The case study that a CTO will actually finish reading.
Ninety-three percent of agency case studies are abandoned before the second scroll. The seven percent that get read to the end share a specific structure, a specific tone, and a specific set of numbers. Here is what to write instead of the case-study template every offshore agency is using.
A US CTO clicks a case-study link on an agency site and gives it about eleven seconds before deciding whether to keep scrolling. In those eleven seconds she is looking for one thing: whether the case study is a real engineering story or a marketing artifact with an engineering veneer. If it reads as marketing, she leaves. If it reads as engineering, she reads to the end and the agency has bought a serious pitch conversation.
The eleven-second test is the entire case-study problem, and almost every offshore agency's case-study page fails it in the same way. The failure mode is not writing quality. It is structural. The template being used is a customer-marketing template, and CTOs do not read customer marketing.
The template CTOs recognize as real
A case study a CTO reads to the end has four sections and one sidebar. Section one: the specific technical problem, stated in the terms an engineer at the client would have used. Not "the client faced scale challenges." Something like "the checkout service was returning 500s on 4% of requests during peak hours, traced to a Postgres connection-pool exhaustion pattern that only manifested under a specific cart-abandonment flow."
Section two: what was tried before the agency arrived. This is the section every marketing template deletes, and it is the section that establishes credibility. "The client's own team had increased pool size twice and added read replicas; neither worked because the root cause was a synchronous webhook call inside the transaction." This paragraph does more trust work than every award and logo on the site combined.
Section three: what the agency actually did, in enough technical detail that another engineer could evaluate the choice. "We refactored the webhook call out of the transaction, moved it to a background queue with idempotency keys, and added a circuit breaker on the downstream service." Not "we implemented modern engineering best practices."
Section four: what happened, with a number and a duration attached. "500-rate dropped to 0.2% within seven days. Peak-hour throughput doubled. The pattern held through the following Black Friday, at 4x the prior year's volume."
The sidebar is the engagement shape. Team size, seniority mix, duration, budget range, and the point of contact on the client side, ideally with a title and a link to a LinkedIn profile. The sidebar is what a CTO forwards to her CFO when she wants to hire the agency.
The numbers that carry weight and the ones that do not
Two kinds of numbers matter. Operational numbers that would appear in a runbook or a postmortem — error rates, latency percentiles, throughput, MTTR, cost-per-transaction. And business numbers that would appear on a board slide — revenue impact, retention impact, cost saved, deal size unlocked. Both are fine. The CTO wants operational, the CFO the CTO forwards it to wants business. Include both.
The numbers that do not carry weight are the ones agencies love: percentage of code coverage improved, hours saved on "manual processes," "customer satisfaction increased by 30%" with no source. Those numbers read as marketing-department authorship and they cost the case study its engineering credibility. Delete them from the drafts.
Naming the client, or how to write around it
Named clients read as three times more credible than anonymized ones, but only if the naming actually clears legal on both sides. Do not name a client without written permission. Do not use a logo without a signed logo-usage agreement.
For clients who will not consent to being named, write the case study around the industry, the stage, and the shape of the business, in a level of specificity that makes clear the case is real. "A Series B fintech, roughly 80 engineers, processing $200M annual GMV." That level of detail is checkable enough to be credible without exposing the client. Vague framings — "a major financial services firm" — read as fabricated even when they are not.
The three case studies to publish and the twenty to keep
The site should carry three to five case studies on the top level. Not twenty. The buyer is not going to read twenty. The buyer is going to read the first two carefully, skim the third, and decide. Twenty case studies of uneven quality damage the average.
The three or four published cases should be the ones that best match the narrow positioning claim. Every offshore agency has done a case that was interesting but off-brief; keep those in a private folder to send during specific pitches, and keep the public case-study page tight, narrow, and matched to the pitch the site is running.
The section every case study should end with and almost none do
The best case studies end with a paragraph titled "what we would do differently." One or two decisions that, in retrospect, the team would have made differently, and why. A named tradeoff that was chosen for good reasons at the time and turned out slightly wrong.
That paragraph does trust work no other section can do. It signals a team that runs post-mortems, that has calibrated judgment about its own work, and that will not oversell in the sales conversation the case study is designed to produce. Offshore agencies who add this paragraph to their existing case studies see the case-study-to-first-call rate rise inside a quarter.
- 01Eleven-second test: real engineering story or marketing artifact. CTOs abandon the second in eleven seconds.
- 02Four sections plus a sidebar: technical problem in the client's language, what was tried before, what you did with specifics, what happened with numbers.
- 03Operational numbers plus business numbers. Delete anything that sounds like it came from the marketing team.
- 04Three to five case studies on the top level, all matched to the narrow positioning. Twenty of uneven quality damages the average.
- 05End each case with a "what we would do differently" paragraph. It does more trust work than every award on the site.
Get vetted. Get listed. Get the paper that survives the memo.
Twelve-minute intake, three-day turnaround. A passing scorecard is the shortest path from a good deck to a serious shortlist.