What a credible process considers
These elements keep the analysis tied to live use.
- feature timestamps
- pre-assignment availability
- contract and sale fields
- stage history
- outcome proxies
- pipeline validation
Build predictions from what was known before the assignment.
Prevent target leakage in sales models by excluding contract, post-appointment, stage, and outcome fields unavailable at decision time.
Strategy guideUse your own evidence
Start with historical appointments, realistic eligibility rules, and outcomes that have had time to mature.
Your Opportunity
What is data leakage in a sales prediction model?
Data leakage occurs when training features contain information created after the prediction decision or information that directly reveals the outcome. It produces impressive metrics that will not survive live use.
What matters most
- feature timestamps
- pre-assignment availability
- contract and sale fields
- stage history
Build the analysis around the decision
The useful prediction is not merely whether an opportunity will close. The system must estimate how the expected outcome changes across the eligible representative choices, using only context available at assignment time.
The ranking becomes an operational policy only after eligibility, capacity, territory, and workflow requirements are applied.
Common analytical mistakes
These shortcuts can make model performance look better than it is.
- using sale date
- using contract price
- using final stage
- using post-appointment notes
- fitting preprocessing before the split
How to evaluate the result
| Measurement | Why it matters | How to use it |
|---|---|---|
| Held-out log loss | Tests whether the method is accurate, stable, calibrated, and economically useful. | Compare appointments that have had enough time to reach an outcome |
| Brier score | Tests whether the method is accurate, stable, calibrated, and economically useful. | Compare similar periods, lead sources, and assignment approaches |
| Precision-recall performance | Tests whether the method is accurate, stable, calibrated, and economically useful. | Compare appointments assigned to recommended representatives |
| Recommendation alignment | Tests whether the method is accurate, stable, calibrated, and economically useful. | Review prediction accuracy alongside actual sales and revenue |
| Revenue per appointment | Tests whether the method is accurate, stable, calibrated, and economically useful. | Check for changes in lead mix and team performance before changing your approach |
From CRM history to a live, measurable decision
Connect your sales history
Isotope Labs organizes your appointments, representative assignments, and outcomes into a history we can evaluate, using information available before each assignment.
Test predictive accuracy
We test predictions against historical outcomes kept separate from model training to identify reliable matches between opportunities and sales representatives.
Backtest realistic assignments
We evaluate historical assignment scenarios that reflect representative eligibility, territories, availability, and workload limits.
Track adoption and sales results
See which recommendations your team uses and how those appointments perform, including close rate and revenue per appointment as outcomes become available.
Common questions
What is data leakage in a sales prediction model?
Data leakage occurs when training features contain information created after the prediction decision or information that directly reveals the outcome. It produces impressive metrics that will not survive live use.
What data should be used?
Use information known before assignment, a stable representative identifier, appointment date, and a mature won or lost outcome. Use reliable information that was available before the appointment was assigned.
How should performance be validated?
Use held-out data, compare against simple baselines, review ranking and probability metrics, and simulate the actual assignment constraints.
What should be monitored after launch?
Monitor recommendation coverage, alignment, mature win rate, revenue per appointment, calibration, agent capacity, and changes in source or opportunity mix.
Continue the evaluation
Strategy guide
Distribute high-value home-service appointments by expected fit.
Compare rotation, rules, and machine-learning-driven assignment for home-service sales teams that need to improve conversion from existing demand.
Strategy guideBalance workload without sacrificing the value of better matching.
Compare equal rotation, constrained optimization, and unrestricted assignment when balancing sales-rep workload and conversion.
Strategy guideIs your CRM data ready for assignment modeling?
Review the fields, definitions, and history needed for a credible evaluation.
Complimentary fit review
Discover where better sales assignment could improve your results.
Share a representative CRM export or connect your data. Isotope Labs will evaluate data readiness, representative-performance variation, and the potential value of model-guided assignment before a live rollout.
- CRM data-readiness review
- Historical model evaluation
- Backtesting that reflects your team and assignment rules
- Plain-language opportunity review