manufactura.ai
recipe No. 023
Book a call
Recipes/No. 023
Not tested yet Agent Email

Spot subscribers likely to cancel while there is still time

You find out who left from the monthly numbers, long after they decided to go. This agent learns the signals that came before past cancellations. Each week you get a list of subscribers at risk, each with its signal and one action.

01Yield

What one run makes.

A weekly list of at-risk accounts

Every week you get a list of the active accounts that match your cancellation signals. Each one shows the signal that flagged it, the likely reason and one suggested action. A person works the list and approves every message or call.

02Ingredients

What you bring.

  • 6 months of canceled accounts and accounts that stayed
  • Usage per account, such as logins or lessons watched
  • Billing data with plan, renewal date and failed payments
  • Support contact history per account
  • A minimum number of accounts before a signal counts

03Equipment

What the run uses.

  • A usage export from your app
  • An export from your billing system
  • A spreadsheet for the weekly list
  • An email tool with automatic payment recovery emails
  • Your AI chat

04Method

Six steps you can follow yourself.

  1. 1
    Export 6 months of accounts

    Export the last 6 months of canceled accounts and the ones that stayed. Include how often each used the product, whether they made a real start, support contacts and failed payments. Set last month's cancellations aside for a test.

    • Decide the minimum number of accounts a signal must show before you trust it. This keeps out signals that fit only a handful of accounts.
  2. 2
    Find the warning signals

    Ask your AI chat which signals showed up before most cancellations but not in accounts that stayed. Ask for at most 3, each with how many canceled accounts had it.

    • Tell it to leave out any pattern below your minimum. Write down the 2 or 3 signals that pass.
  3. 3
    Test on last month's cancellations

    Score the accounts you set aside against your signals, using only their data from before they left. Check how many the signals would have flagged.

  4. 4
    Score every account each week

    Each week, give your AI chat the current account data and your signals. Ask it to list the accounts at risk, with the signal, the likely reason and one action for each.

    • An action could be a check-in, a fix, an offer to change plan or a call.
    • Tell it to use only the data you give it.
  5. 5
    Send failed payments to recovery

    Accounts at risk only because a payment failed go to your payment recovery emails. Check that those emails are live and enrolling the accounts.

    • A failed payment is a separate problem from an unhappy customer, so keep the two apart.
  6. 6
    Work the list with approval

    A person works the list and approves every message or call before it happens. Contact goes only through channels the customer agreed to.

    • For a check-in, name one specific thing they did, praise something real and ask one question.
    • Keep contact helpful, with no pressure or fake urgency. Keep the cancel option easy to find, and follow your local rules on canceling.
    • If there are too many flags to work, raise the threshold for a flag instead of skipping the list.

05Ways to run it

Follow the method yourself today, or book a call.

Yourself
Free stepsFollow the six steps above with your own tools.
One click
Not built yetAsk for it on a call. What owners ask for gets built and tested next.
On a call
Open nowBook a call and I reply by email to set a time.
Not tested yet. The steps come from my playbook library, rewritten for trading businesses. I have not run this version yet, so this page shows no cost or result.