A Three-Year-Old Case Study Reports on a Platform That’s Gone

You find an entry in a case study library with a result you would happily take, then you notice the date. Published three years ago. Nothing on the page tells you whether the thing that produced that number still exists. The result held. The platform underneath it did not stay still.

A case study reports two things, and only one of them is written down

Every case study is two claims stacked together. The first is explicit: this happened. An agency ran something, measured it, reported the outcome. Time does not weaken that claim.

The second is never written down: that the system which produced the result behaved in a particular way, and that the way it behaved is stable rather than a temporary setting. There is no expiry date on an assumption you did not know you were making.

A reader who checks the baseline, the window, the attribution and the sample, then stops, has done half the job: confirmed that a conclusion was reached properly, not that its premise still holds. Borrowing a conclusion whose premise has moved is how a well documented result turns into a bad plan. Most checks in how to read a marketing case study assume a fixed system. The date asks whether it is.

What tends to move underneath a case study without the write-up saying so

Categories, not events.

• Ranking and distribution logic. What gets shown, to whom, and how far it travels.
• Ad auction dynamics and pricing. Who else is bidding, and what attention costs as a result.
• Measurement definitions. What the platform counts as a conversion, a view or an engagement. A metric can keep its name and change its meaning.
• Available formats and placements. Whether the surface a campaign used still exists, and still sits where it sat.
• Policy on the tactic itself. Whether the approach is still permitted, still tolerated, or now restricted.

To point at a specific instance of any of these changing, you owe the reader a dated source published by the platform itself. Without one, the category stays general. A half remembered update, stated as fact, sounds like evidence.

A hypothetical case, to see the mechanism

What follows is hypothetical. No agency, no campaign, no platform, and none of it should be read as a real reported result.

Picture a hypothetical agency running one tactic twice, four years apart: short videos built to be watched to the end. In hypothetical year one, distribution leans hardest on completion rate. The work is built for that, and the agency reports hypothetical reach at four times the client’s baseline.

In hypothetical year five, distribution leans instead on whether a viewer sends the video to somebody else. Same brief, same posting rhythm, and hypothetical reach lands slightly under baseline. Nothing went wrong. The tactic was aimed at a signal that had stopped being the one that mattered, and the year one write-up could not have warned anybody, because back then it was not a variable.

There is no fixed shelf life for a case study

It is tempting to draw a line at two years: older is void, newer is live. Easy to apply, and wrong, because it measures the calendar instead of the thing that decides the question.

A case study published six months ago, resting on a format since restricted, tells you almost nothing about tomorrow. A three-year-old one resting on something durable can still guide you. An old date is a prompt to look closer, not a verdict.

Before you borrow a tactic from an older case study

1. Separate the publish date from the campaign date. An entry can go up long after the work ran, and the conditions belong to the campaign, not to the posting.
2. Find the named mechanism. Does the write-up name the format, ranking factor or ad type it depended on, or only the outcome? A result with no named mechanism cannot be checked, because there is nothing specific to look up.
3. Look for a more recent instance of the same tactic working. The strongest single check you have: a recent example is direct evidence that the mechanism is still live.
4. Check the tactic against current platform policy. A mechanism can survive while permission to use it does not. If the approach has been restricted or deprecated, the old result is inapplicable whether or not the machinery still works.

That is the same work as verifying what backs up a case study number, pointed at the conditions instead of the arithmetic.

Reading the date as part of the claim

The fix is not to distrust old case studies. An older entry is a historical data point about what worked under a dated set of conditions, and read that way it stays useful. Note the date every time as a stated condition of the claim, in the same breath as the sample size and the attribution window, not as a timestamp on the page.

Read more from the library

Browse the case study library, or submit your own. For what a library leaves out by design, every case study library is a selected sample is the companion read.

FAQ

How old does a case study have to be before I should stop trusting it?

There is no age at which one switches off. The useful question is whether the specific thing the result depended on, a format, a ranking factor, an ad type, is still in place. A recent result built on something since withdrawn transfers worse than an old one built on something durable.

Does a platform algorithm change actually invalidate an old case study?

It does not invalidate the reported result: that campaign ran and those numbers were measured, and the platform changing afterwards does not undo either. What a change can invalidate is the expectation that repeating the tactic now would produce the same outcome, because the system producing it is not the same system. Keep the record, drop the forecast.

What is the single most useful check before reusing an old case study’s tactic?

Go looking for somebody who has made the same approach work lately. One current instance settles what the old entry cannot say about itself: whether the mechanism is still there to be used. No amount of confidence in the original result gets you to that answer.

Sources

• How to read a marketing case study, fetched 2026-09-06.
• Self-reported versus verifiable numbers, fetched 2026-09-06.
• Survivorship bias in a case study library, fetched 2026-09-06.
• The case study library, fetched 2026-09-06.
• Submit your case study, fetched 2026-09-06.