The agency case study that wins business: building proof that travels

Most agency case studies are written for the agency. They record what the team did, in the order the team did it, in the language the team used internally. They are accurate, often handsome, and almost useless to anyone deciding whether to hire you.

The case study is the most reused asset an agency owns. The same piece of proof ends up in a credentials deck, an awards entry, a journalist's inbox and, increasingly, in the sources an AI engine draws on when a marketing director asks it for recommendations. Those four readers want different things from the same material. An agency that builds the proof once, properly, gets four uses from it. An agency that builds it for itself gets a nice page on its website.

This is the shape of what each reader needs, where case studies usually break, and the levers that make one piece of proof travel.

A case study is proof, and proof has four readers

The useful way to think about a case study is not as a piece of content but as a claim with evidence attached. The claim is that your agency produced a specific result for a specific client. The evidence is whatever lets a stranger believe it.

Each of the four readers applies a different test to that claim. The buyer asks whether it is relevant and defensible. The jury asks whether it is ambitious and attributable. The journalist asks whether it tells their readers something beyond your client. The AI engine asks, in effect, whether the claim is stated clearly enough and corroborated widely enough to repeat.

A case study that passes one of those tests and fails the other three is common. One that passes all four is rare, and it is rare because it has to be designed that way from the start rather than rewritten for each use.

What a buyer needs from a case study

A buyer, and the procurement team behind them, is reading for risk. The question in their head is whether this agency has solved a problem like theirs, at a scale like theirs, and whether they can defend the choice to someone more senior.

That means the result has to be expressed in terms the buyer's organisation already uses: a commercial or marketing outcome, with a baseline and a period, rather than an engagement metric that only means something inside the agency. It means the client's context needs to be visible - sector, scale, the brief as the client would describe it - so the buyer can see the resemblance to their own situation. And it means the senior people who did the work should be named, because the buyer is hiring them.

The case study also has to survive being forwarded. A marketing director who champions your agency internally will paste a paragraph into an email to a finance director who has never heard of you. If the proof only works with the agency narrating it, it does not work.

What an awards jury needs from it

Juries read for ambition, craft and effect, and they are sceptical of results they cannot attribute to the work. The gap between what juries reward and what the trade press covers is real, and it is the subject of what awards juries reward versus what the trade press writes about. For case-study purposes the practical point is narrower: juries want to see the working.

The Drum's reading of more than 500 entries to The Drum Awards for B2B, published on 7 September, is a useful illustration. Just over half of the entries credited AI with helping deliver a better business or marketing outcome, and entrants, in The Drum's words, "were happy to show their working". The receipts were specific: production costs down by 55% in one case, more than 500 targeted assets built without adding core headcount in another, first drafts cut from between three and six hours to under an hour.

Only 5% of entries were willing to say on the record that AI had a hand in the creative itself. Entrants were comfortable claiming efficiency with numbers attached, and far less comfortable making claims about authorship. Juries notice where a case study is precise and where it goes quiet.

The same analysis found that more than 75% of entries reporting a brand outcome also reported leads, pipeline or revenue. Juries are increasingly reading brand and commercial effect together, which is exactly what a buyer wants to see as well.

What a journalist needs from it

A trade journalist has no interest in your client's campaign as such. They are interested in what it shows about a category, a channel, a buyer behaviour or a way of working that their readers would recognise.

So the case study that gets coverage contains a finding that is bigger than the client. A number that generalises. A named person, on the agency or client side, prepared to say something specific and slightly contestable about why it worked. And a reason it is relevant now rather than eighteen months ago, when the work actually ran.

Most case studies contain none of those, because they were written to describe the project rather than to support an argument. The raw material is often there. It is buried in the results section, or in what the strategist said in the debrief, and never made it onto the page.

What an AI engine needs from it

Engines assemble answers from sources that state a claim plainly and are corroborated elsewhere. A case study that lives only as a designed PDF in a credentials deck, or as a scroll-heavy page where the result appears in an image, gives an engine very little to work with.

What helps is the unglamorous part: the agency, the client, the problem and the outcome stated in a single sentence of plain text; the same figures appearing consistently across the agency's site, an awards shortlist and any coverage; and third-party pages that mention the work without the agency having written them.

This is still early territory. The same Drum analysis found only 3% of B2B entries reported a measurable result from reaching AI answer engines, and described marketing to machines as work that "has barely started". For agencies, that is an opening. Very few are building their own proof with this reader in mind.

Why most case studies fail all four

The failure patterns are consistent enough to be diagnostic.

The case study is told as a chronology - brief, insight, idea, execution, results - so the result arrives last and gets read least. The outcome is expressed as a vanity metric the client's CFO would not recognise. Adjectives do the work that numbers should: "record-breaking", "industry-leading", "unprecedented". The client is anonymised as "a leading retailer", which removes most of the credibility in a single phrase.

Then the structural ones. The figures differ between the website, the awards entry and the pitch deck, because each version was rewritten separately under deadline. Client approval to name them, quote them or publish the numbers was never secured, so the strongest version of the proof can only be used in private. And the agency's contribution is not separated from everything else that happened, so a sceptical reader discounts all of it.

A case study that has to be rewritten for every audience was never really proof - it was a draft.

The levers that make proof travel

None of these is a template, and each has more judgement in it than it first appears. They are the points where one well-built case study separates from twenty adequate ones.

1. A single claim a stranger could repeat. One sentence that names the client, the problem and the outcome. Everything else in the case study exists to support that sentence, and if the team cannot agree on it, the case study is not ready.
2. A number with its denominator. The result, the baseline it moved from and the period over which it moved. A percentage without the base is the first thing a procurement lead or a juror discounts.
3. Permission secured early. Client sign-off to be named, quoted and to publish figures, negotiated while the relationship is warm rather than at the awards deadline. This has the longest lead time of anything here and is the one most agencies leave until it is too late.
4. Honest attribution. What the agency did, what the client did and what the market did, separated clearly - including where AI was used. Claims that acknowledge their limits are believed more, not less, and the decision about disclosing AI in client work now sits squarely inside this lever.
5. Third-party corroboration. A shortlisting, a trade mention, a client quote on the record. Proof the agency did not write itself is what buyers forward and engines repeat.
6. One source of truth. The same figures and the same claim everywhere the case study appears, maintained as a single master rather than a set of versions drifting apart.

Each of those is a lever rather than a step. Pulling any one well is a piece of work with its own calendar, and the difficulty sits in the sequencing between them and in the conversations with the client they depend on.

Where the proof breaks between shortlist and pitch

The moment a case study is most tested is not when it is written. It is after an agency has been shortlisted, when someone senior at the prospect goes looking for evidence and finds either a well-built page that confirms the shortlist or a thin one that quietly undoes it.

That is the credibility gap between shortlist and pitch, and case studies sit at the centre of it. A buyer who has heard good things will read your best case study before your first meeting. Whatever it says, or fails to say, forms the view they bring into the room.

Build it once, before you need it

The practical conclusion is that a case study should be built at the point the result is known, with all four readers in mind, rather than assembled under deadline pressure for whichever use arrives first.

An agency that does this finds that the awards entry, the trade pitch, the credentials slide and the website page stop being four pieces of work. They become four cuts of the same proof - the same principle behind the post-awards atomisation playbook, applied earlier in the life of the work. That is a large part of how we approach an agency's reputation: the evidence comes first, and everything else is distribution.

If your best work is sitting in case studies that only your own team finds convincing, that is worth a conversation.