Learning how to write a B2B case study is not the hard part: Problem, Solution, Results is universal, which is exactly why it differentiates nobody. The three things that actually determine whether a case study closes deals are getting real numbers approved by your client's legal team, structuring it so an AI assistant will cite it when a buyer asks "who has solved this for a company like us," and handling the murky attribution problem that services businesses face and product companies don't.
This post covers all three. And because a post about case studies should demonstrate rather than describe, I'm using a real one as the worked example throughout: the Masai School organic growth work: Instagram 26K to 117K, LinkedIn 50K to 160K. You'll see the artefact and the method at the same time.
Key Takeaways
- Problem-Solution-Results is table stakes. Nobody has ever lost a deal because their case study lacked a framework.
- The real bottleneck is approval. Most case studies die in a client's legal or comms review, not in the writing.
- Anonymisation tiers and relative metrics let you publish numbers that absolute figures would never get past legal.
- Around 74% of B2B buyers now use LLMs to verify vendor claims, which makes a well-structured, retrievable case study the single most valuable content asset you own.
- To be AI-citable, a case study needs a self-contained direct answer, named specifics, dated figures, HTML tables, and a plain statement of the situation it applies to.
- Services businesses should publish contribution, method, and correlation honestly rather than claiming causation they can't defend.
- One case study with an approved, verifiable number beats ten vague testimonials.
Why the Standard Advice Fails
Search the topic and you get the same article twelve times: use the Problem-Solution-Results structure, include a quote, add a metric, use a photo of the client.
All correct. All useless as differentiation, because your competitor read the same article. And none of it addresses the three reasons case studies actually fail in practice.
They never get published, because the client's legal or comms team won't approve the numbers.
They never get found, because they sit in a gated PDF or a /customers/ page structured for a human browsing your site, not for a buyer asking a question somewhere else.
They aren't believed, because a services case study claiming "we grew revenue 340%" invites the obvious question of what else was happening that year.
Fix those three and the framework barely matters.
Part One: How to Get the Numbers Approved
This is the actual bottleneck, and I've almost never seen it written about seriously.
Here's the typical sequence. You write a great case study. You send it to your champion. Your champion loves it. Your champion sends it to comms. Comms sends it to legal. Legal asks who approved disclosing performance data. Six weeks pass. The champion changes jobs. The case study dies.
Ask at the right moment
The single highest-leverage change: get approval rights into the conversation at the start of the engagement, not at the end. A clause in the SOW: "Client agrees to participate in one case study, subject to approval of specific figures disclosed", converts a six-week negotiation into a two-week review.
Failing that, the second-best moment is immediately after a visible win, while your champion is enthusiastic and has internal capital to spend. Waiting until "we have a full quarter of data" usually means waiting until nobody cares.
The anonymisation tiers
Most companies treat this as binary: named client or no case study. It's a spectrum, and each step down the ladder is dramatically easier to approve.
Tier 1: Fully named, absolute numbers. "Masai School grew Instagram from 26,000 to 117,000 followers." Maximum credibility, hardest approval. Worth fighting for with one or two flagship clients.
Tier 2: Named client, relative numbers. "Masai School grew Instagram followers 4.5x." Same credibility on the brand, no disclosure of absolute scale. Approval rate jumps significantly, because the sensitivity is usually about revealing size, not about admitting improvement.
Tier 3: Named client, no metrics, detailed method. "Here's exactly what we did for Masai School." Approval is usually trivial. Less persuasive on outcomes, but surprisingly effective with sophisticated buyers who care more about method than headline numbers.
Tier 4, Descriptive anonymisation with full metrics. "An Indian coding bootcamp with 200+ hiring partners grew organic social reach 4.5x in 14 months." The description does the credibility work. This is the workhorse tier and it's badly underused.
Tier 5: Category anonymisation. "A Series B edtech company." Weakest, but publishable when nothing else is.
Ladder down until you hit yes. Publishing at Tier 4 in three weeks beats publishing at Tier 1 in never.
Relative versus absolute metrics
Legal teams object to absolute numbers far more than relative ones, because absolutes reveal business scale: which touches competitive sensitivity, investor communications, and sometimes disclosure rules.
So lead with relative where you must:
- "Reduced cost per acquisition by 41%" instead of "reduced CPA from ₹4,200 to ₹2,478."
- "Grew organic pipeline 3.2x" instead of naming the pipeline value.
- "Improved conversion from 2.1% to 4.7%", percentages of percentages are often approved even when revenue figures aren't.
A useful trick: offer the client both versions in the approval request and let them choose. People approve things faster when the decision is a selection rather than an edit.
The approval request email
Send this to your champion, not to legal. Your champion navigates internally far better than you can.
Subject: Case study draft, need your read before anything goes anywhere Hi [Name], I've drafted a short case study on the work we did together. Before it goes anywhere near publication, I want you to have full control over it. Attached is the draft. Three things worth knowing: 1. Nothing publishes without your written sign-off. If you'd rather we didn't publish at all, that's a completely fine answer and it changes nothing between us. 2. The metrics are flexible. I've included two versions of the results section: one with absolute numbers, one with relative percentages only. If the absolute figures are sensitive, we use the relative version. If both are sensitive, we can describe the work without metrics, or describe your company without naming it. 3. Edit anything. Track changes, rewrite sections, cut whatever you like. Your quote in particular: I've drafted something, but please replace it with whatever you'd actually say. If it's useful, I'm happy to send a version formatted for your internal comms or legal team, and I can join a call to answer questions directly. No rush on timing, let me know what works. [You]
Why it works: it removes risk before it's raised, it offers pre-built fallbacks so the answer is "which version" rather than "yes or no," and it gives an explicit permission to decline, which paradoxically makes people far more likely to agree.
When they still say no
Ask for one of three smaller things instead: a two-line quote for your site, a LinkedIn recommendation for you personally, or permission to reference the work verbally in sales calls without publishing it. All three are usually yes, and any of them is better than nothing.
Part Two: How to Make a Case Study AI-Citable
Here's the angle nobody in this SERP has touched, and it follows directly from how buyers now research.
McKinsey's 2026 B2B research found that roughly 68% of buyers now use LLMs during vendor research, and around 74% use them to verify vendor claims. So the question a buyer actually types is not "case studies." It's something like: "Which agencies have grown organic social for an Indian edtech company, and what results did they get?"
Your case study either surfaces in that answer or it doesn't exist.
The structural requirements
Publish it as an HTML page, not a PDF. A gated PDF case study is invisible to every system doing the answering. This one change matters more than everything else on this list.
Open with a self-contained direct answer. The first paragraph must state, without any surrounding context, who the client was, what the problem was, what you did, and what the result was. It should be liftable as a complete answer to the buyer's question. If your first paragraph is scene-setting prose, rewrite it.
Name the specifics. Industry, country, company stage, team size, timeframe, channels, tools. A model matching "a company like us" needs attributes to match on. "A leading company in the education space" matches nothing.
Date everything. "Between March 2023 and May 2024." Recency is a ranking factor in synthesis, and undated results read as unverifiable.
Use real HTML tables for before/after data. Not screenshots. Not images of charts. A table is machine-readable; an image of a table is not.
State the applicability explicitly. Add a section that says, in plain language, "This approach applies to organisations that have X, Y, and Z: and is a poor fit if you have W." That sentence is what lets a system correctly match you to a buyer's situation, and it's also the most honest and disarming thing you can put on a sales page.
Include the method, not just the outcome. Outcome-only case studies read as claims. Method-plus-outcome reads as evidence, and gets summarised more favourably.
Get it corroborated off-domain. The client posting about the results on LinkedIn, a press mention, a podcast where the number is stated aloud, each of these creates an independent record that supports your claim during verification.
The question-title move
Title and structure the page around the question a buyer would ask, not around your client's name. "How Masai School Grew Instagram from 26K to 117K" beats "Masai School Case Study" on every dimension: it's more retrievable, more clickable, and more useful as a standalone answer.
Part Three: The Worked Example, Deconstructing the Masai Case Study
Now the useful bit. Here's a real case study, followed by an explanation of every decision in it.
The case study
How Masai School grew Instagram from 26K to 117K and LinkedIn from 50K to 160K through organic social.
Masai School is an Indian coding bootcamp operating an income-share model, which means its business depends on two audiences simultaneously: prospective learners deciding whether to commit a year of their lives, and hiring partners deciding whether to interview those learners. Both audiences research on social platforms before they ever visit the website. When I took over organic social, Instagram sat at 26,000 followers and LinkedIn at 50,000, with content that looked like most edtech content: announcements, motivational quotes, and course promotions.
Over the engagement, Instagram grew from 26,000 to 117,000 followers and LinkedIn from 50,000 to 160,000, entirely through organic content with no paid amplification of follower growth.
What we changed:
- Split the content strategy by audience and platform. LinkedIn became the hiring-partner and outcomes channel: placement data, learner stories with named employers, and posts about the hiring problem that talent leaders were publicly discussing. Instagram became the learner channel, the reality of a career switch, day-in-the-life, and the objections nobody was addressing.
- Made outcome data the core asset. Placement outcomes were the one thing Masai had that no competitor could replicate. Instead of treating them as a bottom-of-funnel proof point, we made them the recurring content spine.
- Moved from brand voice to human voices. Learners, instructors, and team members on camera and in first person, rather than a brand account speaking in the third person.
- Addressed objections publicly. The hardest questions, the income-share terms, the failure cases, whether it's worth it, got direct content instead of avoidance. This was the single biggest change in engagement quality.
- Built a repeatable production system. A weekly cadence tied to cohort milestones, so content production didn't depend on inspiration.
What this applies to: organisations with genuine proprietary outcome data, a multi-sided audience, and the willingness to publish specifics rather than generalities. It applies poorly to companies with no differentiated outcome data and no appetite for publishing real numbers.
Timeframe: the growth described occurred over the course of the engagement, measured from platform analytics at start and end.
Now the deconstruction
Every choice above was deliberate. Here's why.
The opening paragraph is a complete answer. Read it alone, with no context. It tells you the client, the country, the business model, the two audiences, the starting numbers, and the ending numbers. If a system lifts only that paragraph, the case study still works. Most case studies open with "In today's competitive landscape…" which lifts as nothing.
The specifics are named and matchable. Indian. Coding bootcamp. Income-share model. Instagram and LinkedIn. Organic. Two-sided audience. Each of those is an attribute a buyer's situation can match against. Compare with the generic version, "a leading education company grew its social presence significantly", which matches nothing and persuades nobody.
The metrics are Tier 1 absolute. 26K to 117K, 50K to 160K. These are absolute numbers on a named client, which is the hardest tier to get approved: but follower counts are publicly visible, which is exactly why this tier was achievable here. That's a general principle worth stealing: push for absolute numbers on metrics that are already public. Nobody's legal team objects to disclosing something anyone can see by visiting the profile.
The scope is stated honestly. "Entirely through organic content with no paid amplification of follower growth." This is a scoping statement, and it does two jobs. It makes the result more impressive, and it preempts the obvious sceptical question. Every credible case study should anticipate its own biggest objection in the results paragraph.
The method section is specific enough to be stolen. Five concrete changes, each of which a reader could implement. Case studies that hide the method to protect the secret sauce convert worse, because buyers correctly interpret vagueness as either weak results or weak understanding.
The applicability statement is a filter, not a pitch. "It applies poorly to companies with no differentiated outcome data." That sentence costs me some leads and wins me better ones. It also does the AI-citability job: it tells a system exactly which buyers to surface this for.
What's deliberately absent: revenue figures, cost per acquisition, and enrolment numbers. Those are commercially sensitive and were never going to be approved. The case study is stronger for not straining to include a number I'd have to hedge.
The tie-in to attribution: follower growth is a legitimate, verifiable, publicly checkable metric. Claiming it drove a specific revenue number would require a causal claim I could not defend, which brings us to the third problem.
Part Four: Case Studies for Services, Where Attribution Is Murky
Product companies have it easy. Deploy the product, measure the metric, attribute the change. Services businesses, agencies, consultants, fractional operators, have a genuine epistemological problem, and most respond by either overclaiming or saying nothing.
The overclaim trap
"We grew their revenue 340%" is a claim you cannot defend if the client also raised a funding round, hired a sales team, launched a product, and rode a market tailwind in the same period. A sophisticated buyer knows this: and Forrester has reported that a substantial share of technical B2B research now runs through conversational AI, where an inflated claim gets sanity-checked in seconds. When they ask an assistant to sanity-check it, they'll get a sceptical answer. Your credibility drops below where it would have been with a more modest claim.
The four honest patterns
1. Claim the metric you actually controlled. I claim follower growth and engagement, not enrolments. Those are the metrics my work directly produced. It's a narrower claim and a defensible one, and defensible claims survive verification.
2. Publish contribution, not causation. "During this period, organic social grew from 8% to 31% of top-of-funnel traffic" is a contribution statement. It's true, it's measurable, and it doesn't require you to prove you caused the revenue.
3. Show the method in enough detail that the causation is self-evident. If you describe exactly what you did and what changed immediately afterwards, a reader draws the causal conclusion themselves. That conclusion is far more durable than one you asserted.
4. Name the confounders yourself. "The company also expanded its hiring partner network during this period, which contributed to reach." Naming your own confounders is counterintuitive and enormously effective. It signals that you understand measurement, which is precisely the competence a buyer is trying to assess.
The comparative structure for agencies
When a client's business has too many moving parts, compare channels rather than periods. "Across the six months, organic social delivered X while paid delivered Y at Z cost" isolates your contribution without requiring you to own the whole business outcome.
What to do when the engagement partly failed
Publish it, carefully. A case study that says "the first approach didn't work, here's what we learned and what we changed" is the most trust-generating asset a services business can produce. It's also the rarest, which is why it stands out. Get client approval for the honest version: champions often approve these more readily than glossy ones, because the honesty protects them internally too.
Distribution: Where Case Studies Actually Get Used
Writing it is half the job.
Ungate it. A gated case study collects a handful of emails and forfeits all retrieval, citation, and internal-forwarding value. The trade is not close.
Give sales the excerpt, not the link. A rep should be able to paste 150 words that stand alone into an email. If your case study only works as a full read, it won't get used.
Publish one page per question, not one per client. A single client engagement can support three case study pages if it answered three different buyer questions. Structure by question.
Get the client to post it. A client sharing their own results on LinkedIn is corroboration, distribution, and social proof in one action. Ask explicitly, and make it easy by drafting a version they can edit.
Refresh annually. Search Engine Land has covered content freshness signals for years, and case studies are among the worst offenders. Add a "as of [date]" line and update it. A three-year-old undated case study reads as a company that stopped getting results in 2023.
Frequently Asked Questions
How long should a B2B case study be? Eight hundred to fifteen hundred words for the main page, with the complete answer in the first paragraph. Long enough to include the method, short enough that a buyer finishes it.
What if the client won't let me use their name? Use descriptive anonymisation with full metrics: "an Indian coding bootcamp with 200+ hiring partners." The description carries most of the credibility, and approval is far easier.
How do I get legal approval faster? Put a case study participation clause in the SOW at the start, send the request to your champion rather than legal, offer pre-built absolute and relative versions, and give explicit permission to decline.
Should case studies be gated? No. Gating a case study destroys its retrieval, citation, and internal-forwarding value in exchange for a handful of email addresses. Gate a diagnostic or a benchmark instead.
How many case studies do I need? Three good ones covering your main buyer situations beats fifteen thin ones. Each should answer a distinct buyer question and name a distinct situation type.
What metrics work best in a case study? Metrics you directly controlled, that are verifiable, and ideally publicly checkable. Publicly visible metrics, follower counts, review scores, published pricing, are both the most credible and the easiest to get approved.
How do I write a case study when attribution is genuinely unclear? Claim contribution rather than causation, name the confounders yourself, and describe the method in enough detail that the reader draws their own conclusion. Modest defensible claims outperform bold indefensible ones in a market where buyers verify.
Can I write a case study without any numbers? Yes: a detailed method case study with a strong client quote works, particularly with sophisticated buyers. It converts less well than a metric version, but it's vastly better than nothing and takes about a week to approve.
How do I make a case study show up when someone asks an AI assistant for recommendations? Publish it as an HTML page, open with a self-contained direct answer, name specific attributes and dates, use real tables, state explicitly which situations it applies to, and get the numbers corroborated in off-domain sources.
How often should I publish new case studies? One per quarter is a healthy cadence for a small team. Consistency matters more than volume, and refreshing existing ones with current dates is often higher-value than writing a new one.
Should I include the client's logo? If approved, yes, but a named client with an approved number in the text does far more work than a logo wall. Logo walls have almost no persuasive value in 2026 because everyone has one.
What's the biggest mistake in B2B case studies? Writing them for your website's customers page instead of for a specific buyer question. That single reframing improves the structure, the title, the opening paragraph, and the retrievability all at once.
If you're sitting on results you've never turned into publishable proof, or on case studies nobody can find, that's the gap I work in: organic growth for Indian edtech and startup brands, built on evidence that survives verification. More on how I approach it at younusfardeen.com.