Job cancellation-risk-flags-tested-against-your-past-cancellations · revised 12 October 2026
Flag customers who may cancel, tested on your own past cancellations first
A list of accounts that look like past cancellations, backtested on months you hold back and compared with a simple rule you pick. If it does not beat that rule, you get the rule instead.
You might be seeing
- Cancellations come as a surprise
- A vendor score has no test you can check
- Support contacts and cancellations seem linked but nobody has counted
No passwords, keys, card details or admin invites needed to start.
What usually happened
A churn score looks convincing but is rarely tested against the buyer's own cancellations, and a simple rule such as "three contacts in thirty days" often does as well. Acting on a weak score wastes effort or, worse, treats customers unfairly.
Who it’s for: Support or customer-success lead who wants to see likely cancellations earlier, and has a vendor promising a churn score.
Usually starts when: A vendor offers a "churn score", and you cannot tell whether it would have caught last quarter's cancellations.
The result: You receive a ranked list of accounts that look like your past cancellations, backtested on months you hold back and compared with a simple baseline rule you pick. If the list does not beat the baseline, you get the baseline and a plain statement of that.
Check whether this job fits
Answer these without sending chats, tickets, files or logins. Nothing is submitted unless you choose to contact us.
What you get
- The backtest result against the baseline
- The ranked flag list with reasons
- The baseline rule and how to run it
- A page on what the backtest does not show
Included
- Pseudonymised account-level history of at least 12 months with cancellation dates
- A baseline rule you choose, such as three support contacts in thirty days
- A backtest on the last three months, held back by you until the final run
- A ranked flag list for a person, each with its three strongest reasons from the fields supplied
- A written statement of what the backtest does and does not show
Not included
- Retention offers, discounts or messages triggered by a flag
- Price changes or refusal of service based on a flag
- Any claim that a named person will cancel
- Special-category data, names or contact details
- Voice or emotion detection
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
On the three held-back months (at least 40 cancellations), the report shows, for the top 10% of accounts on our list and on your baseline rule, with the baseline flagging the same number of accounts, how many of the cancellations each caught (counts and percentages). The list passes only if it caught at least 10 percentage points more than the baseline, and the report states that with about 40 cancellations a gap of that size can still be luck. If not, the baseline is delivered and the report says the list did not beat it.
Evidence: The held-back table with both lists and the actual cancellations.
The held-back months were not used to build or tune the list, and the split is shown in the report.
Evidence: The date ranges for the build months and the held-back months.
Every flag lists its three strongest reasons from the fields you supplied, and none uses names, contact details or special-category fields.
Evidence: Sample flags and the field list.
No message, offer or system change is produced by us: the deliverable is a report and a list file only.
Evidence: The delivered files.
Sign-off. Your analyst reads the backtest and the limits page and you sign off in writing. Payment follows sign-off.
If it fails. If the list does not beat your baseline, we say so and deliver the baseline and its numbers. That is a valid result you may sign off or decline; we do not call it a success.
When it fits, and when we stop
It fits when
- At least 12 months of account-level history with a clear cancellation date for each account
- At least 200 past cancellations, with at least 40 in the three months held back
- A person who reads the flags and decides what to do
We stop and tell you if
- Fewer than 200 past cancellations, or fewer than 40 in the held-back months
- You want flags to trigger offers, price changes or refusal of service
- The export holds special-category data we have not agreed to handle
- Nobody agrees what counts as a cancellation
What could go wrong
The analysis reads an export and writes a report, so nothing in your systems changes. Our copy of the export is deleted at hand-over.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| A flag is acted on as if it were a decision about a customer | Flags go to a person only, carry reasons and are never used to refuse service, change a price or send an offer. |
| The test leaks future information into the past | The last three months are held back by you and unseen until the final run. |
| A weak list looks like a good one | The result is compared with your baseline; if it does not beat it, you receive the baseline. |
| Personal data in the export | Account ids are replaced and names, contact details and special-category fields are removed before sharing. |
A separate automated reviewer checks the data split, the backtest and the counts before hand-over, and your analyst on your side marks the sample and decides what goes live. No human supervisor is included unless your agreement names one. At launch the work is largely automated, and we say so.
How we deliver
We arrange the work and independent review, then show you the result against the agreed checks. You keep authority over your systems.
- Agree the cancellation definition, baseline rule and held-back months in writing
- Receive the pseudonymised export; you keep the last three months back
- Build the flag list on the earlier months only
- Run both the list and the baseline on the held-back months
- Independent automated review of the split and the counts
- Hand over the result, the list and the plain statement of limits
This is a one-off job, not emergency cover or a subscription. We confirm eligibility, the total price, a start window and a delivery date before you accept. Work starts only after agreed inputs, secure access and any needed permissions are in place. Model usage for our build and tests is included in the price; charges on your own platform account, including per-conversation fees after go-live, are yours. An enquiry creates no charge or booking.
Need to keep it working?
Discuss a quarterly re-run of the backtest on newer months.
Ongoing work is separately scoped and quoted: no monitoring, response-time guarantee or automatic subscription is included in this job.
Mention the responsibility you need in your enquiry.
What you can check
This is a new service. We have not delivered this job for a client yet.
Other ways to get this done
- Count how many past cancellers had three or more support contacts in their last thirty days. If most did, that rule is your baseline and may be all you need.
Questions
Will it tell me who will cancel?
No. It lists accounts that look like past cancellations, and the backtest shows how well that worked on months you held back.
What if it is no better than a simple rule?
Then you get the simple rule, and the report says so.
Can flags send a discount?
No. Flags go to a person only.
Send an enquiry
Send us
- What you sell and what counts as a cancellation
- Roughly how many accounts and how many cancellations in 12 months
- The simple rule you would use if you had no tool
- No customer data, logins or keys in the first enquiry
Later, once you agree
- A pseudonymised export (account ids replaced; names, contact details and special-category fields removed) of 12 months or more, through a secure route agreed in writing
- The three held-back months, kept by your analyst until the final run
- A person to read the flags
You keep the data and every login. We work from a pseudonymised export you create, with names and contact details removed, and we never contact an account or change a system.
Email fallback: open your mail app
If website submission is unavailable, review and send the fallback email yourself. An email fallback is not a website receipt. Or write to hello@syntheticindustry.ai with “cancellation-risk-flags-tested-against-your-past-cancellations” as the subject.