The Definition Changed and Nobody Said
How a reporting change becomes a fake trend, the six places definitions change, and the one-line record that prevents years of misreading.
A definition change produces a step in the series, and a step fitted through becomes a trend. That trend then gets extrapolated for years, and the resulting forecast describes a counting convention rather than any demand that exists.
The method in “The Definition Changed and Nobody Said” is only as useful as the definitions and source records behind it. A team exploring read the current platform resource for ethical employee monitoring can add timely work and project context, with transparent settings, limited access and a documented correction route around every consequential use.
This is the most consequential and least glamorous problem in demand data, and it is invisible unless somebody looks for it deliberately.
For an independent reference relevant to “The Definition Changed and Nobody Said”, consult Atlassian project-management guide. Compare the general material with the organisation’s actual horizon, data definitions and decision process.
What it looks like in the data
A level shift at a single date, with the shape before and after otherwise similar. Not a ramp, not a seasonal effect — a step.
- A new system, or an upgrade to one.
- A queue merged or split; a category redefined.
- A channel added to the count.
- A threshold changed: what counts as abandoned, what counts as done.
- An organisational change that moved work in or out.
The giveaway is that it falls on the first of a month or the start of a quarter, which demand does not respect and release schedules do.
The six places it happens
A new system or a system upgrade. A queue merged or split. A category redefined. A channel added to the count. A threshold changed — what counts as an abandoned contact, what counts as a completed task. And an organisational change that moved work in or out.
Each produces the same signature and none of them is a change in demand.
Why nobody says
Because the people who make the change are not the people who use the series, and the change is correct from their point of view. Adding abandoned contacts to the count made the operational report better.
It also made six years of history incomparable, and nobody in that conversation was thinking about the forecast.
Finding them retrospectively
Plot the series and look for steps. Then ask, for each step, what happened on that date — release notes, change records, org charts.
In most series with five years of history there are two or three. An afternoon finds them, and it is an afternoon most planners have never spent.
Separating a step from a real change
A real demand change usually ramps and usually affects the shape as well as the level. A definition change is instantaneous and leaves the shape alone.
Where it is genuinely ambiguous, the change record settles it, which is why the change record is worth finding first.
What to do with a step once found
Three options. Use only the data after it. Estimate an adjustment factor and apply it to the earlier period, labelled. Or keep both and model the step explicitly as a known level shift.
The third is best where the series is short and the step is well understood; the first is safest.
The adjustment factor
Estimated from the overlap if there is one, or from comparing a period before and after with similar seasonal position. It is an estimate, it carries error, and it should be written as such.
An adjusted series with no note becomes, within a year, a series everybody believes was measured.
The change that is still happening
Some are gradual: a channel migrating, a category slowly being used differently. These are harder to find and they look exactly like a trend.
The distinguishing question is whether the total across categories is stable while the split moves. If it is, the trend is a reclassification.
Preventing the next one
A standing arrangement with whoever owns the reporting: tell the planner before a definition changes. Not approval — notice.
It costs nothing and it is usually granted immediately once somebody explains what a step does to a forecast.
The one-line record
A file with one line per change: date, what changed, estimated effect, who to ask. Four columns.
That file is the single highest-value document a planning function keeps, and it takes minutes a year to maintain once it exists.
What it is worth
In one operation's own experience, a step of about six per cent was carried as a trend for two years, producing a far-horizon forecast roughly twelve per cent high and a headcount plan built on it.
The step took an afternoon to find once somebody plotted the series and asked what happened in April.
Changes that affect only part of the series
Harder to see than a clean step: a definition that changed for one category, one site or one channel while the rest stayed as it was.
The total shows a small shift and the composition shows a large one, which is why the composition is worth plotting alongside the total whenever a step is suspected.
The change made for a good reason
Nearly all of them are. The new definition is usually better, and reverting it would make the operational reporting worse to make the forecast easier.
The answer is almost never to ask for the change to be undone. It is to record it, estimate the effect, and set the usable start date.
The broader habit
Treating the series as a measurement with a method, rather than as data that simply exists. Measurements have definitions, definitions change, and changes get recorded.
Everything in this section follows from taking that one sentence seriously.