Deal probability

What it is

Modelled. How likely an open deal is to close won, from 0 to 100, read as a percentage, where a higher number means more likely. It exists to replace the stage percentage. Most companies forecast pipeline by multiplying deal value by a fixed number attached to the stage — stage 3 is 30%, stage 4 is 60%, stage 5 is 90%. That looks precise and is structurally wrong: it says every deal at a stage is identical, and it only moves when someone drags a card. Deal probability moves when the deal moves. It is modelled, not measured. A measured value is reproduced by running the same arithmetic over your data. A modelled value is composed: it weighs several things against each other, it is calibrated against deals that actually closed, and it sharpens as your outcome history builds. It is not a guess and nothing about it is unpredictable — the same inputs always return the same number — but the weights behind it were set by looking at what happened, and they are revisited.

How it is calculated

Modelled. Five engines are each scored 0 to 100 and combined at fixed weights.

Engine	Weight	What it reads
Fit	0.28	Structural customer quality — how well the account matches who you sell best to, the win rate in its segment, whether it is profitable to serve, and how likely it is to stay
Velocity	0.25	Momentum — days in stage against the norm for this segment and stage, the shape of the slowdown, stalling, and slippage risk
Behavioural	0.22	Relationship and engagement — seller consistency, buyer reciprocity, how many people are involved, and days since the buyer last started something
Value	0.15	Commercial structure — deal size against the segment median, discount against policy, term, and whether the product mix is complete
Intent	0.10	Active buying intent — where the deal came from, how recent the signals are, how deep the engagement is, and third-party intent

The weights sum to 1.00. Fit carries the most because customer quality is the strongest predictor of whether a deal closes and stays closed. Intent carries the least because it is not fully available until marketing data is connected. The result is bounded by a floor and ceiling set per segment and per stage from your own historical win rates, so the number cannot land somewhere implausible for where the deal actually is. A deal whose expected close date has passed loses two percentage points a week until the date is updated. A confidence score is published alongside it, and it is not the same thing. Probability is how likely the deal is to close. Confidence is how much weight to put on that number — it reads how much of Beacon is switched on, how well the deal’s records resolve, the deal’s age against the segment median, and how stable your pipeline is.

Confidence	What you are shown
Below 40	A wide range rather than a number
40-69	A narrower range
70 and above	A point probability

Tier	When it applies
High confidence	Probability 65 or above, confidence 70 or above
Medium	Between the two
Low	Probability below 40, or confidence below 40
Stalled	A velocity stall is active
At risk	An anomaly is active

Where it comes from

Source	What it supplies	Required
Your CRM — deal object	Name, owner, account, stage, amount, expected close date, created date, source	Yes
Your CRM — company object	The fit engine’s structural inputs	Yes
Your CRM — contact and activity	The behavioural engine — meetings, emails, calls, documents shared, days since last activity	Yes
Your CRM — stage history	The velocity engine	Yes
Marketing data	The intent engine at full strength	No — the engine runs partially without it
Finance data	Sharper commercial structure in the value engine	No

Every engine runs on whatever is connected, with a wider range where less is. This is the clearest example in Beacon of a value that gets better rather than appearing later: with a CRM alone the average error is around 18 to 22 percentage points; with sales and customer data connected it narrows to 8 to 10; with everything connected and a few months of outcome history it reaches 4 to 6. One hard limit. Where a deal’s records cannot be resolved to a single company with enough confidence, no probability is published at all — not a low one, not a flagged one. The value is suppressed and the reason is named. That limit cannot be overridden.

How fresh it is

A deal gets its first score within an hour of being created in your CRM. After that it recalculates on any qualifying signal — a stage change in either direction, a meaningful engagement shift, new activity — plus a full daily pass and a weekly full recalculation that tests recent scores against what actually happened. Every response carries two timestamps. as_of is the moment the value describes. computed_at is when Beacon last worked it out. Read both rather than assuming a cadence.

Currency and rounding

The probability is a percentage with no money value, rounded to the nearest whole number. A separate projected revenue figure is built from it and is not this value — see below.

What changes it

Any of the five engines moving. Time passing without the close date being updated. A deal’s records resolving better or worse. Winning sets it to 100 and losing sets it to 0. What does not change it: anyone’s opinion of the deal. There is no field for a rep’s own call, and the number does not recompute in a dashboard — the dashboard shows it. The weekly cycle can adjust the engine weights, and it is bounded. It compares the previous week’s closed deals against the probabilities they were carrying, finds where the model was systematically off by segment and stage, and applies adjustments of up to three percentage points per engine on its own. Anything larger goes to a named owner. Weights are not adjusted more than once a week, and structural changes wait at least thirty days. The past-close decay cannot be switched off. If a deal’s expected close date has passed, the right response is to update the date, not to stop the decay.

What it is not

Not deal health. That is whether this becomes a good customer if it closes. This is whether it closes. Both can be high, both can be low, and the interesting deals are the ones where they disagree.
Not the projected revenue figure. A separate value applies this probability to the deal amount to produce expected recognised revenue, and a further one rolls that into a 30, 60 and 90-day pipeline corridor. Those are money; this is a likelihood. They are built from it and are not it.
Not the confidence score. Confidence is how reliable this number is. The two are always shown together and mean different things.
Not a predicted renewal date or a predicted term. Beacon builds a provisional contract shape alongside an open deal, and those are predictions on that shape. They are replaced by the confirmed values at close.
Not deal stage. A later stage is not automatically a higher probability, and treating it as one is the practice this value replaces.

​What it is

​How it is calculated

​Where it comes from

​How fresh it is

​Currency and rounding

​What changes it

​What it is not