GTM Data Science
Get in touch
Your customer base already shows who is worth acquiring. I route that back into your funnel.
Marketing optimizes for conversion. Sales optimizes for win rate. CS inherits whoever made it through. Every function hits its number, and the customer base still degrades. The outcome that matters is lifetime revenue, and it arrives long after the hand-off to CS, where the people who chose who to sell to never see it.
So the system optimizes for the customers you are best at acquiring, not the ones that fit best, that you could grow with. It is what a funnel does when feedback does not flow back. CS feels the outcomes, but does not control the inputs. As a result your customer base slowly drifts and it surfaces years later as eroding GRR, dampened NRR and a forecast nobody trusts.
The fix is not another tool, it is closing the loop. Observe what happens after the sale. Analyze which customers fit best, the ones you are good at keeping and expanding. And route that back to the people choosing who to sell to next.
Work from the inside out. Start where the data is richest and the payback is fastest, inside the base you already have, then push what you learn back toward acquisition. Each step stands on its own, and each one produces what the next step needs. Nobody has to fund a two-year programme to find out whether this works.
01 Churn risk
Composite health scores fail in three ways at once. They over-flag non-churners, they under-flag churners and they lag. The reason is that they are not built on actual observations, but on beliefs. And almost nobody validates the results against churn that actually happened.
The information sits in the historic renewals that have already closed, in what the accounts that left had in common that the ones that stayed did not. Churn risk and its drivers come out of those outcomes. These observations determine which signals count.
The churn risk trajectory per account together with the revenue at stake naturally give you a prioritization. It tells you where it's worth intervening, where to leave alone, and where the account was never going to stay. For monthly contracts, without a clear renewal gate, the model additionally predicts when the account is likely going to churn, giving you enough time to act.
73% of CS leaders say their health scores can't predict churn.
ChurnZero, 2025 Customer Revenue Leadership Study (n=793)
How well does your health score predict churn? Run a quick self-check.
02 Expansion propensity
The same behavioral base answers the opposite question, and it is not symmetric with churn. Seat and tier growth are discrete events. Usage and consumption growth is a trajectory. Accounts also hold flat or quietly contract, which is signal too.
Either way, the ranking that matters is not current ARR. It is expected trajectory: which accounts, given how comparable customers developed, still have headroom, and which are already at their ceiling. That changes where CS attention goes, and it separates the accounts a play actually moved from the ones that were going to grow regardless.
03 LTV-weighted lead scoring
This is the most commoditized of the three, and worthless without the first two, because almost every scoring model is trained on conversion. Train on conversion and you get more of whoever converts. That is the drift, automated in your scoring model.
Once retention and expansion dynamics are clear, LTV can become an intentional component in your scoring model. The firmographic, technographic and early behavioral signature of high-LTV segments goes back into scoring and qualification. The loop runs on leading indicators of lifetime and value, such as time-to-value and adoption, tested against outcomes rather than assumed to matter.
Churn and expansion information flowing into lead scoring closes the loop. Better-fit accounts enter the base, the signal gets cleaner, and improvements compound in the next iteration.
Effort goes to accounts where intervention has the biggest revenue impact, instead of to whoever is loudest or closest to renewal, directly moving GRR & NRR.
Renewal and expansion forecasts built on data, probabilities and segment-level structure, not on sentiment.
Not just which accounts are at risk, but the behaviors you should actually watch as leading indicators of churn and expansion.
Targeting anchored to the segments that retain and expand, and re-anchored as the evidence changes.
I am Peter Eigenschink, an independent data scientist based in Vienna. I focus on go-to-market analytics in B2B SaaS.
I have been working on productive decision systems rather than dashboards: revenue forecasting, scenario planning, and price optimization in B2B, procurement optimization, pricing, and lost sales estimation in retailing, lead management in insurance. I obtained a PhD in quantitative marketing from WU Vienna, focused on statistical modeling, demand estimation and optimization, for which I was awarded the research prize from the Austrian Retail Association in 2026.
Before and alongside that, fifteen years as a software engineer and consultant, across retail, insurance, the public sector and SaaS. I have also run a small SaaS product on my own for close to a decade, which is a useful practical reference point for what I work on in GTM.
Over the past year I conducted a research project on how revenue teams actually make decisions with data. I ran a series of qualitative interviews and conversations with 40+ RevOps and GTM participants across Europe, the US and Asia. The consistent finding: analytical effort concentrates on acquisition, while post-sales behavioral data, the richest longitudinal dataset in the company, stays largely unused. This page is what I concluded from that.
If you own a revenue number at a $5–50M ARR B2B SaaS, your renewal history and product usage data already answer that, and which accounts still have room to grow. Happy to look at it with you.
