A8gent
All articles

Industry

Gym Failed Payment Recovery: What Actually Gets the Money Back

Most failed gym memberships are not people refusing to pay. They are expired cards, replaced cards and empty accounts on the wrong day. That distinction decides your entire build, because a large share of failures cannot be retried at all - the processor will keep scheduling attempts that never execute until a human updates a card.

A8gentAutomation practiceAugust 5, 202613 min read
Contents
  1. Two Different Churns, One Report
  2. Why Retrying Harder Does Not Work
  3. What the Retry Engine Does Do, and How to Configure It
  4. The Member-Facing Sequence
  5. The End State Nobody Decides
  6. A Worked Example, With the Arithmetic Shown
  7. Building It, and What to Watch
  8. FAQ

Two Different Churns, One Report

Almost every gym reports churn as a single number, and that number hides two populations who need opposite treatment.

Voluntary churn is the member who decided to leave. They stopped coming in November, they thought about it over Christmas, and in January they cancelled. Recovering them is a retention and product problem, and it is largely won or lost months before the cancellation.

Involuntary churn is the member who did not decide anything. Their card expired. It was replaced after a fraud alert. The bank declined a recurring charge it did not recognise. Their account was empty on the 1st and had money in it on the 3rd. They still want the membership, they may not even know there is a problem, and in many cases the first they hear of it is when their access card stops working at the door - which is a spectacularly bad way to find out.

The second population is the one this article is about, and it has an unusual property: the member is not resisting you. There is no persuasion to do. There is an administrative task that needs to happen and a person who has not been told about it clearly enough. That makes it one of the highest-yield automations available to a fitness business, and also one of the easiest to build badly - because the obvious build, "retry the card and send some reminders," misunderstands what is actually failing.

Separate the two in your reporting before you build anything. If you cannot currently tell how many of last month's lost memberships were payment failures rather than cancellations, that is your first task, and it takes an afternoon. For where this sits among a studio's other automation candidates, see AI automation for gyms and fitness studios and our broader piece on AI agents for fitness studios and gyms.

Sourcing note. The payment mechanics below are quoted from Stripe's own published documentation, read in August 2026, with the URL given inline. We use Stripe because it documents this behaviour openly and precisely; if you use a different processor or a gym management platform with embedded billing, the principles transfer but the specific codes, limits and end states will not - find your provider's equivalent documentation before designing to these numbers.

Why Retrying Harder Does Not Work

Here is the fact that should restructure your entire build, quoted from Stripe's revenue recovery documentation.

Stripe lists nine hard decline codes for which a payment cannot be automatically retried:

  • incorrect_number
  • lost_card
  • pickup_card
  • stolen_card
  • revocation_of_authorization
  • revocation_of_all_authorizations
  • authentication_required
  • highest_risk_level
  • transaction_not_allowed

And it states plainly what happens: "For these failures, the scheduled retries continue but the payment only executes if you obtain a new payment method." Elsewhere: "Retries continue to be scheduled, and attempt_count continues to increment, but retries only execute after detecting a new payment method. Unexecuted retries don't create a new Charge."

Sit with that for a moment, because it is the whole article. On a hard decline, your dunning schedule is theatre. The counter increments, the dashboard shows attempts, days pass, and not one of those attempts is a real charge. The only event that can ever recover that membership is a human being entering new card details. Every hour your automation spends optimising retry timing on those failures is an hour spent on nothing.

Note in particular lost_card and stolen_card. A member whose card was compromised and replaced is the archetypal involuntary churn case - a loyal member, no intention to leave, a card number that no longer exists. No retry schedule in the world will fix it.

Note also authentication_required. This is a different beast: the issuer wants the cardholder to authenticate, which means a customer action is required rather than a new card. Treating it identically to lost_card in your messaging will confuse the member and lose the recovery.

The practical consequence is a split at the top of your workflow. Read the decline code, and route on it. Soft failures - most commonly insufficient funds - are genuinely a timing problem and belong to the retry engine. Hard failures are an "update your card" problem and belong to your messaging, immediately, with no waiting for a retry schedule that will never fire. Building one undifferentiated sequence for both is the single most common mistake in this workflow, and it costs you the most recoverable members.

Stripe also documents that it will not retry at all when "No payment methods are available," when the card is India-issued, or when a Connect account has been disconnected - worth knowing if any of those apply to you.

What the Retry Engine Does Do, and How to Configure It

For soft declines, the processor's own retry logic is usually better than anything you would build, and you should let it work.

Stripe describes Smart Retries as using an AI model with "time-dependent, dynamic signals," giving examples including "the number of different devices that have presented a given payment method in the last N hours" and "the best time to pay (payments made for debit cards in certain countries might be slightly more successful at 12:01 AM in local time zones)."

The configuration surface is documented and specific:

  • Smart Retries can be set to a number of retries within a period of 1 week, 2 weeks, 3 weeks, 1 month or 2 months. Stripe states: "The recommended default setting is 8 tries within 2 weeks."
  • A custom retry schedule allows "up to three retries, each with a specific number of days after the previous attempt." So switching off Smart Retries to hand-roll a schedule costs you five of the eight attempts. Have a reason.
  • Retries use the first available payment method in a documented priority order: subscription default payment method, then subscription default source, then customer default payment method, then legacy customer default source.

That ordering carries a trap worth spelling out, because Stripe flags it directly: "if a subscription has a default_payment_method, but you only update customer.invoice_settings.default_payment_method, Stripe continues to retry on the subscription's default_payment_method." In plain terms - a member can successfully add a new card and still fail, because the new card went into the wrong field. If you build your own update page rather than using a hosted one, test this specific path deliberately. It fails silently and it looks exactly like a member who did nothing.

ACH is a different clock entirely. Many gyms run bank debits to avoid card fees on low-margin memberships. Stripe's table of local payment method retries lists ACH Direct Debit with retryable failures limited to insufficient funds, a maximum of 2 retries, and a maximum retry period of 40 days, subject to ACH Direct Debit mandate requirements - and notes that by default Stripe does not automatically retry local payment methods at all unless you enable it. Bacs, SEPA and AU BECS are listed at 2 retries over 30 days; ACSS and NZ BECS at 1 retry over 30 days. If your members pay by bank debit, a card-shaped dunning sequence is wrong on both cadence and content.

Hook your automation to the documented events rather than polling: the invoice.payment_failed webhook carries attempt_count, and next_payment_attempt on the invoice indicates when the next collection will happen - though Stripe notes that for automations users, next_payment_attempt moves from invoice.payment_failed to invoice.updated. See webhook and polling vs webhook.

The Member-Facing Sequence

Given the above, the messaging has one job for hard declines and a different one for soft.

For hard declines: immediate, specific, and one tap to fix. Do not wait for a retry cycle that cannot fire. The message should say what happened in human terms ("your card was declined - it may have been replaced or expired"), what it means ("your membership is still active for now"), and give a single link to a hosted card-update page. Not a login. Not "call us." A link that opens a payment form on a phone.

The friction of that update path is the entire conversion lever. Every additional step - remembering a password, finding the right menu, downloading an app - loses members who fully intended to pay you. If you measure one thing about this build, measure the completion rate of the update page, and treat a low number as a design problem rather than a member problem.

For authentication_required: different copy. The card is fine; the bank wants the member to confirm. Saying "your card was declined, please update it" sends them to replace a perfectly good card and often fails again. Say the bank needs them to confirm the payment and route them accordingly.

For soft declines: acknowledge, then wait. A single, gentle, non-alarming notice, then let the retry engine work. Messaging a member every day about an insufficient-funds decline that will clear itself on payday is how you convert an involuntary lapse into a voluntary cancellation. This is the case where less automation performs better, and it is worth resisting the vendor instinct to add touches.

Channel and timing. Email for the detail, SMS for the urgent nudge, and the app or door system if you have one. If you send SMS over a US 10-digit long code you are in A2P 10DLC: Twilio's documentation describes it as "the standard that United States telecom carriers have put in place to ensure that SMS traffic to US end users through long code phone numbers is verified and consensual," requires a registered Brand and Campaign, and notes unregistered senders "will receive additional carrier fees for sending unregistered traffic," with daily throughput to T-Mobile from 1,000 segments for a Sole Proprietor brand up to 2,000 for Low-Volume Standard. See our A2P 10DLC guide, the glossary entry and opt-in.

The door is a message too. A member who is turned away at the turnstile learns about their billing problem in the most humiliating way available, in front of other people, at the moment they were about to do something good for themselves. If your access control can be told to let a lapsed member in while showing a polite prompt at the desk, that is worth more than three emails. Decide this deliberately - it is a business policy question, not a technical one, and it should match whichever end state you choose below.

The End State Nobody Decides

When retries are exhausted, something has to happen, and Stripe documents exactly three options for a subscription:

  • Cancel the subscription - "The subscription changes to a canceled state after the maximum number of days defined in the retry schedule."
  • Mark the subscription as unpaid - "The subscription changes to an unpaid state... Invoices continue to be generated and stay in a draft state."
  • Leave the subscription past-due - "The subscription remains in a past_due state... Invoices continue to be generated and charge the customer based on retry settings."

Stripe adds: "After the final payment attempt, we make no further payment attempts. Changing your subscription settings only affects future retries." That last sentence matters operationally - fixing your configuration does not rescue the members already past the end of the schedule.

Each option is a different business posture and gyms rarely choose consciously:

Cancel is clean and final. The member is gone, the relationship is reset, and if they come back it is a new signup - which may mean a new joining fee, which may be the thing that stops them coming back. Good for hygiene, bad for a member who simply lost a card.

Unpaid keeps the record alive without generating charges, which suits a business that wants to keep the door open to a returning member without accumulating a debt they will resent.

Past-due keeps invoicing. Be careful here: a member who returns after four months to a balance covering four months of a membership they did not use has a grievance, and it is a reasonable one. Whether that is right depends entirely on your contract terms and how you handle it in practice.

Whichever you choose, the access control policy must match the billing state, and in most gyms these are two different systems maintained by two different people. A member who is canceled in billing and still has a working key fob is a revenue leak; a member who is past_due and locked out is a complaint and a cancellation. Write the mapping down and test it.

The other decision worth making explicitly is when a human gets involved. Our position is that after the automated sequence has run and failed, a long-standing member is worth a phone call from a person - not a fourth email. The value of a recovered annual membership comfortably exceeds a few minutes of staff time, and the conversation frequently surfaces something the automation could never see. Build the handoff as a task with the context attached, and see human-in-the-loop.

A Worked Example, With the Arithmetic Shown

Illustrative model. We report no client results and you should replace every input with figures from your own billing reports, all of which are readily available.

Baseline. A studio has 620 members at an average of $89/month. Monthly billed revenue: 620 x $89 = $55,180.

Suppose the monthly payment failure rate is 6 percent: 620 x 0.06 = 37.2 failed charges a month. This is a number your processor reports directly - use the real one, because failure rates vary enormously by member mix and payment method.

Split by decline type, which is the step this article exists to argue for. Suppose the split is 65 percent soft (mostly insufficient funds) and 35 percent hard:

  • Soft: 37.2 x 0.65 = 24.2 failures. These are largely recovered by the retry engine whether or not you build anything. Suppose the processor's retries recover 80 percent unaided: 19.4 recovered, 4.8 remaining.
  • Hard: 37.2 x 0.35 = 13.0 failures. The retry engine recovers none of these - every one requires a new payment method. Without a working update path, all 13 are lost.

So the addressable population is 13.0 + 4.8 = 17.8 memberships a month, and roughly three-quarters of it is the hard-decline group that no amount of retry tuning touches.

The value of a recovery is not one month. This is where most calculations understate the case badly. A recovered member does not pay $89 once; they continue. If your average remaining member lifetime at that point is, say, 9 months, each recovery is worth 9 x $89 = $801 in future billings. Use your own retention curve rather than this figure, and use contribution rather than revenue if your marginal cost per member is material.

Break-even. Suppose the build costs $3,000 once plus $50/month - your quotes; automation build cost covers the drivers. First-year cost: $3,000 + $600 = $3,600. At $801 of lifetime value per recovery, break-even is $3,600 / $801 = 4.5 recovered memberships across the entire year.

Against an addressable pool of 17.8 a month - 214 a year - that is a required recovery rate of 4.5 / 214 = 2.1 percent. Which is to say: if a clearly worded message with a one-tap card update page recovers even one member in fifty, the build has paid for itself.

Two honest caveats. First, if your gym management platform already does this, you are paying to rebuild an included feature - check before you scope, and check specifically whether it distinguishes hard from soft declines, because many do not. Second, the lifetime-value multiplier is doing a lot of work in that calculation and it is the input most often inflated. Use a measured retention figure, not an aspirational one. Our ROI of AI agents piece covers this class of error and payback period the framing; for general budget ranges see AI automation cost.

Building It, and What to Watch

Build order. First, split your reporting so you can see failed payments separately from cancellations, and see hard separately from soft. Second, make the card-update path as short as it can possibly be and measure its completion rate. Third, wire the hard-decline branch to fire immediately on the failure event. Fourth, add the soft-decline branch with deliberately minimal messaging. Fifth, define the end state and align access control to it. Sixth, add the human handoff for high-value members. Everything before step three is worth doing even if you never automate anything.

Reliability details that decide whether it survives. Webhook handlers must be idempotent - payment events can be redelivered and a duplicate "your card failed" message to a member who has already fixed it is the most damaging error in this workflow. Key on invoice ID plus attempt count. See idempotency and retry and backoff.

Always re-check state immediately before sending. Between the failure event and your message going out, the member may have paid. Check, then send. This one check prevents the majority of embarrassing sends.

Alert on silence. If your webhook endpoint breaks, no failure events arrive, and a dashboard showing zero failed payments looks like excellent news. It is the same hazard as any exception-driven automation: alert when the volume of events drops to zero, not only when errors occur. See observability and error handling.

What to measure. Recovery rate split by decline type - these are two different systems and one blended number tells you nothing. Completion rate of the update page. Time from failure to recovery. And the count of members who reach the end state, which is your true involuntary churn and the number the whole project exists to reduce. Do not report emails sent.

What not to do. Do not offer a discount to recover an involuntary failure - the member was not price-sensitive, they had a card problem, and you have just taught them that letting a payment fail reduces their bill. Do not send more than the sequence justifies. And do not let the automation cancel memberships silently: whatever end state you choose, a human should be able to see the list of members about to hit it, because that list is also your most valuable outbound call list of the month.

If you want a second opinion on whether this is your highest-value first build, the free automation audit works through the alternatives, and automation services explains how we scope it. The sibling problem - the member who stopped showing up before they stopped paying - is covered in attendance drop-off automation.

FAQ

Why do retries fail to recover so many gym memberships?
Because a large share of failures are hard declines that cannot be retried at all. Stripe lists nine such codes - incorrect_number, lost_card, pickup_card, stolen_card, revocation_of_authorization, revocation_of_all_authorizations, authentication_required, highest_risk_level and transaction_not_allowed - and states that for these "the scheduled retries continue but the payment only executes if you obtain a new payment method." The attempt counter increments while no real charge ever happens. For those members, the only thing that recovers the money is a new card being entered.
How many times will Stripe retry a failed subscription payment?
With Smart Retries you set a number of attempts within a window of 1 week, 2 weeks, 3 weeks, 1 month or 2 months, and Stripe states "The recommended default setting is 8 tries within 2 weeks." If you switch to a custom schedule you can configure at most three retries, each a set number of days after the previous attempt - so hand-rolling a schedule costs you five attempts, and you should have a specific reason for doing it.
Is ACH different from card for gym billing?
Substantially. Stripe's local payment method table lists ACH Direct Debit as retryable only for insufficient funds, with a maximum of 2 retries over a maximum retry period of 40 days, subject to ACH Direct Debit mandate requirements - and notes that automatic retries for local payment methods are off by default and must be enabled. Bacs, SEPA and AU BECS are listed at 2 retries over 30 days; ACSS and NZ BECS at 1 over 30 days. If your members pay by bank debit, a card-shaped sequence is wrong on both timing and wording.
What should happen when retries run out?
Stripe documents three end states: cancel the subscription, mark it unpaid (invoices continue to generate but stay in draft), or leave it past-due (invoices continue and charge per retry settings). Choose deliberately, because each is a different posture toward a member who lost a card rather than one who left. Then make sure your access control matches the billing state - a cancelled member with a working key fob is a leak, and a past-due member locked out at the door is a complaint that becomes a cancellation.
Should I message a member on the day the payment fails?
For a hard decline, yes - immediately, because no retry can ever succeed and every day of waiting is wasted. For a soft decline such as insufficient funds, no. A single gentle notice and then silence while the retry engine works performs better, because messaging someone daily about a shortfall that will clear on payday is an efficient way to convert an involuntary lapse into a deliberate cancellation.
Why did a member's new card still fail?
Very often because it was saved to the wrong field. Stripe retries using a documented priority order and warns that "if a subscription has a default_payment_method, but you only update customer.invoice_settings.default_payment_method, Stripe continues to retry on the subscription's default_payment_method." If you built your own update page rather than using a hosted one, test this path explicitly. It fails silently and looks identical to a member who never updated anything.
Should I offer a discount to recover a failed payment?
No. Involuntary churn is not a price objection - the member had a card problem, not a value problem. Offering money off teaches your membership base that letting a payment fail reduces the bill, which is an expensive lesson to teach. Spend the effort on making the card-update path shorter instead; the completion rate of that page is the real lever and it costs nothing per member.
What if my gym management software already handles dunning?
Then check what it actually does before building anything, and check one thing specifically: whether it distinguishes hard declines from soft ones. Many platforms run a single undifferentiated reminder sequence against both, which wastes the retry window on members who need a new card and over-messages members who simply need payday to arrive. If yours does that, the highest-value change may be a configuration and copy change rather than a new build.

If this resonated

Module 02 is the long version of this post.

Eval set, refusal pattern, kill switch. Build it in a weekend.

All postsLast verified 2026-08-05

Done for you

Want this built for your business?

Tell us the workflow. We scope, build, and ship the agent with guardrails and the numbers to prove it worked. The scoping call is free.

Free scoping call. You own the code.

Request a scoping call