Praxentic

Foresight

Explainable churn-risk model — score, simulate, understand
Interactive demo · sample workspace
Accounts scored
Active book of business
Avg. churn risk
%
90-day predicted probability
High-risk accounts
Risk ≥ 60
MRR at risk
Σ (MRR × churn probability)

Customer risk scores

Customer Tenure Usage Tickets Disc. Top risk factors Churn risk

What-if simulator

live re-score
%
Product usageweekly active %
Tenuremonths mo
Support ticketslast 90 days
Discountoff list price%
Drag a slider to re-score

Why this score — risk attribution

Feature importance

gradient-boosted model

Global weights the model places on each signal, learned across the full book. The what-if panel shows how these translate into one account's risk. Drag a slider to spotlight its bar.

Risk distribution

Selected (what-if) Selected (actual) Portfolio skews low-risk with a heavy tail — where retention effort pays.
How the score is computed (transparent, no black box)
Each account's 90-day churn probability is a logistic function of four normalized risk signals. Every term is inspectable — nothing is hidden:
risk = σ( 5.5 × ( Σ wᵢ · fᵢ − 0.5 ) )  where σ(x) = 1 / (1 + e⁻ˣ)
  • fᵤₛₐ𝓰ₑ = (100 − usage) / 100 — lower engagement, higher risk · weight 34%
  • fₜᵢ𝒸ₖₑₜₛ = min(tickets / 12, 1) — support friction · weight 28%
  • fₜₑₙᵤᵣₑ = clamp((24 − tenure) / 24) — newer accounts churn more · weight 22%
  • f_disc = discount / 60 — deep discounts flag price sensitivity · weight 16%
The same function scores the table, drives the what-if gauge, and produces the per-account attribution above. Weights shown in Feature importance. Sample data only.