Calibrated to your win rate. Not a generic baseline.
Every CRM claims AI-powered deal scoring. Most pick a number out of a black box. Nynch's deal probability is a Bayesian update: a prior from your closed deals, combined with live signals, and continuously calibrated against your actual outcomes. The score is auditable, the calibration is reportable, and the model learns from you, not from "consultants in general."
Three steps. One probability.
Bayesian inference combines a prior belief with new evidence to produce an updated belief. That is the right shape of model for deal probability, where every deal starts in a different place and every interaction adds new evidence.
Starts from your closed deals
The prior is built from your historical wins, losses, and no-decisions. New users get a sensible default rooted in relationship-led growth research, replaced by a personalised prior as soon as you have enough closed deals.
Updates from live signals
Engagement frequency, response cadence, stakeholder coverage, competitor mentions, deal velocity, commitment completion, and proposal feedback all feed in as evidence. Each signal moves the probability up or down by an auditable amount.
Outcome feedback closes the loop
When a deal closes won, lost, or no-decision, the actual outcome updates the model's weights. The Suggestion Scorecard reports calibration error over time. By month three, predictions typically land within 10 percentage points of reality.
The signals that move the score.
None of these are exotic. They are the signals you would track in a spreadsheet if you had infinite time. The point of Bayesian Deal Probability is that you don't have to.
- Engagement frequency. How often are real conversations happening?
- Response cadence. How quickly are replies coming back?
- Stakeholder coverage. Are you single-threaded or multi-threaded?
- Champion strength. Is the inside advocate active or quiet?
- Competitor mentions. How often does a competitor name show up?
- Deal velocity. Is the deal moving faster or slower than your average?
- Commitment completion. Are mutual action plan items being kept?
- Close-date drift. Has the close date been moved? How many times?
How Bayesian probability compares to generic AI win prediction.
Most "AI win prediction" features use a single black-box model trained on a SaaS-wide dataset. Nynch's Bayesian model is yours by month three.
| Capability | Generic AI win prediction | Nynch Bayesian Deal Probability |
|---|---|---|
| Prior built from your past deals | No (SaaS-wide) | Yes |
| Auditable: which signals moved the number | No | Yes |
| Reports calibration error | No | Yes (Suggestion Scorecard) |
| Model weights update from your outcomes | No | Yes |
| Different score for the same deal under different reps | No | Yes (per archetype, per book of business) |
| Day-1 score and day-90 score reflect different priors | No | Yes |
By month three, two consultants pursuing the same kind of deal will see different probabilities for the same prospect, because their priors come from different closed-deal histories. That is correct behaviour, not a bug. The model is calibrated to who is actually doing the selling.
If a buyer asks "isn't this just regression to the mean across all your customers?" the answer is no. Each customer's prior and weights are private and never pooled.
See the calibration report on a live account.
Book a 30-minute demo. We will walk through one customer's calibration curve over their first 90 days, and show you the per-deal probability breakdown for an active opportunity.