Causal retention analytics for subscription businesses

Your win-back campaign might be causing the churn it's meant to stop.

Most teams target whoever looks most likely to leave. Field data says that group barely overlaps with the customers an offer can actually save — and some campaigns make churn worse. Slyth Labs reads your last campaign's own control group and tells you, in dollars, which it was.

Book a Demo See the model work ↓

No deck first. We'll run your numbers live on the call.

What a campaign readout finds

from published field data

6–16%

overlap between "most likely to churn" and "actually savable" customers

+56%

relative churn increase from one well-meant "switch plans" campaign

2.7×

campaign profit gain from re-ranking by value instead of risk

CSV in, one dollar figure out — that's the whole first engagement

Why this happens

Three findings the retention-software industry mostly ignores.

Not hypotheses — outcomes from randomized field experiments at real subscription businesses. This is the evidence Slyth Labs is built on.

6% – 16%

Overlap between a company's "top 10% most likely to churn" list and its "top 10% an offer can actually save" list, across two field studies. Risk and save-ability are mostly different customers.

Ascarza, 2018 — JMR

6.4% → 10.0%

Three-month churn before vs. after a proactive "switch to a cheaper plan" campaign at a telecom — a 56% relative increase. Reminding someone to review their subscription can make them reconsider having it at all.

Ascarza, Iyengar & Schleicher, 2016 — JMR

€1,872 → €5,026

Campaign profit at a TV subscriber base, moving from a classic churn-risk model to one trained directly on expected profit lift per customer — same customers, same budget, different ranking.

Óskarsdóttir et al., 2022 — Mkt. Science

How Slyth Labs works

One file in. One dollar number out.

No integration project, no waiting on a data team for the first read. We don't need a trained model to tell you this — we need your control group.

Send one past campaign

A CSV export: who got the offer, who didn't, and who was still active 30–90 days later. Most retention tools already log this — you're likely sitting on it.

We compare who you targeted to who you should have

Same math as the calculator below, run on every decile of your real customers — risk-ranked vs. lift-ranked vs. profit-ranked.

You get one report, one number

Dollars gained or lost from how the campaign was targeted, plus the exact customer list you should've left alone.

Campaign profit-lift estimator

input → output

Customers contacted10,000

Churn — control (no offer)9.0%

Churn — treated (got the offer)10.5%

Customer value / month$29

Offer cost / customer$4

–

Move the sliders to match a real campaign.

Formula: (control − treated churn) × N × ARPU − (N × cost)

Sample output

What a readout looks like

Illustrative numbers from a mid-size subscription base — not a live customer's data.

Customers ranked by predicted churn risk, deciles 1–10 of 10,000 contacted
Decile	Control churn	Treated churn	Δ pp	$ impact
1 — highest risk	79.4%	82.7%	+3.3	−$2,860
2	61.2%	63.0%	+1.8	−$1,490
3	48.5%	47.9%	−0.6	+$410
4 — best lift	57.0%	47.8%	−9.2	+$6,280
5	33.1%	27.4%	−5.7	+$3,920
6 … 10	—	—	≈0	≈$0

Free first read, no model required.

Send one campaign's data on the call. We'll tell you, in dollars, whether it helped — before you're asked to pay for anything.

Book a Demo →

Get started

Book a demo

30 minutes. Bring one campaign's numbers if you have them — we'll run the readout live.

No slideware. We work from your actual control-vs-treatment numbers on the call, not a generic deck.

You'll know within the call whether your current targeting is likely helping, hurting, or roughly break-even.

Nothing to integrate first. A CSV export is enough to start — no API access required up front.