Job metric-change-alert-with-segment-breakdown · revised 12 October 2026
Tell me when orders or refunds suddenly change, and which product or channel moved
For up to five daily metrics, a script you run writes an alert when a value leaves its usual seasonal range, naming the products or channels behind it. Figures recompute; no cause is claimed.
You might be seeing
- A drop in orders or a spike in refunds was found days late
- Dashboards show numbers but nobody is told when one leaves its usual range
No passwords, keys, card details or admin invites needed to start.
What usually happened
A plain threshold alarm either fires every weekend or never. A seasonal range needs history, and an alert that only says 'orders down' sends someone digging. A language model adds nothing to detecting the change: at most it helps draft the alert wording.
Who it’s for: E-commerce or operations lead with a daily export who learned of a drop in orders days late.
Usually starts when: Orders fell for four days before anyone noticed, or a refund spike was found at month-end.
The result: For up to five daily metrics, the script writes an alert message each time it finds a value outside a stated seasonal range, naming the segments that account for most of the change, and you send it on by email or chat because we connect to nothing. Every figure in an alert recomputes from your file, and an alert never states a cause. A missing or stale file raises a 'no data' alert instead of silence.
Check whether this job fits
Answer these without sending files or logins. Nothing is submitted unless you choose to contact us.
What you get
- The script and settings
- The backtest report with every false alert listed
- Sample alerts with their recomputation
- A 'no data' alert
- Run steps
Included
- Up to five daily metrics with at least two seasonal cycles of history before the 12-month backtest (about three years)
- A seasonal range for each metric, worked out by a statistical method
- A breakdown by up to three dimensions, such as product, channel and region
- A backtest over the last 12 months against anomalies you identify and hold back
- A script and settings you run yourself, and the alert wording
Not included
- Root causes or explanations
- Running or hosting the monitor for you
- Access to your shop, database or analytics account
- Forecasts
- Personal data
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
On the last 12 months of your history, with each day's range fitted only on earlier days, at least 4 of your 5 hidden anomalies are found, with no more than 2 false alerts in any month; a false alert is an alert on a day you mark as normal after seeing the list. This is a small test: 4 of 5 shows only that the true rate is at least 34% with 95% confidence.
Evidence: The backtest report beside your hidden list.
Every figure in a sample of 10 alerts recomputes from your file.
Evidence: The 10 alerts and your recomputations.
No alert states a cause. A script search for causal words finds none.
Evidence: The search output.
3 missing days, which you remove from the file and we do not see beforehand, produce 3 'no data' alerts and no 'no anomaly' message.
Evidence: The planted days and the alerts.
The alert wording is a fixed template filled from the figures, and two runs on the same file give identical alerts.
Evidence: Both outputs compared.
Sign-off. You inspect the backtest, the false alerts and the sample alerts, and sign off in writing. Payment follows sign-off. You run the monitor yourself.
If it fails. If the held-back test is not passed, you do not pay for this fixed scope. You keep the results, the failures and your hidden anomaly list, and we agree whether to stop or re-quote. No surprise work.
When it fits, and when we stop
It fits when
- At least two seasonal cycles of daily history for each metric before the 12-month backtest, about three years in all
- A daily export you generate yourself, with no personal data
- You pick 5 past anomalies in the last 12 months and keep them hidden from us until the test
- A named person receives and acts on the alerts
We stop and tell you if
- History is shorter than two seasonal cycles plus the 12-month backtest
- You want root causes
- The export is not daily, or changes shape without notice
- You cannot identify any past anomaly to test against
What could go wrong
The job reads copies you send and writes files. It changes nothing in your systems, so there is nothing to revert. Our copies of your files are deleted when the job ends.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| Too many alerts, so people ignore them. | The backtest lists every false alert, with a cap of 2 in any month, and the cap is agreed in writing. |
| An alert hides a cause that is guessed. | The wording check fails any alert that contains a cause. Only where and how much are stated. |
| A missing file is read as 'no anomaly'. | A stale or missing input raises a 'no data' alert, tested with 3 missing days you remove. |
A separate automated reviewer checks the seasonal ranges, the recomputation and the no-data case before hand-over and does not see the hidden anomalies. A named person on your side picks the 5 hidden anomalies, checks the false alerts and signs off. No person on our side is included unless your agreement names one, and at launch the work is largely automated.
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 metrics, the dimensions, the false-alert cap and the severe errors
- Receive the history; you keep 5 anomalies hidden from us
- Build the seasonal ranges, the breakdown and the alert wording
- Backtest on the last 12 months and compare with your hidden list
- Independent review of the recomputation and the no-data case
- Hand over the script and settings; delete our copies
This is a one-off job, not emergency cover or a subscription. We confirm eligibility, the price, a start window and a delivery date before you accept, and work starts only after inputs, a secure handoff route and permissions are in place. Model usage up to a cap agreed in writing is included, and an enquiry creates no charge or booking.
Need to keep it working?
Want the alert settings reviewed and tuned monthly? Ask us about monthly tuning; it would be quoted.
Ongoing work is separately scoped and quoted: no monitoring, response-time guarantee or automatic subscription is included in this job.
Explore an ongoing managed lane, or 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
- Metaplane's free plan includes monitoring for a small number of tables. Try it first if your data sits in a warehouse it supports. www.metaplane.dev
- Your analytics or shop platform may have built-in alerts. Check their false-alert rate on last year's data.
Questions
Does it use AI to find the change?
Detection uses statistics. The script you run uses fixed alert wording. A model, if used at all, only helps us draft that wording while building.
Will it say why orders dropped?
No. It names where the change is, never why.
Do you run it for me?
No. You get a script and settings and run it yourself.
Send an enquiry
Send us
- The five metrics and how they are exported
- How many years of daily history you have
- No real records, personal data or credentials in the first enquiry
Later, once you agree
- A redacted or aggregate copy of the history by an agreed secure route
- Your hidden list of 5 past anomalies, held until the test
- A named person who receives the alerts
- A written cap on model usage
You keep your data and your monitor. We work on an aggregate export you send by an agreed secure route, return a script and settings you run yourself, and delete our copies when the job ends. We connect to nothing.
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 “metric-change-alert-with-segment-breakdown” as the subject.