Infinure
Lead Scoring Engine
Demo · Kolin Membership · B2C MX
Dashboard
Lead Inbox
Workflows
Models
Lead value
March 2026 · Day 22 of 31
7d
Month
YTD
Expected pipeline · Marchi
$658K
of $720K targeti · 91% with 9 days remaining
$0$720K
On pace to close 94 deals this month. 6 short of target.
The shortfall traces to paid social. CPA rose 23% this weeki; audience drift detected upstream.
Cumulative closes vs targetDay 22 — observed
Pipeline composition
1,041 leads · 94 expected closes · $658K total
Current month
TOP
Top tier — score ≥ 0.30
125 leads/mo · 38% close rate · 48 expected deals
$336K
51% of pipeline
MID
Mid tier — 0.10 to 0.30
489 leads/mo · 8% close rate · 39 expected deals
$273K
42% of pipeline
BASE
Base tier — below 0.10
427 leads/mo · 1.6% close rate · 7 expected deals
$49K
7% of pipeline
Method. Expected pipeline = (leads × calibrated P(close)) × first-month ticket ($7K MXN).
Lead quality and return by channel
Ranked by marginal contribution
Trailing 30 days
Total spend $48.2K · Total pipeline $658KTotal contribution +$610K
Method. Contribution = (expected closes × $7K first-month ticket) − channel spend.
Score ↓
Today
Top decile
128 leads
Lead
Score
Next action
Source
This lead
How the model decides

How the model decides

LightGBM classifier · trained on 8,400 leads with 982 conversions · AUC 0.83 holdout · Brier 0.11 · isotonic calibration.

Top eight signals — global SHAP

Requested call
0.34
Days to renewal
0.28
Annual payment
0.22
Age bucket
0.18
Knows a member
0.15
Has insurance
0.13
Referral channel
0.11
Time on form
0.08
3 active workflows · click any to edit
Leads processed
1,041
100% automated
Expected closesi
94
+27 vs manual scoring
Pipeline generated
$658K
First-month expected
SDR hours savedi
~35h
vs manual triage
TOP
Top tier — call today
if score ≥ 0.30 → push to HubSpot · senior advisor · 1h SLA
125
leads/mo
$336K
pipeline
ACTIVE
→
MID
Mid tier — quote and retarget
if 0.10 ≤ score < 0.30 → quote email · Meta retarget
489
leads/mo
$273K
pipeline
ACTIVE
→
BASE
Base tier — long-term nurture
if score < 0.10 → biweekly newsletter · 60-day review
427
leads/mo
$49K
pipeline
ACTIVE
→
Add a new rule — e.g. "Annual-payment referrals" → VIP flow
Active
When it fires — trigger
Conditions reference signals from the model.
MATCH ALL
MATCH ANY
What it does — actions
Run in order when a lead matches the trigger.
Estimated impacti
Recomputed live from the current cohort. Shifts as conditions change.
Leads / mo
125
12% of total flow
Expected conv.
38%
avg per matching lead
Expected closes
48
per month
Pipeline
$336K
first month
Model performance · Last refit March 18, 2026
Holdout
6mo
12mo
The two-model approach

Each lead carries two probabilistic estimates. Their product is the unit decision is made on.

One classifier predicts the probability of closing within 30 days. A second regression predicts the customer's lifetime revenue, conditional on closing. Multiplied together, they yield expected revenue per lead — a single number with a heavy-tailed distribution. The top decile of leads is not ten percent more valuable than the median; it is roughly twenty times more valuable.

Classifier · LightGBM

Propensity to close

Probability that a lead converts within 30 days of form submission, given form responses, source, and session features.

AUC · Holdout
0.83
95% CI 0.81 — 0.85
Brier score
0.11
After isotonic calibration
Calibration · predicted vs observed10 deciles · Wilson 95% CI
Points fall on the diagonal. Bin sizes shown by marker; bars indicate Wilson 95% confidence intervals on the observed close rate.

Top SHAP signals

Requested call
0.34
Days to renewal
0.28
Annual payment intent
0.22
Age bucket
0.18
Knows a member
0.15
Regression · Gamma GLM

Customer lifetime value

Expected gross revenue from a customer over a 36-month horizon, conditional on conversion. Estimated from cohort retention curves and ticket history.

R² · Holdout
0.71
Adjusted, 12-month actual revenue
MAPE
18%
Median absolute % error
Predicted vs actual · 12mo revenue (MXN)Holdout cohort · n = 240
Predictions cluster around the y = x diagonal with widening dispersion at higher LTV. Model under-predicts the highest-value cohort by ~12% on average.

Top regression coefficients

Annual payment
+38%
Plan type · family
+31%
Knows a member
+22%
Age 60–69
+14%
Has insurance
+8%
Lead value · joint distribution of P(close) × LTV
7d
Month
YTD
Expected revenue per lead — joint distribution
Each point is one lead from the master cohort. Color encodes the product. Marginal histograms show each model's distribution; iso-curves are loci of constant expected revenue.
n = 1,041 · March 2026
Distribution stats
Median lead — eValue
$2.4K
Most leads sit here.
Top decile — eValue floor
$48K
A lead in the top 10% is worth roughly 20× the median.
Top decile share of total
61%
10% of leads carry the majority of expected revenue.
Tail weight (Gini)
0.72
Comparable to early-stage SaaS.
Method. eValuei = P̂(close)i · L̂TVi. Predictions independent given features (residual correlation 0.04). Iso-curves at $5K, $25K, $100K. Cohort sampled from a Gaussian copula linking Beta(α, β) for P(close) and LogNormal(μ, σ) for LTV; channel-conditional means.
RoutingKnowing P(close) tells you who to call.
RetentionKnowing LTV tells you who to keep.
AllocationKnowing P × LTV tells you where the next dollar comes from.