SIGNAL · 2-WEEK PREDICTABILITY TEST

Prove your churn and expansion were predictable.

In two weeks, on your own data, we replay your last year with the outcomes hidden, then show you what we would have called before it happened.

1,900 €
One-shot, excl. VAT
2 weeks
From extract to readout
1 extract + ~1h
Your total effort

HOW IT WORKS

We replay your history and keep score.

We take what was known at a point in time, hide everything after it, and score every account. Then we check what actually happened.

What we know

everything up to date t

observed events

date t
lead time

Will they churn or grow?

SAVEcancelled before renewal
GROWupgraded plan tier
NOTHINGno change

The result: a portfolio-wide number, in euros, on your own data, before you commit.

THE PREDICTABILITY BACKTEST

The signal is measurable.

Here is what the backtest returns. Two models, scored monthly, one month ahead.

Churn

risk model
18.9xAlerts are 19 times better than chance.
37%When we flag an account, the churn happens within the month 37% of the time, against a 1.9% base rate.
21%Tuned for precision: we catch a fifth of churn events while keeping every alert worth acting on. Move the threshold and you trade catch rate against noise.

861 churn account-months to detect

Expansion

intent model
24.0xAlerts are 24 times better than chance.
55%When we flag an account, the upgrade happens within the month more than half the time, against a 2.3% base rate.
60%Three upgrades in five, caught before they happen.

820 expansion account-months to detect

Measured on a simulated B2B CRM SaaS dataset (ChurnSim, Gold, SSRN 2023): 12 months of history, a 1-month horizon, one score per account per month, at the F1-optimal threshold. Your board is computed the same way, on your data.

SIMULATED ACCOUNTS · SCORED MARCH 2, 2026

Two accounts. Same plan. Same renewal date.

On paper, twins. One is about to leave, one is about to surge. Could you tell which?

What the usage data already knew

ACCOUNT ATLAS

Standard plan€500 / moRenews Mar 1625 seats

Churn drivers

  • Tenure in the silent-churn zone

    351 days, nearing the 1-year disengagement mark

  • Meeting cadence has stalled

    7-day momentum near zero against the 90-day pace

  • Locked in a 6-month billing cycle

    no interim invoice, so disengagement runs unchecked

  • Scheduling volume is fading

    30-day activity far below the 90-day pace

ACCOUNT BLOOM

Standard plan€500 / moRenews Mar 1625 seats

Expansion drivers

  • Seat ceiling hit

    100% seat use, 0 of 25 left

  • Intense recent engagement

    meeting, edit and search activity surging

  • Actively working the pipeline

    opportunity activity rising over 30 days

  • New but fast-ramping

    76 days in, all 25 users active

THE CALL · SIMULATED CHURNSIM ACCOUNTS

14 days before renewal, the call was already made.

Two probabilities, two verdicts, two euro figures, while both still looked identical on the invoice.

Atlas

15.8%

churn probability

2.2% on Feb 2, then 15.8% on Mar 2: eight times the 1.9% base rate, and rising.

SAVE−€500 / mo−€6,000 / yr

pre-renewal save

Bloom

96.5%

expansion probability

Model score on Mar 2, the observation date.

GROW+€500 / mo+€6,000 / yr

estimated upside, upgrade to Advanced

What actually happened

Atlas churned March 16

Flagged 2 weeks earlier, on Mar 2.

−€6,000 / yr

Bloom laddered Standard → Advanced → Premier

Two upgrades in six weeks.

+€18,000 / yr

WHAT YOU KEEP

The deliverable: the Signal Board.

Everything we build to run the backtest is yours.

Metrics Table

One row per account, your metrics, computed from your raw events.

Top Drivers

The behavioural signals that most separate churn from expansion.

Executive Readout 60 min

A live walkthrough, plus a shortlist of actions for the next 90 days.

The Signal Board: churn and expansion recall, precision and lift, with a tunable alert threshold

Let's prove it on your data

Follow Obviable