How to Predict Churn in Usage-Based SaaS.
The three signals that show usage churn before revenue falls, how to measure them from Stripe Meters, and what to do next.
To predict churn in usage-based SaaS, compare each customer's usage with their own baseline, watch whether the average of recent invoices is shrinking, flag accounts with no usage for a set window, and rank them by revenue at risk. Act while they still use the product: a check-in naming what they stopped using, a one-off usage credit, or a committed-spend plan.
In usage-based SaaS, most churn doesn't start with a cancel. Usage fades, invoices shrink, and the subscription can stay active with nothing billed. To predict it, compare each customer's usage with their own baseline, watch invoice size over several periods, flag accounts that go idle, and rank the result by revenue at risk. Then act while the customer still uses the product.
Why usage-based churn is hard to see
On a flat plan, a customer who leaves cancels, and the cancel is a clear event. On a usage plan, a customer who leaves usually just stops using the product. Nothing fails and nothing is canceled, so tools that wait for a cancel or a failed payment never fire.
Stripe's own MRR doesn't help much here either. Its Billing analytics MRR leaves out metered (usage-based) products entirely, and usage-based MRR is a preview feature you have to request (Stripe docs, checked Nov 14, 2026). For a usage-only business, the MRR chart can miss most of the revenue.
The three signals to watch
| Signal | What it looks like | How to measure it |
|---|---|---|
| Usage fades | Calls, tokens or jobs fall week by week, with no cancel and no failed card | Each customer's recent usage against their own earlier baseline, not one threshold for everyone |
| Invoices shrink | Each invoice is a little smaller while the total still looks fine | The average of the last 3 invoices against the 3 before |
| Accounts go idle | The plan stays active and nothing is billed | No usage for a set window, such as 28 days |
How to predict churn, step by step
1. Get usage per customer
If you bill with Stripe's Billing Meters, each meter event already carries a customer, a value and a timestamp. Stripe's Meter Event Summary returns a customer's total usage for a time period, and the Meter Usage Analytics API, in public preview, returns usage by day or other windows for up to 5 meters per request (Stripe docs, checked Nov 14, 2026). If you bill another way, product analytics such as PostHog gives the same picture from events.
Note that Stripe processes meter events asynchronously, so recent usage can take a while to show in summaries. That is fine for spotting a trend over weeks.
2. Set a baseline for each customer
A team of three and a team of three hundred use the product at very different levels. Compare each customer with their own recent history, for example the last few weeks against the weeks before. A fixed threshold flags small customers forever and misses large ones that are falling fast.
3. Combine the signals
One quiet week is noise. A fall against baseline that lines up with shrinking invoices is a pattern. Score each customer on how many signals fire and how far usage has fallen, and keep the reasons next to the score so whoever acts knows what changed.
4. Rank by revenue at risk
A small customer at high risk can matter less than a large customer at moderate risk. Multiply each customer's recent revenue by their risk, sort by the result, and start at the top.
5. Act while they still use the product
The moves that fit usage plans are different from flat plans:
- A check-in that names what they stopped using. One short note in your name, with one question and no offer.
- A one-off usage credit for valuable customers whose usage dropped, to bring them back to the product.
- A committed-spend plan when usage is steady but the customer complains about price: a lower unit price for a monthly minimum.
Skip the pause offer. A usage-only customer who stops using the product already pays little or nothing, so a pause gives them nothing new.
What Stripe does and doesn't do
Stripe's usage alerts notify you when a customer goes over a usage threshold, with up to 25 alerts for each meter and customer pair. They are built for limits and upsells; we found no alert for falling usage in the docs. Stripe's Usage pages in Billing analytics show aggregate usage and usage revenue across meters that sum or count (Stripe docs, checked Nov 14, 2026).
Stripe now recommends Metronome, its usage billing platform, for new usage-based integrations. Its docs note that basic Billing Meters reconcile usage at invoice time, while Metronome tracks usage in real time (Stripe docs, checked Nov 14, 2026).
See money at risk in your own Stripe, and the first five customers to act on.
Where North Metric fits
North Metric reads usage from Stripe Meters for usage and hybrid plans, and from PostHog for flat and seat plans. It flags fading usage against each customer's own baseline, uses a sharp fall in the last-3-invoice average as a risk reason, and counts no usage in 28 days toward an idle customer rate. Every customer gets a 0–100 risk score, the list is ranked by money at risk, and each at-risk customer gets one drafted move to approve. Usage billed through Metronome isn't read yet; PostHog covers it. See churn prevention for usage-based SaaS.
Sources
- Stripe docs: Billing analytics, checked Nov 14, 2026.
- Stripe docs: How usage-based billing works, checked Nov 14, 2026.
- Stripe docs: Analyze and query meter usage, checked Nov 14, 2026.
- Stripe docs: Monitor usage with alerts, checked Nov 14, 2026.
Frequently asked questions
How do you predict churn for usage-based SaaS?
Compare each customer's recent usage with their own baseline, check whether recent invoices are shrinking, and flag accounts with no usage for a set window. Score customers on those signals and rank them by revenue at risk.
Is MRR meaningful for usage-based SaaS?
Only with care. Stripe's Billing analytics MRR excludes metered products, and its usage-based MRR is a preview feature (checked Nov 14, 2026). A steadier measure averages each customer's last few usage invoices.
Can Stripe alert me when a customer's usage drops?
Stripe's usage alerts fire when a customer goes over a threshold, for limits and upsells. We found no alert for falling usage in Stripe's docs, checked Nov 14, 2026, so a drop needs its own tracking.
When does a usage-based customer count as churned?
A common rule is usage at zero for whole billing periods while the subscription stays active. Before that point the customer is fading, and a check-in can still work.