AI subscription cancellation flow: turn cancel requests into saved subscribers
A step-by-step playbook for an AI subscription cancellation flow: find out why subscribers cancel, match each reason to one save offer like a skip, pause or swap, apply it in Recharge or Skio, and always honor a firm request to cancel.
Subscription brands on Shopify with a steady flow of cancel requests
You need
Yuma connected to Shopify, your helpdesk and your subscription app, plus Claude connected through the Yuma MCP
Main metric
Share of saved subscribers still active after 60 days
Time to first results
Steps 2 to 7 fit inside a 30-day free trial
What problem does this solve?
Subscription brands lose a large share of subscribers through support: a customer writes in to cancel, waits in a queue, and gets either a generic retention script or a cancellation with no attempt to understand why. Many of those customers only needed a skip, a pause, a different frequency or a product swap. Agents spend hours on routine subscription changes, and the reasons people leave stay buried in tickets nobody analyzes.
How does it work?
Connect Yuma to Shopify and your subscription app (Recharge, Skio, Loop and others) so the agent can see each subscription and take actions like skip, pause, reschedule, change frequency, swap products, apply a discount or cancel.
Connect Claude to Yuma through the Yuma MCP and run the subscription save prompts on the last 90 days of cancellation requests. Group them by reason and count how many could have been solved with a change instead of a cancel.
Build a save-offer map with one offer per reason: too much product gets a skip or a longer frequency, price gets a one-time discount, travel gets a pause, taste or fit gets a swap, a move gets an address update.
Write a cancellation playbook in the Yuma Handbook. Ask for the reason if the customer hasn't given one, make the single most relevant offer, and apply it instantly in the subscription app. If the customer declines or repeats the request, cancel right away and confirm it.
Add tests before you publish, including a customer who firmly says "cancel now" (the agent must cancel without pushing back), an upset customer, and a refund request on a recent charge.
Roll out on part of your volume first and read transcripts daily during the first week.
Measure every week with the reporting prompt: save rate by reason, which offers get accepted, and how many saved subscribers are still active 30 and 60 days later.
Adjust the offer map based on what actually retains subscribers, update the Handbook and rerun the tests.
What's the biggest win?
Cancel requests get an answer and an action in the same conversation, at any hour, instead of a queue and a generic script. Customers who only needed a skip, pause or swap stay subscribed, agents stop spending their day on routine subscription changes, and the brand finally gets clean data on why subscribers leave, by reason and by product.
What's required to run this?
Make one offer per conversation and match it to the stated reason. Keep discounts one-time and capped so saves don't quietly erode margin. Track whether saved subscribers are still active after 30 and 60 days, not just whether they accepted the offer in the moment. Keep the reason categories stable so you can compare month to month. Run steps 2 to 7 on your own tickets with a 30-day free Yuma trial.
What are the constraints?
Clear cancellation requests must always be honored, and the flow has to follow the auto-renewal and consumer protection rules in each market you sell in. Available actions depend on the subscription app; each platform supports a different set of changes. A save that cancels again next month inflates the save rate, which is why the 30 and 60-day check matters. Heavy discounting can retain subscribers at a loss.
See Yuma in action
Want Yuma to run this playbook for you?
Turn this into a live playbook in your Handbook, tested before customers see it.
A save-offer map pairs each cancellation reason with a single offer. This is a starting point built from the reasons subscription brands see most often. Replace it with the reasons and actions you find with Prompt 1 and Prompt 2 below.
Cancellation reason
Save offer
Skip the offer when
Too much product piling up
Skip the next order or move to a longer frequency
The customer has already skipped several times in a row
Price
One-time discount on the next order, capped
The customer has already used a save discount
Travel or a temporary break
Pause for a set number of weeks
The customer gives a firm end date that is far away
Taste, scent or fit
Swap to another product or variant
The complaint is about quality, not preference
Moving
Update the shipping address
The new address is outside your shipping zones
Quality issue or damaged order
No offer: fix the problem and cancel if asked
Always route to a human if the order needs a refund review
No reason given
Ask once for the reason, then cancel if there is no answer
The customer has already said they want to cancel now
The prompts for this playbook
Use these with Claude connected to Yuma through the Yuma MCP. Replace everything in [brackets]. Claude proposes; a human reviews and publishes in Yuma. Save them in a Claude project or skill so anyone on the team can rerun them.
Prompt 1: why subscribers cancel
Search every Yuma ticket from the last [90] days where a customer asks to cancel a subscription.
Return:
1. Total cancellation requests and distinct customers.
2. Reasons grouped into stable categories (for example: too much product, price, travel or temporary break, taste or fit, quality issue, moving, shipping problems, no reason given), with count and share for each.
3. For each reason, how many could likely have been solved with a skip, pause, frequency change, swap, discount or address update instead of a cancel.
4. The top products or SKUs mentioned in cancellations.
5. Two ticket IDs and one short customer quote per reason.
Prompt 2: build the save-offer map
Using the reasons from Prompt 1 and our subscription app's available actions ([list them, e.g. skip, pause, reschedule, change frequency, swap, discount, cancel]), propose one save offer per reason.
For each reason give: the offer, the exact wording the agent should use in our brand voice, and when NOT to make an offer (for example quality complaints, or a customer who is clearly upset). Keep discounts one-time and no higher than [X%].
Prompt 3: write tests for the cancellation playbook
Write [12] tests for our cancellation playbook, using real tickets where possible. For each test give the customer message (personal data removed), the expected reply, the expected action in the subscription app, and why the case matters.
Include at least:
- a customer who firmly says "cancel now" (expected: cancel immediately, no offer pushback)
- a customer who declines the first offer (expected: cancel and confirm)
- an upset customer with a quality complaint
- a refund request on a charge that just went through
- a customer with more than one active subscription
Prompt 4: weekly save report
For the last 7 days of cancellation requests in Yuma, report:
- total requests, saved subscribers and save rate
- save rate by reason and by offer
- offers that were declined most often
- 3 example conversations: one clean save, one respectful cancel, one that went badly
Keep it under 200 words.
Prompt 5: did the saves stick?
Find customers who accepted a save offer [30 / 60] days ago. Tell me how many have since cancelled, written in again to cancel, or are still active, broken down by reason and offer. Flag any offer whose saves mostly cancel within 60 days.
Tests to run before you publish
Prompt 3 writes these for you from real tickets. These five cases are the minimum. If any of them fails, fix the Handbook and rerun before going live.
Customer says
Expected action
Why it matters
"Cancel my subscription now, please."
Cancel immediately and confirm. No offer.
A clear request must always be honored
"I have too much coffee left."
Offer one skip or a longer frequency, apply it if accepted
The most common save, and the easiest to get right
"No thanks, just cancel."
Cancel and confirm after the declined offer
One offer per conversation, never a sequence
"The last bag arrived damaged. Cancel everything."
No offer. Handle the damage, cancel, and route a refund review to a human
Retention offers on a quality complaint read as tone-deaf
"I have two subscriptions and want to stop one."
Confirm which subscription, then act only on that one
Avoids cancelling the wrong plan
Metrics to track
Metric
How to measure it
Why it matters
Save rate by reason
Saved subscribers divided by cancel requests, split by reason (Prompt 4)
Shows which offers fit which reasons
60-day retention of saves
Share of saved subscribers still active 60 days later (Prompt 5)
The real test: a save that cancels next month is not a save
Offer acceptance by offer
Accepted offers divided by offers made
Spots offers customers ignore
Discount cost per retained subscriber
Total save discounts divided by subscribers still active at 60 days
Keeps saves from quietly eroding margin
Time to resolution
Time from first message to the subscription change
Cancel requests answered at any hour, without a queue
Customer example: Javvy Coffee
At Javvy Coffee, a global consumer subscription brand, subscription cancellations and subscription questions were the largest share of support requests. Yuma now handles 70% of all customer queries, with subscription cancellations and subscription inquiries among the top automated requests.
A subscription cancellation flow is the set of steps a customer goes through when they ask to cancel. A good one finds out why they want to leave, offers one relevant alternative such as a skip, pause or product swap, and cancels straight away if the customer still wants to go.
How does AI reduce subscription churn?
An AI agent answers every cancel request at once, at any hour, reads the subscription in your subscription app and makes the offer that matches the stated reason. Customers who only needed a skip or a pause stay subscribed, and every conversation is tagged with a reason, so you can see why subscribers leave.
Which subscription apps does this playbook work with?
Yuma connects to Recharge, Skio, Loop Subscriptions and other subscription apps on Shopify. The actions available, such as skip, pause, swap or discount, depend on what each app supports.
Will the AI agent make it hard for customers to cancel?
No. The playbook makes one offer per conversation and cancels right away if the customer declines or repeats the request. Your tests should include a customer who says "cancel now", and the expected result is an immediate cancellation. Follow the auto-renewal and consumer protection rules in each market you sell in.
How do I know if a save offer worked?
Look at how many saved subscribers are still active 30 and 60 days later, not only at whether they accepted the offer. Prompt 5 in this playbook gives you that breakdown by reason and by offer.
