A6 · Evidence case · independent-audit (RevOps team internal analysis, shared at SaaStr 2024)

ML-based lead scoring model trained on 18 months of closed-won/lost data; scored inbound leads in real time and routed to SDR or AE based on score band

Pipeline-to-close rate +31%; SDR time on high-intent leads increased from 34% to 67% of capacity; ACV of model-routed deals 19% higher — The model only works as well as the outcome labels. Teams that skip 'reason for loss' hygiene in CRM see model drift within 6 months.
moderate evidence neutral Last updated 2026-06-18

What was built

ML-based lead scoring model trained on 18 months of closed-won/lost data; scored inbound leads in real time and routed to SDR or AE based on score band

Company type: Mid-market SaaS (HR tech)

Tier map

Tier 1 execution: pipeline quality optimization

Tier 1 · Execution C6 Agentic hub →

Human-in-the-loop design

Human sets score-band thresholds quarterly; routing is automated; AEs can override routing with a flag

Results

Pipeline-to-close rate +31%; SDR time on high-intent leads increased from 34% to 67% of capacity; ACV of model-routed deals 19% higher

Source trust

neutral independent-audit (RevOps team internal analysis, shared at SaaStr 2024)

This case is rated neutral — the source has no commercial stake in the outcome and the methodology is described in sufficient detail to evaluate the claims. These are the highest-confidence cases in the database.

Failure mode observed

Model overfit to one customer segment; required retraining when company entered enterprise market

Transferable lesson

Key lesson: The model only works as well as the outcome labels. Teams that skip 'reason for loss' hygiene in CRM see model drift within 6 months.

How to cite

@misc{shalvi_gtm_evidence_predictive_lead_scoring_saas_2026,
  author    = {Singh, Shalvi},
  title     = {ML-based lead scoring model trained on 18 months of closed-won/lost data; scored — Agentic GTM Evidence Case},
  year      = {2026},
  note      = {Source trust: independent-audit (RevOps team internal analysis, shared at SaaStr 2024). Methodology: moderate.},
  url       = {https://shalvisingh.com/gtm/evidence/predictive-lead-scoring-saas}
}

Singh, S. (2026). *ML-based lead scoring model trained on 18 months of closed-won/lost data; scored — Agentic GTM Evidence Case*. GTM World Model. Retrieved from https://shalvisingh.com/gtm/evidence/predictive-lead-scoring-saas

← All evidence Annual Index → C6 Agentic hub →