Naali Subscription Case Study: 64% to 78% Order 2 Retention on Stay AI

Naali sells supplements in France, as a €5M to €10M brand with more than 10,000 subscribers. Subscriptions run on Stay AI, emails on Klaviyo.

We started in April 2026. The lifecycle flows went live from 22 May. The rebuilt cancel flow shipped on 30 July, with a second version on 6 August. We read the results on 8 September.

Order 2 retention for monthly subscribers went from 64% to 78%.

What was Naali's subscription problem?

Naali signed up subscribers fast and lost them almost as fast. Of every 100 subscriptions started, 68 reached a second order and 38 reached a third.

Monthly churn ran between 15% and 21%. In November and December 2025, more subscribers left than joined.

And there was nothing in place to hold them. Naali had no subscription flows in Klaviyo at the start.

What did the audit find?

The audit found that a quarter of cancels never meant to subscribe, and that the save rate Stay AI showed was mostly people closing the window.

• Accidental subscribers. In February 2026, 20.4% of cancel reasons were "subscribed by mistake" and 4.0% were "just wanted to try". That is a checkout problem. No retention email fixes it.
• Too much product. 11.5% said they had too much stock at home. That is a frequency problem.
• A flattering save rate. Stay AI counts a closed window as a save. The original flow reported 63.8% across 29,512 sessions. Take out the 51.3 points of abandons and the real save rate was 12.5%.
• Discounts that did little. In the old flow, education saved 32 subscribers who said they saw no results, where a discount saved 12. A 20% offer on one product was accepted 0.6% of the time.
• Payment recovery. 38.2% of failed payments were recovered before our work.

Stay AI breaks saves down by reason. "No reason given" saved nobody. "Subscribed by mistake" was the biggest row.

What did RetainUP build for Naali?

We built 13 lifecycle flows in French on Stay AI events, rebuilt the Stay AI cancel flow twice, and tested milestone gifts.

The offer

We tested a free gift on order 3. It shipped on 19 June 2026 and showed no lift: 52.4% of subscribers with the gift reached order 4, against 53.0% without. We now test any gift against a holdout before it goes to everyone.

Stay AI setup and cancel flow

Each cancel reason now gets its own answer. "I don't see results" gets education on how long the product takes to work, with the 90-day guarantee. "Too much stock" gets a delivery delay.

Fewer people closed the window. Abandons fell from 51.3% to 36.5%, so more of them reached a real decision, and more of those took an offer. The method is in cancel flow routing.

Payment recovery

Payment recovery was left for the next phase. Nothing in this build aimed at it, and the recovery rate held flat: 38.2% before, 36.6% after.

Lifecycle flows

Thirteen flows, email and SMS, all in French: order upcoming, onboarding, renewal, payment failed, an SMS reminder before each order, paused, change date, cancel and winback.

From June to August 2026 the flows reached 59,271 recipients and were credited with 422 orders. Before May, they reached nobody, because none existed.

What were Naali's results?

Order 2 retention for monthly subscribers went from 63.9% to 78.2%, and the real save rate from 12.5% to 19.8%.

• Order 2 retention, monthly: 63.9% for September to December 2025 cohorts (4,302 subscriptions). 78.2% for April to July 2026 (4,440). Every cohort counted had lived long enough to be billed a second order.
• Order 2 retention, quarterly: 53.5% (2,909 subscriptions) to 57.7% (5,225).
• Save rate: 12.5% on the original flow across 29,512 sessions. 19.8% on version 2 across 3,329.
• Early cancels: the share of a monthly cohort cancelling between day 7 and day 26 halved, from 7.0% to 3.5%.
• Time to cancel: monthly subscribers who did cancel stayed a median 60 days, up from 33.
• Recovery rate: flat, 38.2% to 36.6%.

On the monthly book alone, the order 2 gain is worth €31,000 to €38,000 a year.

What did not move at Naali?

The charge window, cancels that skip the flow, and the frequency mix all went the wrong way. The next phase aims at them.

• The charge window. Cancels between day 27 and day 34 rose from 29.2% to 40.4% of a monthly cohort. They spike on the charge date and three days before it, when the reminder lands.
• Cancels that skip the flow. Naali's own team cancelled more subscriptions on customers' behalf, from 15% of cancels in March to about 30% by August. 70% of those were "subscribed by mistake" or "did not know".
• The frequency mix. Quarterly went from 39% to 57% of new subscriptions, and quarterly converts to a second order about 20 points worse. So the blended order 2 rate held at 67%.

The fix for the first is the reminder timing. The fix for the second is the checkout, where the accidental subscribers come from. Read subscription cancellation timing for why cancels cluster around the charge.

What can you copy from Naali?

You can copy six moves on any subscription app: a real save rate, reasons before offers, education for "no results", order 2 by frequency, gift holdouts and reminder timing.

1. Take abandons out of your save rate. Stay AI, Recharge, Skio and Loop can all count a closed window as a save. Subtract it before you judge an offer.
2. Read the reasons before you write the offers. "Subscribed by mistake" is a checkout fix, not a cancel flow fix.
3. Answer "no results" with education. At Naali it saved nearly three times what the discount did.
4. Read order 2 retention by frequency. A blended number hid a 14 point gain on monthly.
5. Test gifts with a holdout. The order 3 gift felt right and did nothing.
6. Watch the three days before each charge. If cancels spike there, move the reminder.

FAQ

How much did Naali's order 2 retention improve?

For monthly subscribers, from 63.9% to 78.2% on fully mature cohorts of 4,302 and 4,440 subscriptions. For quarterly, from 53.5% to 57.7%.

Why is Stay AI's save rate higher than the real one?

Stay AI counts an abandoned cancel session as a save. Naali's original flow reported 63.8%, and 51.3 points of that were people closing the window. Offers accepted were 12.5%.

Which cancel offers worked best at Naali?

Education and delivery-delay offers beat every discount. For subscribers who saw no results, education saved 32 where a discount saved 12. A 20% offer on one product was accepted 0.6% of the time.

Did the free gift on order 3 work?

No. 52.4% of subscribers who got it reached order 4, against 53.0% who did not. Test any gift against a holdout first. Our order 2 gift case study shows timing matters more than the gift.

Can RetainUP work on Stay AI in another language?

Yes. Naali's 13 flows and its cancel flow run in French. Our Stay AI retention setup covers the settings.

How do I find where my own subscription program leaks?

Start with the free Subscription LTV Scorecard. It scores your program on 48 checkpoints from public data and lands in your inbox within 24 hours.

About the author

Rahi Jain

Rahi Jain is a subscription LTV specialist who helps $1M to $10M DTC brands reduce subscription churn and maximize LTV on Recharge, Skio, Stay AI and Loop.

More about Rahi