A real case study of how we built a churn prediction and retention automation system that saved $180K in annual recurring revenue.
This is the story of how we helped a B2B SaaS client identify at-risk customers before they churned — and recover $180K in ARR inside a single quarter. The names are anonymised; everything else is exactly how the engagement unfolded.
The Problem
Our client sells a workflow management tool to mid-market operations teams. At the time we got involved they were running roughly 8% monthly churn on their starter tier, which was quietly eroding most of the gains coming from new sales. Worse, most churned customers left without warning — no cancellation reason, no downgrade, no support tickets. The CSM team would look at the weekly churn list on Monday morning and have no idea what had happened.
The root cause was straightforward: nobody was watching the leading indicators. Usage data lived in the product, billing data lived in Stripe, support tickets lived in HubSpot, and nobody had connected the three.
The Solution
We built a churn prediction pipeline in n8n that pulls weekly snapshots of product usage, billing events, and support sentiment, scores each account on a 0–100 risk scale, and triggers a graduated set of retention plays the moment the score crosses a threshold.
The scoring model is intentionally simple — a weighted linear combination, not a black-box ML model. The business team can read it, audit it, and tune the weights when something feels off. That trust matters more than a couple of percentage points of accuracy.
The retention plays are tiered. A score over 60 triggers a personalised email from the CSM. Over 75 triggers a calendar invite for a 30-minute check-in. Over 85 triggers a same-day Slack alert to the head of CS, with the account's history pre-loaded into HubSpot.
The Implementation
The whole pipeline took just under four weeks to ship end-to-end. Week one was data plumbing — getting product events into Snowflake, Stripe events into HubSpot, and ticket data wired up to sentiment scoring. Week two was the scoring engine itself, plus a Looker dashboard so the CS team could see who was at risk and why. Week three was the retention playbooks, written collaboratively with the CSM team so they actually used them. Week four was tuning thresholds against historical churn data.
We did not change a single line of product code. Everything sits in the integration layer, which means it can be unplugged, rewired, or scaled up without product engineering involvement.
The Results
In the first full quarter after launch, the client recovered $180K of ARR that the model had flagged as at-risk and the CS team had then saved through targeted outreach. Monthly churn on the starter tier dropped from 8.1% to 4.3%. The CSM team's "we have no idea what happened" Monday morning meeting was replaced with a 10-minute review of last week's saves.
A few things drove the result more than the model itself. First, the CSM playbooks were written by the CSMs, not us, so they actually felt natural to run. Second, the threshold-based triggers meant nobody had to remember to check the dashboard — the workflow nudged them. Third, every save was logged back to the account record, so the team could see in real time which plays were working and which were not.
What We Would Do Differently
With hindsight, we would have invested another week into the post-save follow-up sequence. Many of the customers we saved in Q1 came back into the at-risk bucket six months later, because the underlying problem — they were not getting enough value from the product fast enough — was never really addressed. A churn save is a stay of execution, not a win. The retention model needs to feed back into onboarding and product feedback loops to compound long-term.
If you are running a SaaS business with anything north of 3% monthly churn on a self-serve tier, this kind of pipeline is almost certainly the highest-ROI automation you can ship this quarter.