Annual trend reports: a practical reference for 2027

The trend report is a genre, and like every genre it has conventions that tell you more than its contents do. Learn how one is produced and you can read any of them in twenty minutes, take one decision from it, and score it a year later. This page is the reference for that routine. It names no report and quotes no figure, because the routine is the same whichever one is on your desk.

What to take away

• A report is produced backwards: the publisher's reason for publishing comes first, the narrative second, and the data is gathered to fill the narrative. Read it in the order it was made, not the order it is printed.
• Separate description from prediction on every page. The descriptions are usually fine. The predictions are the marketing.
• Keep last year's report and score it. Nobody else will, and a report's track record is the only evidence of its worth.

How a report gets made

Every publisher of an annual report has a reason that is not "the public needs to know". A tool vendor wants to be the source people cite. A platform wants a version of the year that flatters its surfaces. An agency wants to look like it can see ahead. A survey house wants press coverage. None of these is disreputable. All of them shape the document.

From the reason comes a narrative: a handful of themes the publisher wants associated with itself. Then comes the data, and the data comes from what the publisher can reach cheaply: its own product's usage, a survey of its own customers or a panel it can rent, and public sources. The survey methodology section, when there is one, is where the reachable population is quietly described. A report about "marketers" is about the marketers who answer that publisher's surveys.

Finally the charts are made, and the ones that do not support the narrative do not appear. This is not fraud. It is editing, and the result is a document that is systematically more confident and more coherent than the year it describes. The general version of this filter, in which interesting results get published and dull ones do not, is publication bias, and a trend report is a small private instance of it.

Reading one in twenty minutes

Read it in the order it was produced.

1. The publisher and the sponsor. Two minutes. Who pays, what they sell, and which of the report's themes would help them sell it.
2. The methodology. Five minutes, and if there is none, stop and downgrade everything that follows. Who was asked, how many, how they were reached, when, and what the response rate was. Note any question wording quoted.
3. The definitions. Three minutes. What counts as an "active" anything, what the time period is, what region. A trend is a change in a defined quantity, and a report that changed the definition since last year is describing two different quantities.
4. The charts, without the captions. Five minutes. Read the axes first. Percentages without bases, changes without the level they changed from, and growth rates on tiny starting numbers are the usual three.
5. The predictions, marked. Five minutes. Go through with a pen and mark every sentence that is about the coming year rather than the past one. These are the sentences to score in twelve months. They will be far fewer than the report's tone suggests.

Description or prediction

A description says what a measured quantity did over a stated period in a stated population. It can be wrong, but it can be checked, and it is usually roughly right because the publisher would be embarrassed otherwise.

A prediction says what will happen. It cannot be checked until later, so it costs the publisher nothing now, and it is where the narrative lives. Predictions in trend reports cluster around whatever the publisher benefits from your believing, and they are phrased to be unfalsifiable: a format "will continue to matter", a behavior "is set to grow".

The reading discipline is to strip the verbs. For each sentence, ask whether it reports a number or forecasts one. Report the numbers to your team. Treat the forecasts as the publisher's wishes, which they largely are.

Descriptions have a limit too. A description of last year, however accurate, is information about a system that has since changed. The argument that ranking systems change faster than any description of them is made in what a ranking input can and cannot be, and it applies with equal force to any report about what worked.

Take one decision

A report earns its reading time if it changes one thing you do. Choose the one thing before you close the document, and write it down with the page that prompted it.

Good candidates are decisions you were already weighing, where the report supplied a description that tipped them: a surface to try, which your change register will already have logged; a format to stop; a segment to ask about. Poor candidates are decisions the report invented: a new priority that did not exist before you opened it. If every report you read generates a new priority, the reports are running your plan.

Most of the decisions worth taking from a report are about the product layer of platforms, which changes slowly enough that a yearly document can describe it usefully; the difference between that layer and the faster ones is in what beginners should know about platform change. Content-layer trends in an annual report are already over, and how a content trend rises and dies explains why a yearly document cannot catch one.

Score last year's

This is the step that nobody does and everyone should.

• Keep every report you read, with the predictions marked.
• Twelve months on, go through the marked sentences and grade each: happened, did not, cannot tell.
• Tally by publisher. After two or three years you have a track record for each source that no amount of production value can fake.
• Drop the sources that score badly. Read the ones that score well first the following year, and give their descriptions extra weight because their predictions earned it.

The scoring also teaches you something about yourself: which predictions you believed, and why. Usually it is the ones that agreed with what you already thought, and the record shows those were not more accurate than the rest.

Common questions

Are reports from platforms themselves more reliable than third-party ones?

More reliable on the platform's own usage numbers, because they see them directly. Less reliable on what they mean, because the platform is the interested party. Read the methodology and the sponsor line with the same care either way.

How many reports should I read a year?

Fewer than you receive. Three or four, chosen by track record, read properly, beat twenty skimmed.

A report contradicts my own numbers. Which should I believe?

Yours describe your audience; the report describes its sample. Both can be right. The interesting question is what is different about your audience, and that is a question worth an afternoon.

What if there is no methodology section at all?

Then you have a document of opinions with charts. Read it as opinion, take nothing from it as a number, and note the absence in the track record.