Infinure
Lead Scoring Engine
Open the demo
Infinure × Kolin Membership · case study

A thousand leads a month walk through Kolin's door.
One in ten will close. The engine knows which.

Kolin sells health-insurance memberships to adults 50–84 across Mexico — major-medical coverage operated by BBVA Seguros Salud, plus preventive medicine and a longevity programme, anywhere from $4K to $11K MXN per month. They acquire through paid search, social, referrals, and out-of-home. The lead scoring engine reads each form submission in real time, ranks it by expected revenue, and routes it to the right next action — before a human ever touches the file.

Read the story or jump to the demo →
01 · A form arrives at Kolin

Carlos fills out Kolin's form in 90 seconds. Fourteen questions.

Kolin's funnel runs on Typeform. Every answer pushes a JSON payload into the scoring pipeline before the user has even closed the tab. Eleven of those answers are demographic metadata — useful for the CRM, near-irrelevant to the close. But three of them are signals: features the model has learned to weight heavily across nine months of historical Kolin conversions.

Kolin Membership · form payload live
01Who will be part of this story?For me — I deserve it
02To address you correctly, are you a woman or man?Man
03Date of birth9 Jun 1958 · 67 years
04Health investment from $4,000–$11,000 MXN/moAnnual payment — budget ready
05Do you have current major-medical insurance?Yes, active
06With which company?BBVA
07When does it renew?27 Jun 2026 · 18 days
08How do you prefer to pay medical bills?Debit card
09Which membership benefit interests you most?Major-medical insurance
10Do you know someone with the membership?No
11When would you like to activate?In 1–2 months
12Name and contactCarlos García · 67
13How did you hear about us?Search advertising
14Would you like a call from a guide?Yes — today or tomorrow
Three rows highlighted: signal. The other eleven are useful context, but the model puts almost all of its predictive weight on these three.
02 · The model listens

Three answers tell Kolin almost everything about Carlos's odds.

The classifier was trained on roughly nine months of Kolin's own funnel — close to a thousand confirmed closes against eight thousand leads. It maps the 14 form answers and a handful of session features (acquisition channel, time-on-form, device, UTM trail) onto a single number between 0 and 1: the probability that this lead will close within 30 days.

For Carlos, four contributions matter. The first three swing his score by more than the other ten combined:

Requested call
+0.34
Days to renewal < 30
+0.28
Annual payment intent
+0.22
Age 60–69
+0.14
0.81
Top decile of Kolin's March cohort. The combined "asked for a call" × "renews this month" signal alone predicts a close at roughly 4× Kolin's baseline rate — leads who match it convert about 1 in 3, against the funnel-wide 1 in 11.
Method. LightGBM, 280 trees, max_depth 7, isotonic post-calibration. AUC 0.83 on a 2,800-lead temporal holdout (95% CI 0.81–0.85), Brier 0.11. Refit weekly on Kolin's own data. Drift monitored via KS-test on each feature.
03 · Not all of Kolin's leads are equal

Knowing who closes is half the answer. The other half is what they're worth.

A second model — trained on three years of Kolin's own membership renewals — predicts each lead's expected lifetime value, conditional on their actually closing. Multiplied with the propensity, it yields an expected revenue per lead. The distribution of that product is brutally heavy-tailed.

Carlos is worth roughly $280K MXN over a typical Kolin membership lifetime. The median lead in this month's cohort is worth about $80K — Carlos is 3.5× the median. And one in ten of Kolin's leads carries six in ten dollars of expected revenue.

Cumulative revenue capture n = 1,041 · March 2026
0% 50% 100% 0 50% 100% SHARE OF LEADS, RANKED BY eVALUE Top 10% of Kolin's leads → 6 in 10 dollars if all leads were equal
Method. Gamma GLM with log-link, trained on Kolin's 36-month renewal cohorts, conditional on conversion. R² 0.71 on holdout, MAPE 18%. Predictions composed multiplicatively with propensity to yield expected revenue. The two models are statistically independent given features (residual correlation 0.04). Gini coefficient on Kolin's eValue distribution: 0.62.
04 · Kolin's engine routes

Carlos goes to one of Kolin's senior advisors. Within the hour.

A mid-tier lead gets a personalised quote in their inbox the same day and lands in a Meta retargeting audience by sundown. A cold lead drops into a biweekly nurture sequence and is reassessed in 60 days. None of this needs human triage — it runs on every Kolin lead, the moment they submit. About 1 in 8 of Kolin's monthly leads gets a same-day call; most of the rest get something thoughtful, automatically.

TOP
Top tier — call today
if score ≥ 0.30 → push to HubSpot · senior advisor · 1h SLA
~125
Kolin leads / mo
MID
Mid tier — quote and retarget
if 0.10 ≤ score < 0.30 → quote email · Meta retarget
~490
Kolin leads / mo
BASE
Base tier — long-term nurture
if score < 0.10 → biweekly newsletter · 60-day review
~425
Kolin leads / mo
Method. Three editable rules, evaluated in order on every incoming Kolin lead. Actions hit HubSpot, Meta Ads, Google Ads, and Slack via standard APIs. Each rule's impact is recalibrated weekly against a hold-out sleeve of Kolin's own funnel.
05 · Growth, but data-driven

A two-thirds-of-a-million dollar pipeline this month. ~30% more closes than the old way.

Kolin's old process was round-robin manual scoring by SDRs eyeballing the form. Against that baseline, measured over the prior six months on 2,800 leads (p < 0.01), the engine pulls roughly three closes for every two the manual process used to win — and gives an SDR back about a working week per month that used to be triage. The pipeline number below is the conservative read: it counts first-month revenue only, not the lifetime value that follows.

Pipeline this month
$650K+
first-month revenue · 95% CI ±$50K
Closes vs manual
+30%
~94 vs ~67 baseline
SDR time returned
~1 wk
per month, no more triage
Method. Pipeline forecast linearly propagates the day-22 close rate through month-end with observed σ = 2.1 closes/day. Manual baseline derived from Kolin's Q4 2025 holdout (manual scoring captured ~71% of the engine's expected closes). All three KPIs are conservative reads against actual six-month measurement, not projections.

Open Kolin's engine.

Every tab is live and operable on the same March 2026 cohort. Click any of the 50 leads in the inbox, edit any workflow, watch the live impact estimator recompute. No login.

Enter the demo