--- name: case-study description: Multi-brand case study writer. Takes working notes from a delivered engagement (Slack threads, kickoff doc, weekly recaps, final deliverable, raw bullets) and produces a publish-ready case study in the right shape for the brand it's about. Three shapes built in: EAA (AI consulting outcome story), Boxpress.io (SaaS-buyer story), FUEL (marketing services story). Each output includes the full case study, a headline-ready quote, a social snippet, and 3-5 sales-deck slide bullets. Brand voice is loaded from a documented voice doc so the case study reads in that brand's actual voice, not generic consultant-speak. trigger: /case-study --- # /case-study Take the messy notes from a delivered engagement and turn them into a publish-ready case study, sized and voiced for the brand it's about. Three brands built in. Each one has a different audience job and a different shape. This is for the moment between "the work is done" and "we have a sellable artifact." Most consultants leave case studies on the floor because the writing is the part nobody schedules. This skill turns 30 minutes of paste-and-edit into a finished asset. ## Usage `/case-study ` runs against the default brand (Elite AI Advantage). `/case-study --brand ` switches the brand shape and voice. `/case-study --brand --interview` runs in interview mode: the skill asks you 5-7 targeted questions one at a time to extract the case study from your head when you don't have written notes ready. `` can be: pasted Slack threads, a kickoff doc, weekly recap notes, raw bullet points, the final deliverable, or any combination. Length doesn't matter. The skill pulls signal from messy inputs. If the user types `/case-study` with no notes, default to `--interview` mode. If they specify a brand but no notes, same thing. ## What you must do when invoked ### Step 1, parse inputs - **Brand** (default `eaa`). Recognized: `eaa`, `boxpress`, `fuel`. Reject anything else with the three valid options listed. - **Mode**: notes mode (notes pasted) or interview mode (`--interview` or no notes). If both notes AND `--interview` are passed, run interview mode using the notes as priors. - **Notes**: everything after the flags. Strip leading/trailing whitespace. ### Step 2, load the brand voice Read the brand voice document. Try paths in order: 1. `content/brand-voices/.md` (relative to the current working directory; canonical when running inside the EAA repo) 2. `~/.claude/skills/case-study/voices/.md` (skill-local fallback) 3. `~/.claude/skills/voice-check/voices/.md` (shared with the voice-check skill, same docs) If no voice doc resolves, output: `Brand voice doc for "" not found. Drop it at content/brand-voices/.md or ~/.claude/skills/case-study/voices/.md and re-run.` Then stop. ### Step 3 (optional, interview mode only), extract from the user If running in interview mode, ask exactly these 7 questions, one at a time. Wait for the answer before asking the next. Skip a question if the answer is in the notes already. 1. **Who was the client?** (Real name OK; we'll anonymize in the published version. Industry, company size, and role of the person you worked with.) 2. **What was the situation when they came in?** (The actual problem in their language, not yours. What was bleeding? What did the calendar / inbox / P&L look like?) 3. **What did you build / ship / deliver?** (The literal thing. A workflow, a tool, a campaign, a piece of software.) 4. **How long did it take? What did the engagement cost?** (Timeframe + dollar range. The dollar number is for the sales-deck slide, not the public case study.) 5. **What's the measurable outcome?** (Numbers, not adjectives. Hours saved, revenue moved, deals closed, cost cut, conversion lifted, traffic up. If you don't have a number for something, say so.) 6. **What's the one quote you'd hang on the wall?** (From the client, or from you. If neither, paraphrase a Slack message.) 7. **What's the one thing you'd repeat for the next client like this one?** (The pattern that worked. The framework, the tool, the call you made.) After question 7, generate the case study. Don't ask follow-ups; if signal is missing, mark the field as "Detail TBD: " in the output and let the user fill it in. ### Step 4, generate the case study in the brand's shape Each brand has a fixed shape. The shape is non-negotiable; it's the format the brand's audience expects. #### EAA shape, AI consulting outcome story Audience: AI-skeptical SMB owners with budget; mid-market marketing/ops directors; occasional enterprise. They want proof that AI consulting produces a defensible outcome they could explain to a CFO. ``` # ## The situation 2-4 short paragraphs. Anonymized client (industry, size, role of the person who hired you). The literal pain in their language: hours bleeding, deals slipping, vendor stack costing too much, team drowning in repetitive work. One specific scenario the reader can picture. ## The work 2-4 short paragraphs. The literal thing you built. Name the tools (Claude, GPT, n8n, Zapier, custom Python, whatever it actually was). Name the workflow that changed. Specifics over abstractions. ## The outcome 3-6 bullets, one number per bullet: - [N] hours/week saved on [specific task] - [$N] in monthly cost cut by retiring [specific vendor or process] - [N]% lift in [specific KPI] within [timeframe] - [N] deals/leads/signups (whatever the business cares about) - [N]-day turnaround on [thing that used to take longer] ## What we'd repeat 1-2 paragraphs. The pattern that worked. The framework, the tool, the call you made. This is the part that compounds: it tells the next prospect "this is the kind of thinking they get from EAA." ## The one-line takeaway A single sentence that summarizes the case in a tweet-length form. ``` #### Boxpress shape, SaaS-buyer story Audience: cigar shop owners, brand owners, lounge operators. They want proof that the platform actually delivers, not that someone holds their hand. Different from EAA: the case study is about the platform doing the work, not about a consultant being smart. ``` # ## The operator 2-3 short paragraphs. Real operator type (B&M shop, brand, lounge, online), region, size (number of locations, employees, SKU count), tenure in the industry. Optional: which boxpress.io plan they're on. ## The setup 2-3 short paragraphs. What they were doing before. The actual problem: inventory drift, distributor reconciliation pain, member retention dropping, online-vs-in-store mix unbalanced, whatever. Industry-specific language is fine; the audience is operators. ## How they're using boxpress.io 3-5 bullets, each one a specific feature or workflow: - Use [specific feature] to [specific outcome] - Wired up [integration] to [system they already use] - Built [report/dashboard] their team checks every [day/week] ## The numbers 3-6 bullets, one number per bullet: - [N]% reduction in [inventory variance / shrinkage / reconciliation time] - [$N] recovered annually from [process] - [N] hours/week back for [role: owner / manager / counter staff] - [N] units/month sold-through that weren't tracked before ## What this unlocks 1-2 paragraphs. The next thing the operator can do BECAUSE the platform handled the basics. (Re-buy the right cigars, expand to a second location, run the lounge events without manual prep, etc.) ## The one-line takeaway Single sentence, operator-to-operator framing. ``` #### FUEL shape, marketing services story Audience: founders, ops people, CMOs at SMB or mid-market companies who buy marketing services. They want CAC, conversion, attributable revenue, and a story that doesn't read like an agency case study every other agency could write. ``` # ## The brief 2-3 short paragraphs. The client (industry, size, stage), what they wanted (specific goal: lower CAC, higher LTV, more demos, retention lift), what they'd already tried (and why it failed). ## The campaign 2-4 short paragraphs. The actual creative + channel mix. Named platforms (Meta, Google, LinkedIn, TikTok, podcast, whatever). Named messaging shifts. Specific A/B tests if relevant. No "we leveraged a holistic approach to the customer journey." ## The numbers 6-10 bullets, one number per bullet, hardest numbers first: - CAC: $X -> $Y (Z% drop) - Conversion rate: X% -> Y% on [specific page / channel] - Attributable revenue: $X in [timeframe] - Cost per [meaningful action]: $X - ROAS: Xx blended, Yx on [best-performing channel] - [N] qualified leads/demos/calls in [timeframe] ## What we'd skip next time 1-2 paragraphs. Honest. The thing that didn't work, the channel that wasn't worth it, the test that lost. Catty if appropriate. This is the FUEL signature: most agency case studies don't admit anything failed; this one names what didn't. ## The one-line takeaway Single sentence. Catty allowed. Honest required. ``` ### Step 5, output the supporting artifacts After the main case study, output these three additional pieces, in this order: ``` --- ## QUOTE FOR THE WALL A 1-2 sentence pull quote, attributed (real or paraphrased), suitable for a homepage / sales deck / proposal. No em-dashes. No emojis. ## SOCIAL SNIPPET A single 240-280 character post that summarizes the case in [brand]'s voice. Twitter/X-shaped (short, punchy, specific number in the first line). Use it as-is for X, expand for LinkedIn/IG. ## SALES-DECK SLIDE BULLETS 3-5 bullets fit for a customer-facing deck. Each one stands alone (the prospect skims, doesn't read in order). Lead with the strongest number. ``` ### Step 6, voice rules The whole case study runs through the brand voice doc. Specifically: - **No em-dashes** (the global rule, enforced in every brand voice doc). - **No emojis** in the case study body. Allowed in the social snippet only if the brand voice explicitly allows them. - **No "Excited to share," "We're proud to announce," or "thrilled."** Open with the substance. - **No "leveraged," "synergy," "holistic," "ecosystem," "drive impact," "10x," "game-changer."** - **Quotes inside the case study** should sound like real humans wrote them. Slack-message-shaped, not press-release-shaped. ("My ops lead got her Fridays back" beats "Our operational efficiency improved markedly.") - **Numbers everywhere.** If a section has zero numbers, rewrite it or cut it. Abstract claims have no shelf life. - **Anonymized but specific.** "A 12-person SaaS team" is anonymous. "A growing technology company" is generic and useless. The reader should be able to picture the client. ### Step 7, what to skip - **Don't write the case study at the wrong length.** EAA case studies are 600-1,000 words. Boxpress is 500-900. FUEL is 600-1,200. If the notes are too thin to hit the lower bound honestly, the output is shorter and the user gets a "Detail TBD" footer naming what's missing. - **Don't invent numbers.** If the notes say "saved a lot of time," the case study says "Detail TBD: hours saved per week" in the bullet, not "saved 30 hours/week." - **Don't rephrase a Slack message into something it didn't say.** Quotes preserve the original tone; clean only what's actually broken. - **Don't write the conclusion.** End on the takeaway sentence + the three supporting artifacts. No "and that's how..." closer. ## Hard rules for this skill itself - **No em-dashes in your output.** Every brand voice doc forbids them; the skill enforces them. - **No emojis in your output** (except in the social snippet if the brand voice explicitly allows them, which the EAA / Boxpress / FUEL docs currently do not). - **No preamble.** Start with the headline. End on the sales-deck bullets. - **Don't skip sections.** Every shape requires every section. If a section is genuinely missing data, write "Detail TBD: " inline so the user knows what to fill in. - **No "Note:" / "Disclaimer:" / "Caveat:" footers.** The case study is what it is. ## Calibration notes The user is Jake McCluskey. He runs three brands: - **Elite AI Advantage** (eliteaiadvantage.com): AI consulting for SMB and mid-market. Solo operator. Premium positioning. Audience is AI-skeptical buyers with budget. - **Boxpress.io**: SaaS for the cigar industry (B&M shop owners, brand owners, lounges). Operator-focused. Audience is busy retail/wholesale operators who don't have time for vendor BS. - **FUEL UP Your Marketing**: marketing services partnership. Audience is founders/ops/CMOs tired of agency hype. Voice is honest + funny + catty when the industry deserves it. Each brand's voice doc is the canonical source. The shapes above are the structure; the doc is the voice. Both layers apply. ## Example shape (abbreviated, EAA brand) User: `/case-study --brand eaa A SaaS founder with 12 people. Their ops lead was spending 10 hours/week on data entry between Stripe, HubSpot, and Notion. Built a Claude + n8n flow that ingests Stripe webhooks and writes the right rows into HubSpot and Notion. 3 hours of setup. Now her ops lead is back to ~30 min/week of monitoring. Took ~2 weeks total. We were paid $8K. They renewed the retainer at $4K/mo for 6 months.` Output: ``` # How a 12-person SaaS team got 9.5 hours a week back without hiring ## The situation A 12-person SaaS company (post-seed, ~$1.2M ARR) had one operations lead handling all post-sale data flow between Stripe, HubSpot, and Notion. Every new customer signup meant 15 minutes of manual work: copy the Stripe receipt, paste the relevant fields into HubSpot, set up the customer's Notion workspace from a template, kick off the welcome email sequence. By the time the company hit ~150 active customers, that 15 minutes had compounded into 10 hours a week. Her Fridays were gone. The founder kept getting Slack messages that read "are we still doing the spreadsheet" because the manual workflow was ALSO drifting (typos, missed Notion setups, a couple of welcome emails that never sent). ## The work Built a workflow in n8n that ingests every Stripe webhook (subscription.created, invoice.paid, customer.updated). Each event fires a Claude prompt that decides what to write where: which HubSpot fields, which Notion template, whether the welcome sequence is the new-customer or expansion shape, whether a Slack notification needs to go to the customer success channel. Total build: 3 hours of setup, including the Notion template structure. Total engagement: 2 weeks (most of week 2 was watching the workflow handle real traffic and tightening edge cases). ## The outcome - 9.5 hours/week saved on data entry (from 10 hours to ~30 min of monitoring) - 0 missed Notion workspace setups in 8 weeks of running (down from ~3/month before) - 100% of welcome emails firing on time (was missing ~5%) - $8K engagement paid back in under 6 weeks at the founder's modeled time cost - Renewed at a $4K/month retainer to extend the workflow into renewal + churn signals ## What we'd repeat The pattern that worked: don't let an LLM drive the workflow logic. Let it drive the field-level decisions inside a deterministic n8n graph. Claude decides "this customer is on the team plan with 4 seats so use the team-Notion template and tag them in the team channel." n8n decides what API call fires when. That split keeps the workflow reliable (n8n doesn't randomly fail) and the AI useful (Claude makes the judgment calls a hardcoded if-tree would get wrong). For the next 5 SaaS engagements like this, we'd start with that architecture. ## The one-line takeaway A 3-hour build paid back $8K in 6 weeks because we put Claude in charge of judgment, not orchestration. --- ## QUOTE FOR THE WALL "My ops lead got her Fridays back, and the founder stopped getting 'are we still doing the spreadsheet' messages." (Founder, 12-person SaaS) ## SOCIAL SNIPPET A 12-person SaaS team had one ops lead burning 10 hours a week on Stripe -> HubSpot -> Notion data entry. 3 hours of n8n + Claude. 9.5 hours back per week. The pattern: Claude makes judgment calls, n8n orchestrates. ## SALES-DECK SLIDE BULLETS - 9.5 hours/week saved per ops lead on cross-tool data flow - $8K engagement paid back in under 6 weeks at modeled time cost - 0 missed Notion workspace setups in 8 weeks of production - Renewed at $4K/month retainer to extend the same architecture - Pattern: Claude makes the judgment calls, n8n handles the orchestration ``` That's the shape. Headline, structured body, takeaway, then the three supporting artifacts. No fluff. No filler. Ready to publish.