People Analytics for a Sales Hiring Funnel
Same LATAM sales team as the onboarding case: 30 to 60 new sales hires a month, onboarding → probation → live sales floor. Data lived in a hiring sheet, CRM exports, a Power BI report that didn't match payouts, and termination records where everyone was "Dismissed".
Problem
Three questions kept coming back and nobody could answer them with numbers: who actually reaches the floor, which leavers we should have kept, and whether paying a fixed salary during ramp-up is worth it. The attrition formula everyone used divided leavers by a headcount of 13 people standing in a stage that 200 people passed through, and returned 1491% "annualized". The arithmetic was right, the meaning was not.
Solution
Built the analytics layer for the LATAM sales funnel: one cleaned dataset, a conversion dashboard, attrition done right, hiring signals with honest caveats, and a payback calculator.
- salaried vs commission hires still working
- 61% vs 14%
- salaried vs commission hires still working
- hiring signals × outcomes
- 22 × 10
- hiring signals × outcomes
- people × fields in the data lake
- 209 × 83
- people × fields in the data lake
How it works
Every number opens into names
Every tile opens into a named list, so any number can be checked person by person.
Only resolved outcomes
Conversions count only resolved outcomes: someone hired three days ago hasn't "failed to reach the floor" yet.
Signals with caveats
A row from the candidate profile: has sales experience · went live 35% · without it 12% · strong · with no sales experience, 100% drop-off within 2 months (n=21). Another one: call-centre experience · 31% vs 32% · no effect, so it was dropped from scoring.
Probation simulation
Switch the lead ceiling (10/12/15/18/20/25) and see who gets cut, how much floor revenue leaves with them, and how conversion shifts.
Outcome
- Leadership wanted to return to commission-only hiring. Salaried hires went from onboarding to the floor at 67.7% (21 of 31) vs 21.6% (11 of 51) on commission, and stayed at 58–61% vs 12–14%. We kept fixed pay.
- About half of the people who left (12 of 23) were worth keeping, which moved the conversation from "hire more" to "lose fewer".
- Call-centre experience was dropped from candidate scoring; sales experience and an offline or teaching background were moved up.
- The probation rule was set on the simulation, not on a gut feeling about the right ceiling.
What I did
- Built a data lake: 209 people × 83 fields, every number traceable to its source.
- Had agents parse 175 CV PDFs (teaching, sales, call-centre experience) and reclassify 108 free-text termination comments into resigned vs dismissed, then re-checked every decision with a second pass.
- Split the metrics: classic attrition for the sales floor (headcount), stage drop-off for onboarding and probation (leavers ÷ entrants).
- Built a conversion dashboard with about 10 tabs: leads, breakdowns, weekly cohorts, newcomers vs veterans, attrition, who leaves, candidate profile, methodology.
- Ranked 22 hiring signals against 10 outcomes, each with a caveat when the sample is thin.
- Built a payback calculator for a sales hire and a simulation of the probation rule ("two sales within N leads").
- Ran a weekly cohort report from late May and took the conclusions to the HRD and the CEO.
Screens
My role
Owned the analysis end to end, from cleaning the data to presenting the conclusions to the HRD and the CEO.
Stack
Google Sheets · Claude (artifacts, and agents for parsing) · Supabase + Next.js (data lake page in the LMS) · Huntflow API for CVs.
Next case
Hiring Funnel Automation