Job metric-jump-or-drop-possible-causes-ranked-with-the-data-behind-each · revised 12 October 2026
When a metric jumps or drops, get ranked possible causes with the data behind each
For one metric change you get ranked possible causes. Each one shows the data that supports it, the data that would rule it out, and a next check. No cause is stated as fact.
You might be seeing
- A key number changed and three people give three stories
- Nobody has checked the story against the data
No passwords, keys, card details or admin invites needed to start.
What usually happened
A fluent explanation of a metric change is easy to produce and easy to believe. Practitioners report that automated root-cause guesses can be wrong even when they point near the real cause. The honest form is a ranked list of hypotheses, each with the data for and against it, and "no candidate found" when the data and change log give none.
Who it’s for: Ops, product or e-commerce lead whose daily export shows a sudden change in orders, sign-ups or refunds and who has hours, not days, to find where to look.
Usually starts when: A number moved sharply, three people have three explanations, and nobody has lined up the change log against the data.
The result: For one metric change in one export, you receive a ranked list of up to five possible causes. Each cause shows the data that supports it, with a query you can re-run, the data that would rule it out, and the next check. Causes come from your change log and the data's segments, a change with no candidate is reported as such, and no cause is stated as fact.
Check whether this job fits
Answer these without sending files, tickets, code or logins. Nothing is submitted unless you choose to contact us.
What you get
- The ranked possible causes with supporting and ruling-out data
- The queries or pivots
- The segment tables
- The re-run command
- A note on what the analysis cannot show
Included
- One metric and one date range in one export of up to 1 million rows, with no personal data
- Your change log: releases, price changes, campaigns, outages, promotions and stock events, with dates
- Comparison of the change against segments (product, channel, country, device) and against the change log, by code
- Up to five ranked possible causes, each with supporting data, ruling-out data and next check
- The queries or pivots and a one-line command to re-run all of them
Not included
- Stating any cause as fact
- Forecasts, advice or a decision on what to do
- Access to your warehouse or analytics logins
- Personal data or user-level event logs
- Causal claims from tests you did not run
- More than one metric change per job
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
A script recomputes every figure in the report from the export and finds them equal within the stated rounding (100%, because this tests code). A figure that cannot be reproduced is labelled an estimate.
Evidence: The recomputation log.
Of 12 past changes, at least 10 have the known cause in the top three proposals, including at least 5 of the 6 changes you keep from us until the test run. For the held-back 6, 5 passes has an exact one-sided 95% lower bound, rounded down, of 41%; the other 6 are development cases and carry no rate claim.
Evidence: The results table with your known causes.
All 3 planted changes with no change-log entry and no segment explanation are reported as "no candidate found". One invented cause fails this smoke test; it supports no general rate claim.
Evidence: The planted-change table.
Your named reviewer reads every proposed cause and confirms it is labelled possible, includes the data that would rule it out and gives a next check. One cause asserted as fact or lacking either item fails the whole test.
Evidence: Your named reviewer's marks for every proposed cause.
A planted instruction inside a cell of the export changes no cause and no figure, and a second run gives identical output.
Evidence: Both outputs compared. A smoke test of one planted instruction.
Sign-off. You re-run the acceptance checks with the steps we hand over, read the results and sign off in writing. Payment follows sign-off.
If it fails. If an agreed check fails, you do not pay for this scope. We hand over what we found and agree whether to stop or re-quote; no surprise work.
When it fits, and when we stop
It fits when
- One export with the metric by day and its main segments
- A change log with dates, or a statement that none exists
- A person who knows the business and checks each proposed cause
- 12 past metric changes whose cause you know, 6 kept from us until the test run
We stop and tell you if
- You want a single definite cause stated
- The export holds personal data or user-level events
- There is no history before the change
- You want live access to your analytics tool
What could go wrong
The job reads an export and reports, so there is nothing to undo. Our copies of your files are deleted when the job ends.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| A proposed cause is wrong but believed. | Every cause is worded as a possibility, shows the data against it and gives a next check. The cause that fits the data best may still be wrong. |
| No cause is in the data. | The job reports "no candidate found" and lists the segments that moved, instead of inventing a cause. |
| A figure in the report is wrong. | Every figure is recomputed by script from the export, and any that cannot be is labelled as an estimate. |
| Text inside the export steers the analysis. | Cells are treated as data. A planted instruction in the export is tested and must change nothing. |
A separate automated reviewer checks the work and the test results before hand-over. Your named person checks each proposed cause against what they know and decides what to do. No human supervisor from us 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 metric, dates, export shape and change log
- Receive the export and log; add your 12 past changes
- Break the change down by segment and line it up with the log
- Write each candidate cause with its for and against data
- Check every figure recomputes; independent automated review of wording
- Send the ranked causes, queries and re-run command
This is a one-off job, not emergency cover or a subscription. We confirm eligibility, price, start window and delivery date in writing before you accept, and work starts only after the agreed inputs are in place. An enquiry creates no charge or booking.
Need to keep it working?
Discuss the analyst retainer if you expect to ask this kind of question every month.
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
- Break the metric down by one segment at a time in a spreadsheet and compare with your change log.
- Use the anomaly alerts in your analytics tool and follow their segment hints.
Questions
Will you tell me why it happened?
We tell you what might explain it, with the data for and against each possibility. You decide which is right.
How is this different from the metric alert job?
That one says when a number moved and which segments moved. This one proposes why, with evidence, and never states a cause as fact.
What if the change log is empty?
Then the job can only show which segments moved, and it says so.
Does a pass mean it will be right next month?
No. A pass describes items like the test sample on the day of the test. It is not a promise about future items, and it does not replace your own checking.
Send an enquiry
Send us
- Which metric, which dates and how large the change is
- What the export holds
- Whether a change log exists
- No logins, keys, customer data or payment details in the first enquiry
Later, once you agree
- The export through a secure route
- The change log
- 12 past changes with their known causes, 6 kept from us
- The name of the person who checks each cause
You keep your analytics and every login. We work on the export and change log you send through a secure route, and hand back files and a report. We never ask for warehouse or analytics logins.
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-jump-or-drop-possible-causes-ranked-with-the-data-behind-each” as the subject.