Campaign case studies: a practical reference for 2027

A campaign case study is a marketing document about marketing. It was written by someone with an interest in the conclusion, using data they selected, about a campaign they chose to write about rather than one of the ones they did not. That does not make it worthless. It makes it evidence of a particular kind, and the skill is knowing how much weight it can carry.

This page is about reading them and about writing honest ones. It contains no examples of real campaigns, because a case study whose details you cannot verify is exactly the thing it is warning you about.

What to take away

• Before any number, establish the author's position, because it predicts which facts were included.
• Certain phrasings recur because they let a writer imply more than the data supports.
• Ask them in this order, because the early ones make the later ones unnecessary.
• The point of an internal record is that in a year you will want to know what actually happened, not what you told people happened.

Start with who wrote it and why

Before any number, establish the author's position, because it predicts which facts were included.

An agency case study exists to win business. It will describe a campaign that worked, attribute the outcome to the agency's method, and omit the campaigns that did not work. This is not dishonesty; it is what the genre is for.

A platform or vendor case study exists to show that the platform's own measurement makes the platform look effective. Watch for the circularity: the tool that reports the result is frequently the tool being sold.

A brand's own published case study is usually a recruitment or reputation document as much as a marketing one, and tends to be vaguer about numbers because legal reviewed it.

A conference talk sits somewhere in between and often contains the most useful detail, because a live audience asks questions.

An internal post-mortem is the most valuable of all and almost never public, which is why building your own is worth more than reading a hundred external ones.

None of these should be dismissed. They should be read with the author's incentive in mind, and the questions below applied harder where the incentive is stronger.

The seven things that are usually missing

The baseline. A result is a difference, and a difference needs a starting point. "Sales rose during the campaign" is not a finding without knowing what sales did in the equivalent prior period, in the same season, and in the absence of the campaign. If the baseline is absent, the number is a level being presented as a change.

The counterfactual. What would have happened anyway. This is the hardest thing to establish and the most commonly skipped. A case study that includes a holdout market, a matched control, or even an honest acknowledgment that no control existed is doing something the others are not.

The cost. Almost every published case study omits the full cost, which makes any efficiency claim unverifiable. Fee, production, product, agency margin, paid amplification behind the content, and staff time all belong in the denominator.

The denominator generally. Percentages without absolute numbers are the oldest device in the genre. A large percentage increase from a small base is a small increase.

The timeframe, and why that one. Windows get chosen. If the reported period ends conveniently at a peak, or extends until the result became positive, the window is doing the work rather than the campaign.

The rest of the marketing. Campaigns run alongside other campaigns. If the brand also launched a product, cut a price, or ran television in the same window, a claim that this channel produced the result requires an argument, not just a chart.

The failures. How many partners were involved and how the results were distributed. A campaign of twelve creators carried by one is a different lesson from twelve creators performing consistently, and only the second supports the method being sold.

Vocabulary that signals a soft claim

Certain phrasings recur because they let a writer imply more than the data supports. Recognizing them is quick and it saves reading the rest carefully.

"Reach" and "impressions" used as an achievement. These describe distribution purchased, not effect produced. A campaign that reached a large number of people and changed nothing has still reached a large number of people.

"Up to." Names the best observed value in the set and tells you nothing about the rest.

"Over-delivered against benchmark." Ask whose benchmark, set when, and by whom. Internal targets can be set to be beaten.

"Record" or "best-ever." True of some metric, over some period, chosen after the fact.

"Drove," "generated," "delivered." Causal verbs attached to correlational data. They are doing work no stated method supports.

"X times return." Meaningless without the attribution model, the window, the margin assumption, and the cost base. The same campaign yields very different multiples under different models, and the one published is rarely the most conservative.

Engagement rate quoted without a denominator. It can be computed against followers, reach, or views, producing very different figures. Two case studies quoting it are usually not quoting the same thing.

"Incremental" used loosely. A word with a precise meaning, the difference against a control, routinely used to mean "attributed." Ask what the control was. If there was none, the word is unearned.

Questions to ask when a vendor presents one

Ask them in this order, because the early ones make the later ones unnecessary.

How was the result measured, and by whom? Was there a control group or a holdout, and if not, what accounts for the counterfactual? What is the full cost, including everything not in the fee? What was the attribution window, and what does the number become at half that window? How many partners were in the campaign and what does the distribution look like, not the average? What else was running at the same time? How many campaigns did you run for clients in this category in the same period, and why this one? What did not work, and what would you do differently?

That last pair is the most informative question in the set. An honest practitioner has a ready answer. A presentation that has no failures in it is a presentation that has been curated past the point of usefulness.

If the answers are unavailable because of client confidentiality, often genuinely true, then treat the case study as a description of an approach rather than as evidence of a result. That is still useful. It is just not proof.

Writing an internal case study worth keeping

The point of an internal record is that in a year you will want to know what actually happened, not what you told people happened. Write it for that reader.

State the decision it should inform. A case study that does not change a future choice is a scrapbook, and the decision it informs should trace back to the aim in strategy and objectives.

Record the plan as it was written beforehand. Objective, mechanism, chosen metrics, and the stated definition of failure. Copy it in verbatim rather than summarizing, because the summary will drift toward whatever happened.

Record what you actually bought. Partners, formats, deliverables, usage terms, timing, and full costs, with the fee recorded beside the rights and exclusivity it paid for rather than on its own; the reasoning is in rates and negotiation and the terms themselves in contracts and disclosure. Include the things that are embarrassing, like the reshoot or the three-week approval delay, because those are the causes you can act on next time.

Record the result with its method attached. Every number gets its definition, its source, its window, and its known limitation beside it. If tracked conversions are a floor rather than a total, say so in the document, not in someone's memory.

Record the confounds. Everything else that was live in the window. Write it even when it weakens the story, because you will otherwise re-learn this the expensive way.

Separate what you know from what you think. Two headings. Observations under one, interpretation under the other. This single habit does more for the usefulness of an internal record than any template.

Say what would change your mind. If the next campaign produced the opposite result, what would you conclude? Answering it now, before you know, is what makes the document a test rather than a story.

Keep the ones that failed. A library of successes teaches you nothing about what to avoid, and the failures are where the transferable lessons are.

How much weight a case study can carry

A single campaign is one observation of a noisy process. Platform distribution varies, audiences vary, timing varies, and a large share of outcome variance in this channel is not under anyone's control. Published cases also exist only because they went well enough to publish, so a collection of them is a sample of successes rather than a sample of campaigns, which is ordinary survivorship bias. That means a single case study, yours or anyone's, supports a hypothesis and does not establish a rule.

Use them for the things they are actually good at: generating ideas about mechanisms, showing how someone structured a deal, revealing what a category's content conventions are, and surfacing failure modes you had not considered. Do not use them to set expectations for your own results, and be especially wary of a case study used to justify a price.

Where a claimed result would change a real decision, the right response is to test it small rather than to believe it. The methods for doing that are in measurement and ROI, and the reason the tracked numbers in most case studies are shakier than they look is explained there too.

Bottom line

Read every case study as an argument made by an interested party. Look first for the baseline, the counterfactual, and the full cost, since the absence of any one of them makes the headline figure unusable. Treat causal verbs and unattached percentages as claims to be checked. Then write your own internal versions properly, keep the failures, and record what would have changed your mind.

Common questions

Are published case studies worth reading at all?

Yes, for method and for ideas. Not for benchmarks, and not as evidence that a similar result is available to you.

What if a case study includes detailed data?

Detail is not the same as rigour. A page of charts built on last-click attribution with no control is precise about the wrong thing. Check for the baseline and the counterfactual first, regardless of how much data is shown.

Does publishing customer results carry any compliance risk?

It can, particularly where a case study functions as a testimonial or where results are presented as typical. The FTC's questions and answers on the consumer reviews and testimonials rule is the place to read the current position, and anything you publish about results should be reviewed by a lawyer rather than by a marketing team alone.

Should we let an agency publish a case study about our campaign?

That is a commercial decision, but negotiate approval rights over the numbers and the claims. Your name attached to an unsupportable figure is your problem, not theirs.

Can we write a case study when the results were mixed?

Those are the most useful ones internally. For external publication, mixed results tend to be more credible than clean ones, and an honest account of what did not work is unusual enough to be memorable.

How do we compare two case studies from different vendors?

Usually you cannot, because they will not share a method. Reissue your own questions to both and compare the answers rather than the documents.