GTM Data Science
Data Science for GTM

Your health score says the book is healthy. Don't wait for the renewals to correct it.

Behavioral churn and expansion prediction for B2B SaaS, from the data you already have.

Check your health score

Why scores fail

Most scores are never tested.

Most health scores are a weighted mix of lagging signals someone expected to matter when the score was set up: logins, tickets, NPS, CSAT, a handful of usage metrics. The weights are a judgement call made at the start, and they are rarely tested against the renewals that actually closed.

Meanwhile the product ships, the customer mix shifts and the motion changes. The score drifts away from the assumptions baked into it, and it fails in three ways at once.

  • Blind. Accounts leave without the score ever going red.
  • Late. The score turns red after the decision has already been made.
  • Noisy. It flags accounts that were going to renew anyway.

73% of CS leaders say their health scores can't predict churn.
ChurnZero, 2025 Customer Revenue Leadership Study (n=793)

None of this shows on the dashboard. The score keeps reporting a healthy base with the same confidence, and only closed renewals show how far it has drifted. Few teams ever monitor the drift, so it stays invisible while it is already costing money.

What it costs

Noisy costs hours, blind and late cost ARR.

A bad score misguides your CSMs. Noisy, and CSM hours go to accounts that were fine, leaving fewer hours for the accounts actually leaving. Blind or late, and accounts that were at risk never get a touch, because they looked healthy. In a pooled or digital tier they are lost without anyone having looked at them.

Measuring what your score's blindness costs does not need a better score. Answer the two questions:

  • How many accounts did you lose last year without a warning? The accounts that churned without the score ever flagging them, and the ARR they took with them. That is how far the score had drifted by last year's renewals and what it has already cost you.
  • How much ARR are you going to lose silently this year? At last year's miss rate, the ARR and the number of accounts likely to walk out without a flag this year. That is what is at stake now, while there is still time to act.

Answer them with a quick self-check.

Work with me

Identify the cause, then fix it.

The self-check puts a rough estimate on what your score has cost you. In a free health score review call, we identify which of the failure modes your score suffers most from. Sizing all three accurately needs your account data.

Free health score review

Thirty minutes, free. If you have not run the self-check, we run it together on your numbers. If you have, we go past blind to late and noisy: how far ahead your flags actually land, and how much of what the score flags was never going to churn. You leave knowing which of the three you are dealing with.

Book a free score review

Health score audit

Accurately measuring the failure modes needs your account data. The audit takes last year's renewals and returns what summary counts cannot reach: how far ahead each flag actually landed, against the lead time your save motion needs, and what your CSMs spent on false alarms, separating the reds that renewed anyway from the ones a CSM saved. Per segment, so you know where the score can be trusted and where it cannot. About a week, for a fixed fee agreed before we start.

Fixing your score and beyond

The audit puts numbers on the score's failure mode. The next step is fixing it:

  1. Churn risk model. Trained and validated on your closed renewals, pushed back to your CSP, monitored so it does not drift the way the health score did. Along the way you learn which signals actually drive churn in each segment. Revenue at risk ranks your accounts for CSMs to act.
  2. Expansion propensity model. The same path for expansion propensity: which accounts are likely to expand, when, by how much, and which signals come first in each segment. Revenue opportunity ranks your accounts for CSMs or AMs to act.
  3. Capacity and revenue planning. Whether next year's goals are reachable with the coverage you have: what the current score lets the team hold and grow, what a better one would, and what that means for headcount and budget.
Churn risk over the months after close. One account's risk rises and is flagged well before renewal, ahead of the window where it is too late to act. A comparison account stays low.
Churn model. Risk rises and is flagged months before renewal, while there is still time to act.
ARR relative to the value at close over the months after close. Some accounts grow, some hold flat and some contract.
Expansion model. Some accounts grow, some hold flat and some contract. The model predicts when they do which.

No new tools required. I work with the CSP, CRM and DWH you already have. The models train on outcomes your accounts have already produced: renewals, churn, upgrades, up-sells, cross-sells and downgrades. I build integrated production systems, not slide decks.

What it buys

Improve the KPIs your book is judged on.

Better churn risk and expansion propensity estimates move four numbers.

  • GRR. Risk flagged while the CSM can still act on it, and accounts you can safely leave alone, both tested against the renewals that actually closed rather than on assumptions.
  • NRR. Expansion surfaced when the account is receptive, not waiting to be asked, and accounts ranked by the ARR they are likely to reach instead of the ARR they carry today.
  • Forecast accuracy. Renewal and expansion forecasts built on segment-level outcomes and likely trajectories, with a range you can defend in a board meeting, rather than on sentiment collected from account reviews.
  • CSM hours. One ranked list carrying churn risk, expansion propensity, and revenue at stake, cut where the team's capacity runs out, so the hours go where intervening actually changes the outcome. Two lists if you split retention and expansion motions between teams.

About me

A B2B SaaS operator with a customer analytics PhD.

I am Peter Eigenschink, an independent data scientist based in Vienna, focused on go-to-market analytics in B2B SaaS. Fifteen years as a software engineer and consultant, and close to a decade operating my own B2B SaaS products, so I know renewals, churn and support load from the operator's seat. For the past six years I have built production decision systems for clients: scoring and routing 30,000+ leads a year across 500+ insurance sales teams, revenue forecasting and price scenario planning for a €100M+ procurement agency, stock optimization and lost sales estimation in retail at similar scale.

My PhD is in quantitative marketing at WU Vienna, the field that models how customers are acquired, retained and lost. I worked with Prof. Thomas Reutterer, known for his research on customer lifetime value estimation. My dissertation was about the sales a retailer never sees: when an item is out of stock the demand is unknown, the loss is invisible and how much to stock next week is unclear. Estimating the risk and timing of a stockout is similar to the problem a renewal book has, where what you need is which accounts are going to leave, how long you have, and what it costs if you are wrong. It won the Austrian Retail Association's research prize in 2026 and went into planning at a large retailer.

Over the past year I spoke with 40+ RevOps and GTM people across Europe, the US and Asia about how revenue teams actually decide with data. The consistent finding: attention mostly goes to acquisition, while post-sales behavioral data, the richest longitudinal dataset in the company, stays largely unused. This page is what I concluded from that.

LTV prediction Churn modeling Time-to-event estimation Causal analysis Experiment design Bayesian methods Machine learning

Do you know why your score costs you?

Blind, late and noisy each need a different repair, and fixing the wrong one costs you a quarter. Thirty minutes on your score, and you leave knowing which one you have.

Book a free score review