Not every lead has the same probability of buying. Propensity models use data and statistical methods to quantify this probability. In B2B sales, they enable objective prioritization and more efficient resource allocation.
What Are Propensity Models?
Definition
A propensity model is a statistical model that predicts the probability of a specific behavior – in sales context, typically purchase probability.
| Component |
Description |
| Input |
Lead/company characteristics |
| Model |
Statistical algorithm |
| Output |
Probability value (0-100%) |
Distinction from Lead Scoring
| Aspect |
Lead Scoring |
Propensity Model |
| Basis |
Often rule-based |
Data-based |
| Weighting |
Manually set |
Learned from data |
| Output |
Point score |
Probability |
| Validation |
More difficult |
Measurable |
Propensity models can be considered an advanced form of lead scoring.
Types of Propensity Models in B2B
Conversion Propensity
| Model |
Prediction |
| Lead-to-Opportunity |
Will the lead become an opportunity? |
| Opportunity-to-Won |
Will the opportunity be won? |
| Trial-to-Paid |
Will the trial user convert? |
Churn Propensity
| Model |
Prediction |
| Churn risk |
Will the customer cancel? |
| Downgrade risk |
Will the customer reduce? |
Upsell/Cross-Sell Propensity
| Model |
Prediction |
| Upsell probability |
Will the customer buy more? |
| Cross-sell probability |
Will the customer buy other products? |
Building a Propensity Model
1. Data Collection
| Data type |
Examples |
| Firmographics |
Industry, size, revenue, region |
| Engagement |
Website visits, content downloads |
| Historical data |
Previous purchases, interactions |
| External data |
Buying signals, news |
2. Feature Engineering
Characteristics that can serve as predictors:
| Feature |
Type |
Source |
| Company size |
Firmographic |
Commercial register |
| Industry |
Firmographic |
Register, website |
| Website activity |
Behavioral |
Analytics |
| Email opens |
Engagement |
Marketing automation |
| Days since last contact |
Timing |
CRM |
| Number of stakeholders |
Deal characteristic |
CRM |
3. Modeling
| Method |
Characteristic |
| Logistic regression |
Classic, interpretable |
| Random forest |
Robust, feature importance |
| Gradient boosting |
High accuracy |
| Neural networks |
Complex patterns |
For many B2B applications, simpler models like logistic regression are sufficient and more interpretable.
4. Validation
| Metric |
Description |
| AUC-ROC |
Discriminative ability |
| Precision/Recall |
Hit accuracy |
| Lift |
Improvement vs. random |
| Calibration |
Do probabilities match? |
Application in Daily Sales
Lead Prioritization
| Propensity Score |
Action |
| >70% |
Handle immediately, senior account executive |
| 40-70% |
Actively pursue, regular process |
| 20-40% |
Nurturing, automated |
| <20% |
Do not actively pursue |
Resource Allocation
| Decision |
Propensity support |
| Which leads first? |
Sort by score |
| Who handles? |
Senior for high score |
| How much effort? |
Proportional to score |
| Which channel? |
Based on preferences |
Forecasting
Propensity scores can feed into pipeline forecasting:
| Opportunity |
Value |
Propensity |
Weighted Value |
| Lead A |
EUR 100,000 |
60% |
EUR 60,000 |
| Lead B |
EUR 50,000 |
80% |
EUR 40,000 |
| Lead C |
EUR 200,000 |
20% |
EUR 40,000 |
Data Sources for B2B Propensity Models
Internal Data
| Source |
Data |
| CRM |
Opportunities, wins/losses, activities |
| Marketing automation |
Engagement, campaigns |
| Product |
Usage (for SaaS) |
| Support |
Tickets, satisfaction |
External Data
| Source |
Data |
| Company data |
Firmographics, financials |
| Intent data |
Research behavior |
| Technographics |
Tech stack |
| News/Events |
Trigger events |
Challenges
Data Quality
| Problem |
Impact |
| Incomplete data |
Features not usable |
| Incorrect data |
Wrong predictions |
| Outdated data |
Irrelevant patterns |
| Too little data |
Model not robust |
Sample Bias
| Bias |
Description |
| Survivorship |
Only successful deals analyzed |
| Selection |
Only certain leads pursued |
| Feedback loop |
Model influences its own data |
Interpretability
| Requirement |
Significance |
| Sales understands score |
Acceptance, usage |
| Explainable factors |
Action recommendations |
| Transparency |
Trust |
Best Practices
Model Development
| Practice |
Description |
| Clean training data |
Won and lost deals |
| Feature selection |
Relevant, available characteristics |
| Cross-validation |
Robust evaluation |
| Regular retraining |
Keep model current |
Implementation
| Practice |
Description |
| CRM integration |
Scores in workflow |
| Thresholds |
Clear action guidelines |
| Feedback |
Sales reports results |
| Monitoring |
Monitor performance |
Organizational
| Practice |
Description |
| Buy-in |
Involve sales |
| Training |
Explain usage |
| Piloting |
Start small |
| Iteration |
Continuously improve |
Measuring Success
KPIs
| KPI |
Description |
| Lift |
Conversion high-score vs. all |
| Efficiency |
Deals per activity |
| Cycle time |
Shortened sales cycle |
| Win rate |
Close rate |
A/B Testing
| Comparison |
Measurement |
| With propensity vs. without |
Conversion, efficiency |
| Different thresholds |
Optimal cutoffs |
| Model versions |
Performance comparison |
Propensity Models and ICP
Interplay
| Concept |
Role |
| ICP |
Defines target group |
| Propensity model |
Prioritizes within target group |
Integration
- ICP filters generally suitable companies
- Propensity model prioritizes by purchase probability
- Sales focuses on high-propensity/ICP match
Technical Implementation
Build vs. Buy
| Approach |
Advantages |
Disadvantages |
| Own model |
Customizable, control |
Effort, expertise needed |
| Vendor solution |
Fast, support |
Less customizable, costs |
| Hybrid |
Balance |
Complexity |
| Category |
Examples |
| CRM-integrated |
Salesforce Einstein, HubSpot |
| Specialized |
6sense, Demandbase |
| DIY |
Python (scikit-learn), R |
| BI platforms |
Tableau, Looker |
Limitations of Propensity Models
What They Cannot Do
| Limitation |
Explanation |
| Causality |
Correlation ≠ cause |
| New segments |
No historical data |
| Black swans |
Unforeseen events |
| 100% accuracy |
Always uncertainty |
Human Factor
| Aspect |
Significance |
| Relationships |
Not fully modelable |
| Timing |
External factors |
| Negotiation |
Individual |
| Intuition |
Complements data |
Propensity models are a tool for support, not a replacement for sales competence.
Conclusion
Propensity models quantify purchase probability based on data and enable objective prioritization of leads. In the B2B context, they combine firmographics, engagement data, and historical patterns into a probability value.
Success depends on data quality, clean modeling, and consistent use in the sales process. Integration into CRM systems and clear action guidelines for different score ranges are decisive.
As a complement to ICP and classic lead scoring, propensity models enable data-driven sales management – provided the human factor is not forgotten.
Identify target customers: With Firmium, you receive the company data that feeds your propensity models.