A case study is a published account of a single campaign, deployment or customer relationship, describing a situation, an intervention and an outcome. In advertising and marketing technology it functions as evidence. Vendors, platforms and agencies use it to show that a product worked for somebody, and buyers use it to decide whether to test that product themselves. The form exists because the alternative is worse for both sides: controlled experiments across a whole customer base are slow and expensive, and most advertising technology is sold before such evidence exists.

Two lineages share the label. One is pedagogical, a written situation used to teach judgement. The other is commercial, a marketing asset produced by a seller about a buyer. The second dominates the advertising industry. When a supply-side platform announces that an early customer cut cost per conversion by 70 percent, that is a case study, and the distance between it and a randomised experiment has been argued over for as long as the format has existed.

What the document contains

The commercial format settled into four parts decades ago: client background, challenge, solution, results. Length runs from one web page to a twenty-page document. The results block carries the weight, and its construction determines whether the whole thing is usable.

Six fields decide that. Client identity comes first: a named advertiser signals that a legal and communications department signed off, while an anonymised one, described as a leading apparel retailer or a global automotive group, signals that somebody declined. The baseline comes second, because a 456 percent increase in form submissions means something different from a base of twelve than from a base of twelve thousand. Third is the comparison: year over year, before and after, and against a randomised control group are three different claims presented in identical language. Fourth is budget, since conversion volume responds to spending independently of whatever the product did, and fifth is duration and flight dates. Sixth is provenance: the vendor's own reporting interface, the advertiser's analytics, or a third party carrying Media Rating Council accreditation.

A worked example shows what is usually missing. When one agency published a year of results running campaigns through an automation platform in August 2026, the release reported a 98 percent cut in a creative refresh task that took 40 hours, a 456 percent year-on-year rise in form submissions, a 71 percent fall in cost per lead and a 14 percent fall in cost per click for one client. The client was unnamed, its sector unstated. No absolute volumes appeared, no media budget, and no comparison period beyond the phrase year over year.

Production follows a standard route: vendor product marketing drafts, the client's communications team approves, and distribution runs through a newswire, the vendor's resources hub, an awards entry, a conference stage and finally the sales deck and request-for-proposal response, where it does its real work.

The evidence question

The IAB and IAB Europe placed platform-reported figures in the weakest of four tiers in their incrementality framework, classing them as a hybrid proxy because they rest on platform attribution rather than an independent control group. Experimental designs sit at the top of the same hierarchy, which is why a randomised holdout and a case study answer different questions even when both report a lift percentage: one estimates what would have happened without the advertising, the other records what happened with it.

Provenance divides into who measured and who was audited. A vendor reporting its own latency improvement, as happened when infrastructure partners published timeout reductions above 80 percent and above 50 percent on their own traffic, describes a mechanism nobody disputes. A vendor reporting a conversion lift attributable to its own scoring product is making a causal claim about a metric it defines, computes and sells. Attention measurement illustrates the gap, since performance claims in that category rest largely on vendor case studies about vendor products.

Origin and evolution

The teaching case came first. Harvard Law School built a curriculum on compilations of court decisions from the 1870s, and Wallace Brett Donham, who studied under that system, became dean of Harvard Business School in 1919 intending to replace lectures with discussion of real business problems. The first published business case, The General Shoe Company, written by Clinton P. Biddle in 1921, ran to one page and described workers leaving a plant 45 minutes early while orders went unfilled. Classroom use became routine by 1924, and the school's Bureau of Business Research was created to produce more.

Advertising built its own case literature around competition. The Effie Awards launched in 1968 under the New York chapter of the American Marketing Association, honouring effectiveness rather than craft; the rights passed to Effie Worldwide in July 2008. The Institute of Practitioners in Advertising began its biennial Effectiveness Awards in the United Kingdom in 1980 and retained every entry, winning or not, in a structured databank that now holds more than 1,600 cases, which turned individual documents into a dataset. Les Binet and Peter Field analysed 996 campaigns entered between 1980 and 2010, covering more than 700 brands across 83 categories with over 200 coded fields per entry, and published The Long and the Short of It in 2013, from which the 60:40 brand-to-activation split entered planning orthodoxy.

Regulation arrived through testimonials. The Federal Trade Commission's Endorsement Guides, at 16 CFR Part 255, permitted advertisers under the 1980 version to describe unusual outcomes provided a disclaimer such as results not typical appeared. Revised Guides taking effect on December 1, 2009 removed that safe harbour, requiring clear disclosure of the performance consumers can generally expect plus substantiation for it, the Commission having found in testing that the old disclaimers failed to correct the impression they were meant to correct. A 2023 revision tightened what counts as a clear and conspicuous disclosure and extended liability across advertisers, endorsers and intermediaries.

Why it matters to the marketing community

Procurement runs on these documents. Business-to-business purchasing benchmarks put the average journey at 272 days across 88 touchpoints and 10 stakeholders, most of it before a vendor knows the buyer exists, which makes the published case the first and often only evidence in the room. Retrieval has raised the stakes again, because much of that evaluation now passes through large language models before any contact with a sales team, and a case study page is precisely the kind of structured claim those systems surface and summarise.

The format also carries results that reach the industry no other way: an automotive group documenting 13 billion measured impressions across 19 countries, a measurement company reconstructing one retailer's sports television buying down to 153 ad minutes, or an exchange reporting a 75 percent fall in cost per site visit for an apparel retailer.

Limitations and disputes

Selection is the structural flaw. Failures go unpublished, so the visible population consists of successes and the base rate behind any headline figure is unknowable. Awards databanks inherit the same problem, since entry is voluntary and costly, and analyses drawn from them have been criticised for circularity when the comparison set is itself defined by execution quality.

Independent checks sometimes disagree. In summer 2024 a grocery retail media operator analysed a verification vendor's attention product and found no uplift in clicks or sales attributable to high-attention placements, while finding conventional viewability correlated with sales. Marketplace data has also undercut single-vendor tests: agentic buyers were measured clearing inventory 13.4 percent below conventional programmatic demand on cost per thousand impressions, against vendor claims of two to four times revenue improvement drawn from internal tests. Authorship conflicts appear in the same territory: a June 2026 study reporting an 87 percent cut in task time drew its data from one of the two compared products, and three of its four authors worked there.

Supplier-published figures dominate categories where the capability is new: clean room results circulate as case studies of overlap rates and click-through lifts against control, written by the suppliers themselves. None of this makes the numbers false. It makes them uncorroborated, a different and more common condition.

Not the same as

A white paper argues a general position and may cite several deployments; a case study describes one. A benchmark report aggregates a vendor's whole measured population, supplying the base rate a single account lacks. A lift or holdout study establishes a counterfactual by randomised suppression, and can be reported inside a case study without the two becoming the same thing. A teaching case presents a situation without resolving it, because its purpose is discussion rather than proof, and it is written by a third party rather than an interested seller.

Recent developments

Agentic advertising has produced the densest run of single-campaign proof points the industry has seen. A supply-side platform reported a 70 percent cut in connected television campaign setup time from a first European test in September 2026, naming no advertiser and disclosing no budget or flight dates, which is how the category has been documented all year. Comparable claims support containerised bidding, where one early customer reported a 70 percent cut in cost per conversion, and platform modelling tools, where an open-source package built by a media seller has been promoted through a customer account reporting 20 percent quarterly revenue growth alongside a budget shift toward that seller.

The counterweight is thin but growing. Trade bodies have standardised the vocabulary of causal evidence, accreditation now covers more third-party metrics, and the largest deployments disclose scale, duration and the survey partner behind the brand results. The document itself remains what it was in 1921: a single situation, argued.

Timeline

  • 1870s: Harvard Law School establishes case-based teaching using compilations of court decisions
  • 1919: Wallace Brett Donham becomes dean of Harvard Business School and begins adapting the method to business
  • 1921: The General Shoe Company, written by Clinton P. Biddle, is published as the first business teaching case
  • 1924: Case teaching becomes routine across the Harvard Business School curriculum
  • 1968: The Effie Awards launch under the New York American Marketing Association
  • 1980: The Institute of Practitioners in Advertising runs its first biennial Effectiveness Awards and begins the Databank
  • 1980: The Federal Trade Commission issues Endorsement Guides permitting a results not typical disclaimer
  • July 2008: Effie rights transfer to Effie Worldwide
  • December 1, 2009: Revised FTC Endorsement Guides take effect, removing the typicality safe harbour
  • 2013: Binet and Field publish The Long and the Short of It, analysing 996 Databank campaigns
  • June 29, 2023: The FTC publishes further revised Endorsement Guides tightening disclosure standards
  • Summer 2024: A grocery retail media operator's internal analysis contradicts a vendor attention product's claims
  • February 2026: An exchange publishes a 75 percent cost per site visit reduction for an apparel retailer
  • June 4, 2026: A 13 billion impression attention programme across 19 countries is published as a case study
  • August 2026: An agency case study reporting a 98 percent task reduction circulates without audit
  • September 2026: A supply-side platform reports a 70 percent setup time cut without naming the advertiser

Summary

Who. Vendors, platforms and agencies write them; client communications and legal teams approve them; advertisers, procurement departments and agency planners read them. Trade bodies including the IAB, the Institute of Practitioners in Advertising and Effie Worldwide curate structured collections, the Media Rating Council accredits the measurement some of them cite, and the Federal Trade Commission governs how atypical results may be presented.

What. A published account of one campaign, deployment or client relationship, structured as background, challenge, solution and results, whose usefulness depends on whether the client is named, the baseline disclosed, the comparison specified, the budget stated and the measurement independently verified.

When. The teaching form dates to 1921 at Harvard Business School, the effectiveness competition form to the Effie Awards in 1968 and the IPA Databank in 1980, and the regulated testimonial form to the revised FTC Endorsement Guides that took effect on December 1, 2009.

Where. Newswires, vendor resource hubs, awards databanks, conference stages, sales decks, request-for-proposal responses and, increasingly, the retrieval layers of large language models.

Why. Because advertising technology is bought long before controlled evidence about it exists, and a single documented deployment is usually the only evidence available. Its weakness is structural rather than dishonest: failures go unpublished, base rates stay hidden, and the party reporting the result is normally the party selling the product.