Solutions / Conversion scoring

Conversion-probability scoring

Every lead and deal in your CRM gets a win probability learned from your own closed history, with the reasons behind it — so the pipeline is worked in order of what will actually close.

Sales RevOps ML
100% of deals
scored, refreshed daily
Reasons shown
not a number without an explanation
Calibrated
70% means 70% closed
Who it is for

Sales and RevOps leaders with a CRM containing at least a few hundred closed deals and reasonably consistent stage data.

Short answer

A model trained on your won and lost history predicts the probability that each open deal closes, from firmographics, source, engagement, stage history and how the deal has actually moved. Every score comes with the factors that drove it, so a rep sees why a deal ranks high, and the forecast rolls up from probabilities that have been calibrated against reality rather than from stage percentages someone set years ago.

The problem

Priority is decided by whoever called most recently, and the forecast is a stage percentage nobody believes.

01

Reps work the loudest deal

Attention goes to the prospect who replied this morning, not the one most likely to sign. Good deals go cold in the middle of the list while someone chases a tyre-kicker.

02

Stage percentages are fiction

Proposal equals sixty percent because a consultant said so in 2019. Roll that up across a pipeline and you get a forecast that is precise, official and consistently wrong.

03

Nobody knows why a deal is good

Even when a score exists, it arrives as a bare number. Reps ignore numbers they cannot explain to their manager, and the tool quietly stops being used.

How it works

From trigger to result, step by step.

01

Audit what the CRM actually holds

Closed-won and closed-lost with dates, stage history, source, engagement, and whether losses are recorded honestly. A CRM where nothing is ever marked lost cannot train anything, and we say that before starting.

02

Build features from behaviour, not opinion

Time in stage, number and direction of touches, response latency, meeting attendance, discount asked, firmographics. Rep sentiment fields are excluded — they predict the rep, not the deal.

03

Train, then calibrate

The model is trained on older deals and validated on recent ones, then calibrated so a stated 70% actually closes about 70% of the time. Uncalibrated probabilities make forecasting worse, not better.

04

Put the score and the reasons in the CRM

A probability, a trend arrow and the top factors land on the deal record in HubSpot, Pipedrive or Salesforce. Views and queues sort by it, so priority changes without anyone learning a new tool.

05

Measure against what actually closed

Every month, predicted versus realised is compared by segment and by rep. Drift is visible, retraining is scheduled, and the accuracy conversation happens with numbers instead of opinions.

Before / after

What changes on the ground.

Today, by hand
×Priority follows the most recent reply
×Forecast built on fixed stage percentages
×Scores arrive without reasons and get ignored
×No comparison between forecast and outcome
With the automation running
›Win probability on every open deal, daily
›Forecast rolled up from calibrated probabilities
›Top factors shown on each deal record
›Predicted versus realised reviewed monthly
What you get

Delivered, not demoed.

›A CRM data audit that says plainly what can and cannot be learned
›A calibrated win-probability model trained on your closed history
›Score, trend and top factors written back to your CRM
›Pipeline views and queues sorted by probability
›Monthly accuracy reporting and a retraining schedule
Built with

We build in your stack rather than moving you onto ours. The list below is what this solution most often connects to — other systems are a scoping question, not a blocker.

Python scikit-learn XGBoost SHAP PostgreSQL HubSpot Pipedrive Salesforce Airflow
Time to production5–8 weeks
Build priceFixed quote
First stepFree mini-audit
Honest limits

When this is not the right solution.

·Under a few hundred closed deals, the model memorises rather than learns. Clear criteria and a disciplined pipeline review will serve you better.
·If losses are never recorded — deals just go quiet — there is no negative class to learn from, and that is a CRM hygiene project first.
·If every deal is bespoke and takes a year, there are too few comparable cases for a model to add much beyond a good qualification framework.

Questions we get about this one

Lead scoring ranks new inbound leads on fit and intent; this predicts whether an open deal closes, using how it has moved. Most teams end up running both, at different points in the funnel.

Bring us the process that hurts.

The mini-audit is free: we take your version of this process apart and tell you plainly whether automating it pays. If it does, you get a scope and a fixed price.