Who this is for
E-commerce or operations lead with a daily export who learned of a drop in orders days late.
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.
What is 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
You receive
- The script and settings
- The backtest report with every false alert listed
- Sample alerts with their recomputation
- A 'no data' alert
- Run steps
What is not included
- Root causes or explanations
- Running or hosting the monitor for you
- Access to your shop, database or analytics account
- Forecasts
- Personal data
What we need from you first
- 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
Never send passwords, keys, customer records or confidential code in the first enquiry. Secure handover is agreed after scoping.
How we check it is done
- 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.
- Every figure in a sample of 10 alerts recomputes from your file.
- No alert states a cause. A script search for causal words finds none.
- 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.
- The alert wording is a fixed template filled from the figures, and two runs on the same file give identical alerts.
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.
When we would stop or decline
- 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
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.
Price and terms
£295 · untested offer price. £295 after the agreed test passes on the held-back part and you sign off. No payment before sign-off.
This is a new service with no published client results. The price is a starting point we have not yet tested with buyers. Nothing is ordered or charged by the enquiry. The full specification is on the Synthetic Industry catalogue.