will have multi-thread engagement, a buyer-initiated timeline, and a legal review already started. Another will have one contact, last touched 19 days ago by your rep, with no commercial conversation on record. They carry the same probability because they are in the same stage. That spread, across five reps on the same team, is wide enough to make the number meaningless as a forecasting input.

What actually predicts close

The teams I have seen run genuinely accurate forecasts, meaning within 5-8% of actual over a rolling quarter, do not rely on stage probability at all. They track a small set of observable, deal-specific variables and they are ruthless about only counting confirmed evidence, not rep assertions.

The three that carry the most predictive weight, in order:

1. Days since buyer-initiated contact. Not days since your rep last sent an email. Days since someone on the buying side reached out unprompted. A buyer who wants to close a deal makes contact. A buyer who is stalling, ghosting, or running a competitive process to validate an existing preference does not. Deals where buyer-initiated contact has gone dark for more than 14 days should be forecast at a fraction of their stated probability regardless of stage. The stage has not changed. The deal has.

2. Multi-thread depth. A single-threaded deal is a hostage situation. If your champion leaves, gets promoted, goes on parental leave, or simply loses internal appetite, the deal goes with them. Genuine multi-thread depth means your team has had substantive conversations with at least two people who sit in different parts of the buying organisation, ideally including someone who controls or influences the budget. Not CC'd on an email. Actual conversations. If you cannot name them, the thread does not exist.

3. Whether a commercial conversation has actually happened. This one is brutal and most reps will tell you it has when it has not. A commercial conversation means the buyer has reacted to a number, asked a clarifying question about pricing, pushed back on a term, or requested a revision. It is not "I sent them the proposal." Sending a proposal is your behaviour. A commercial conversation requires their behaviour. Deals where pricing has been shared but not discussed are not in negotiation. They are in limbo.

If you want a structured way to score these against your current pipeline, the Deal Health / Risk Scorecard gives you a starting framework that applies these variables at deal level rather than stage level.

Why most orgs will not fix this

Building a defensible probability model requires two things most revenue teams do not have in good shape. First, clean historical data: actual close rates by stage, by deal type, by rep cohort, long enough to be statistically meaningful. Second, honest qualification data at the deal level, which means your reps need to have recorded things like "economic buyer engaged: yes/no" and "commercial conversation held: yes/no" with enough consistency that the field is trustworthy.

The moment you try to build that second layer, you find out how thin the qualification discipline actually is. Reps have been marking deals through stages based on their own activity, not on buyer behaviour. Stages represent what the rep did, not what the buyer confirmed. And nobody has checked, because the CRM showed a number that looked like a forecast so the question never got asked.

That is the real reason stage-probability tables persist. Not because they work, but because replacing them with something real means admitting that current pipeline visibility is much lower quality than the board slide implies. That is an uncomfortable conversation. It is also the only conversation worth having.

If you want a baseline on what your forecast accuracy actually looks like before you change anything, the Win Rate & Forecast Accuracy Tracker will show you the gap between stated probabilities and actual outcomes over time. Most teams find it clarifying in the way that a cold shower is clarifying.

What a defensible probability assignment actually requires

A probability number earns the right to exist in a forecast if it is built from at least three inputs: a base rate derived from historical close data for that deal type and size, a modifier based on observed buyer engagement (multi-thread depth, buyer-initiated contact recency), and a binary check on whether a commercial conversation has happened. Without all three, you have a stage label with a percentage sign next to it.

That is not forecasting. It is formatting.

The teams that figure this out do not necessarily build complex models. They usually get there by stripping the forecast back to deals where they can answer yes to all three qualifying questions, and treating everything else as pipeline to be developed rather than revenue to be counted. The forecast gets smaller. It gets more accurate. And the coaching conversation finally has something real to point at.

Your stage-probability table is not wrong because someone made a mistake. It is wrong because it was never designed to be right. It was designed to fill a field.

The tool for this: Deal Probability Scorecard: Three-Factor Forecast Model, free and no signup.