Building predictive Deal Scoring

Introduction
Accurate forecasting remains a constant challenge.

Sales managers know all about the pain of pipelines that swing between optimism and missed targets.

Traditional CRM "probability of closure" fields - often simple percentages based on intuition - don't provide the level of reliability that modern RevOps teams need.

This is where predictive deal scoring comes into its own.
By combining historical data, engagement signals and business context, RevOps teams can move beyond intuition to a transparent, data-driven, predictive model of the current quarter's revenue.

👉 HubSpot, Salesforce and Pipedrive already integrate deal scoring functionalities, while specialized dealrooms such as Qwoty make it possible to associate accurate, up-to-date data with them, to obtain realistic predictive models.

Result:
✅ Better prioritized deals
✅ More reliable forecasts
✅ More focused team energy

Predictive deal scoring is a quantitative measure of the probability of a deal closing, based on actual buyer behavior, history and business context - not just the salesperson's intuition. Unlike probability fields, predictive scoring is dynamic and evidence-based.

This is a major change: predictive scoring doesn't reflect what the salesperson thinks, but what the data reveals about the buyer's intentions.

Most CRMs display scores either on a 0-100 scale, or as a percentage.

The formats seem similar, but the logic differs.

A raw score of 70/100 does not necessarily correspond to a 70% probability of closing - it may represent a relative ranking against other opportunities.

Successful RevOps teams calibrate what "a good score" means. For example:

• 80+: historically associated with win rates 3 to 4 times higher.
• 40-60: indicates a pipeline risk and the need for coaching.
• < 30 : corrélé à des opportunités stagnantes ou de faible qualité.

Forecast accuracy, prioritization and pipeline health signals

Pipeline reviews often reveal the gap between optimism and reality.

A predictive scoring model reduces the noise by highlighting deals supported by real engagement signals.

The results: more reliable forecasts, better prioritization of sales efforts, and early detection of risks in the pipeline.

The minimum viable dataset

A predictive model only works with a data base:

• History of customer/segment deals (won/lost)
• Stage progression
• Key activities (emails, calls, meetings)
• Buyer's intentions (transfer, reading time, number of openings)

Field coverage & data hygiene benchmark

A poorly informed CRM kills predictive scoring.

RevOps teams must impose data hygiene benchmarks to ensure model reliability (e.g. 95% of critical fields completed).

The aim is always to juggle requirements and speed: CRM data completion must not slow down deals, but sales teams must be supported and well supervised: CRM data completion is a challenge for 100% of sales teams.

Alignment with the definition of stages: from "lead" to "closed won".

Harmonizing stage definitions is key to making the model's outputs credible and simplifying pipeline tracking.

This list is not exhaustive, but is based on our own experience. Of course, special cases may require very precise data (planned kick-off date, payment facilities, etc.).

But even with this list of KPIs, deal scoring will be top-notch!

1. Engagement & activity (captured by a DealRoom like Qwoty)

• Number of views of the proposal
• Time spent on proposal (by section: pricing, scope, conditions)
• Number of stakeholders who consulted the document
• Average response time after proposal sent
• Comments or annotations left in the proposal
• Progress of electronic signature actions (initiated, completed)

2. Progression through the sales cycle

• Stage velocity (average speed from one stage to the next)
• Inactivity time in a given stage
• Number of backtracks in the pipeline (e.g. from "Contract Sent" to "Negotiation")

3. Quality of CRM data and fields

• Completion rate of critical fields (budget, decision-maker, deadline, use case)
• MEDDICC-type qualification criteria (identified decision-maker, success criteria, urgency, etc.)

4. Commercial factors

• Annual contract value (ACV)
• Discount rate applied, margin respected (or not with managerial approval)
• Contract duration (1 year, 3 years, etc.)
• Customer segment (SMB, Mid-market, Enterprise)

5. Interactions outside DealRoom

• Number and speed of replies to emails
• Number of meetings with decision-makers
• Level of seniority involved on the prospect side

Two possible models

1. Heuristic baseline: A model based on simple rules (weighting of steps + activity signals) can be deployed in a quarter. Transparent and easy to explain.
2. Machine learning : Training a model on historical data (gains/losses) can detect non-linear patterns. Hubspot's Breeze is a great ally.

Choosing a scale

Whether 0-100 or percentage, the key is to map scores to clear, easily identifiable thresholds (green/yellow/red).

Example of heuristic weighting (out of 100 points)

A simple model might look like this:

• Commitment & activity (40 pts)
  • 15 pts: number of views of the proposal ≥ 3
  • 10 pts: average time spent on the proposal > 5 minutes
  • 5 pts: at least 2 stakeholders have consulted the document
  • 5 pts: comments or annotations left in the proposal
  • 5 pts: electronic signature initiated in the DealRoom
• Velocity & health pipeline (25 pts)
  • 15 pts: steady progress, no stagnation > 30 days in a stage
  • 10 pts: compliance with median time per stage (stage velocity)
• Coverage of strategic fields (15 pts)
  • 15 pts: completion of critical fields (budget, identified decision-maker, deadline, MEDDICC criteria)
• Commercial factors (20 pts)
  • 10 pts: LCA aligned with target KPI (e.g. ≥ threshold defined by segment)
  • 5 pts : remise < 15 %
  • 5 pts: contract duration ≥ 12 months

Final score = 0-100.

👉 Example: a deal with several decision-makers having viewed the proposal, an initiated signature, completed MEDDICC fields and correct pipeline velocity will reach 80+ points, thus considered highly likely.

Advanced patterns

Separate models for new business vs. expansion: signals differ between acquisition and renewal/upsell.
Integrate pricing and competition: pricing intelligence and competitive signals enrich scoring.

Backtesting by cohort

Testing by region, segment or LCA validates that scores reflect actual results.

Demonstrate explanatory factors

Transparency creates trust: showing that "buyer answers" or "executive commitment" explain a high score encourages adoption.

Green/yellow/red thresholds

Linking thresholds to historical success rates (e.g. 80+ = 70% win rate) makes the model actionable. However, this data must be available historically.

Where does the score live?

Ideally, mainly in CRM (HubSpot, Salesforce, Pipedrive), with DealRoom backing up with a dashboard, synchronized in real time.

Score-driven pipeline reviews

The score guides managers towards the next best actions, beyond simple pipeline hygiene.

Views by role

AE: priority deals.
Managers: pipeline health.
Management: forecast reliability.

How does deal scoring differ from lead scoring?

Lead scoring predicts whether a prospect will become an opportunity. Deal scoring predicts whether that opportunity will become a revenue stream.

Why does my CRM display 0-100 when we're talking in percentages?

Each publisher applies its own logic. The key is to calibrate your thresholds according to your past results.

How often should the score be updated?

Near real-time for short cycles. A daily update will suffice for enterprise environments. But the truthfulness of the data at the very moment the sales rep is looking at it, will help him prioritize his actions and therefore, better close 🚀