Case Studies skill for Claude Code
Results with baselines, quotes the customer approved. Claude writes case studies that survive a skeptical reader. It runs an interview guide that digs for numbers, builds an evidence table with baselines, periods, sources, and approval status, picks the angle that matters to the target reader, drafts with a snapshot box and problem-approach-results structure, fact-checks every quote and metric, and manages customer sign-off before anything ships.
- name
case-studies - size182 lines · ≈1,900 tokens
- files1
- licenseMIT
- updated
What Claude sees before loading it
This description is the only part of the skill that sits in context all the time. Claude reads it to decide when to load the full playbook below.
description: Writes customer case studies and engineering case studies built on evidence - customer interviews with a question guide, before-and-after metrics with baselines, timeframes, and honest attribution, approved verbatim quotes, a problem-approach-results structure, a snapshot box for skimmers, engineering versions with constraints, alternatives, and lessons, plus customer approval and legal sign-off. Use when writing a customer success story, a technical case study of a migration or architecture change, a sales one-pager, or when a draft case study reads like an ad with vague claims and no numbers. Try asking
- Write a customer case study from this interview transcript and their usage metrics
- Turn our Kafka migration into an engineering case study other teams can learn from
- This case study draft sounds like an ad. Make it credible with real evidence
Use it when
- Writing a customer success story or sales one-pager
- Documenting a migration or architecture change as a technical case study
- Preparing interview questions for a customer story
- A draft has vague claims, no baselines, or unapproved quotes
Skip it when
- You want a narrative engineering post for the community — use dev-blog-posts
- You are writing a landing page around the results — use landing-page-copy
How to check the result
Every number in the draft maps to a row in the evidence table with a source and an approval, and every quote matches the transcript and has the speaker's sign-off.
The common mistake it prevents
Leading with product features and percentages without baselines or timeframes, which reads as marketing and convinces no one in the target audience.
A case study persuades when a skeptical reader in the same situation sees a specific problem they recognize, a credible account of what was done, and results with numbers, baselines, and a timeframe that the customer has approved. “Acme transformed its workflow and saw dramatic improvements” persuades no one.
Ground rules
- Evidence first. No draft until you have at least one metric with a baseline and a timeframe, or a concrete qualitative outcome you can state precisely.
- Quotes are real and approved. Never invent or “polish” a quote beyond light cleanup the speaker signs off on.
- Attribute honestly: state what else changed during the period and say “contributed to” when the product was one factor among several.
- The customer is the hero; the product is the tool. Lead with their problem, not your feature list.
- Get written approval from the customer (and legal or PR if either side requires it) before publishing names, logos, numbers, or quotes.
Two kinds
| Customer case study | Engineering case study | |
|---|---|---|
| Reader | Prospects in a similar role or industry | Engineers facing a similar technical problem |
| Goal | Show the outcome is achievable for someone like them | Share a transferable approach and its trade-offs |
| Core evidence | Business metrics, quotes, time to value | Technical metrics, architecture, alternatives considered |
| Tone | Plain, specific, confident | Candid about dead ends and costs |
For narrative engineering posts aimed at the community rather than a structured case,
see dev-blog-posts.
Workflow
- [ ] 1. Pick the story and confirm the customer will participate and approve
- [ ] 2. Collect data and run the interview(s)
- [ ] 3. Build the evidence table (metric, before, after, period, source)
- [ ] 4. Find the angle and headline
- [ ] 5. Draft with the standard structure
- [ ] 6. Fact-check every number and quote against sources
- [ ] 7. Customer and legal review; incorporate edits
- [ ] 8. Publish in formats; reuse pieces
1. Pick the story
Good candidates: a customer that matches the target audience, a clear before state, a measurable change, and a champion willing to talk. Confirm early what can be named (company, person, logo, numbers) and whether they need anonymization (“a top-10 US insurer”).
2. Interview
30-45 minutes, recorded with permission, two people if possible (one asks, one notes). Question guide:
- What does your team do, and what does success look like for you?
- Before: what were you doing, what was painful, what did it cost (time, money, risk)? Can you put a number on it?
- What triggered the search for a change? What else did you consider and why not?
- How did rollout go? What was harder than expected? How long until it was useful?
- After: what changed, in numbers if possible? What can you do now that you could not?
- What would you tell a peer considering the same change?
- Who else should we talk to, and what data can you share?
Ask follow-ups for specifics: “How many hours a week?”, “From what to what?”, “Over what period?”, “How do you measure that?”. The best quotes come from question 6 and from describing the old pain.
For engineering case studies, interview the engineers and pull dashboards, incident
records, cost reports, and design docs (see design-docs-rfc) for the facts.
3. Evidence table
| Metric | Before | After | Period | Source | Approved |
|---|---|---|---|---|---|
| Month-end close time | 9 days | 3 days | Q1 vs Q3 2026 | Customer finance team | Yes |
| p95 API latency | 820 ms | 140 ms | 30 days pre/post migration | Datadog dashboard | Yes |
| Infra cost | $48k/mo | $31k/mo | Jan vs Jun 2026 | Cloud billing export | Pending |
Rules: percentages always with absolute numbers somewhere (“cut close time 67%, from 9 days to 3”); state the period; note confounders (“headcount grew 20% in the same period”). Drop any metric the customer will not approve rather than blurring it.
4. Angle and headline
The angle is the one result most relevant to the target reader. Headline formula:
<Customer> <achieved specific result> <by/with how>.
| Weak | Strong |
|---|---|
| Acme Finance Success Story | Acme cut month-end close from 9 days to 3 by automating reconciliation |
| How Globex Uses Our Platform | Globex moved 2 PB to object storage with zero downtime and 35% lower cost |
5. Structure
Customer case study (600-1,200 words):
# <Headline with result>
| Company | <name, industry, size> |
| Challenge | <one line> |
| Solution | <one line> |
| Results | <2-3 metrics with numbers> |
## The challenge
<Their situation and pain, with a number and a quote in their words>
## Why they chose <product>
<Alternatives considered and the deciding factors, briefly>
## The approach
<What was implemented, timeline, who was involved, what was hard>
## The results
<Metrics from the evidence table, each with before/after/period; one quote>
## What's next
<Their next step; signals ongoing relationship>
Engineering case study (1,200-2,500 words): context and constraints (scale, SLOs, team size, deadline) -> the problem with data -> options considered with trade-offs -> what we did, with an architecture diagram -> results with measurements -> what went wrong and what we would do differently -> takeaways a reader can apply.
Writing rules:
- Specific nouns and numbers in every section; cut adjectives that a number could replace.
- Quotes do work the narrative cannot: emotion, credibility, peer advice. One per section at most, attributed with name and title.
- Show the hard part. A rollout with no friction reads as fiction.
- No superlatives (“best-in-class”, “revolutionary”) unless quoting the customer.
6. Fact-check
Re-verify every number against its source row, every quote against the transcript, names and titles against the customer’s current information, and product claims against what the product actually does today. Remove anything that cannot be sourced.
7. Approval
Send the customer a clean draft plus the evidence table, highlighting quotes and numbers needing approval. Expect edits; accept factual corrections, push back gently on removing all specifics (offer ranges or anonymization instead). Keep the written approval with the final version.
8. Formats and reuse
From one approved study: web page, PDF one-pager (snapshot box + results + one quote),
sales slide, social posts (one metric and one quote each; see social-media-posts),
and a short quote bank for landing pages (see landing-page-copy). Reuse only the
approved wording.
Gotchas
- Reporting the best week instead of a sustained period gets called out by informed readers and embarrasses the customer.
- “Up to 10x faster” claims without context invite disbelief; say what was measured.
- Interviewing only the buyer misses the hands-on story; talk to a user too.
- Customer approvals can take weeks; start the request at the interview stage.
- Anonymized case studies still need enough detail (industry, size, scale) to be credible.
Output contract
**Headline:** <customer + result + how>
**Snapshot:** company, challenge, solution, 2-3 results
**Evidence table:** metric, before, after, period, source, approval status
**Body:** challenge -> why chosen -> approach -> results -> what's next
(or engineering: context -> problem -> options -> solution -> results -> lessons)
**Quotes:** <n>, each approved, attributed (name, title)
**Approvals:** customer <date>, legal/PR <date or n/a> About this skill
Written and maintained by the Claude Skills editors to the same editorial standard as every skill in the directory: a selective trigger, a real workflow, known pitfalls and a verification step. First published , last updated . MIT licensed.