Case studies
Finding leverage in how a business already runs
Case studies in growth, GTM, and operations — using data and AI to find what's slowing a business down and fix it. Details are shown by industry rather than by name to protect confidentiality; the situations, builds, and results are real.
Filter case studies by category
Independent Build — AI Recommendation Tool
Shifting the Burden of Product Knowledge From the Customer to the Tool
- AI / Productivity
- Data & Research
- The situation
- I've surfed my whole life, and choosing fins still overwhelms me. So I buy the standard set and move on, and so does everyone I surf with. It's a pattern you see with any product where the specs outrun the shopper.
- The problem
- If the knowledge burden sits on the customer, then people will default to the safest choice. That's the gap: between what I know and what the page expects me to know. FCS has great information on fins, but it's static. It expects you to understand every material and use case, then match that to your own board, waves and style.
- What I built
- SkegJev, a guided fin finder built on FCS's own product data and Jev. You describe your board, waves and what you want in plain words, the way you'd tell an expert at a surf shop. It narrows down the fins as you type and explains every pick in plain English. The same approach would work for any knowledge- or spec-based product: cabinets, shoes, courses and more.
- The result
Anyone can now find products that fit them without knowing anything about the materials and nuances. A few plain details about themselves and how they'll use it can guide them to the best-fitting product. In my case, that pushed me to a performance-level fin I would likely have skipped for the safe standard one.
Public Tool — MCP Server
Turning Every AI Video Summary Into Something You Can Actually Verify
- Data & Research
- Workflow Automation
- The situation
- Between research, tutorials, and building out case studies like this one, I'm constantly pulling material from YouTube — and a quick AI summary isn't enough when the actual wording is what matters.
- The problem
- Research and tutorials sometimes need the exact transcript — the precise quote, the exact step — not an AI paraphrase. But AI assistants can't provide a full transcript because of copyright restrictions, leaving manual copy-and-paste as the only workaround.
- What I built
- ClipScript, an MCP server (Python SDK + youtube-transcript-api, with yt-dlp for chapters) that gives Claude, ChatGPT, and Gemini the real transcript behind any YouTube video — chapter-aware, timestamped, and fully quotable, not just paraphrased.
- The result
No more manually copying and cleaning captions before I can work with them — every summary now comes with the real transcript behind it, exact quotes included. Available on GitHub right now.
Marine Infrastructure
Turning 36 U.S. Districts and Territories Into an Always-On Project Pipeline
- Workflow Automation
- Data & Research
- Growth
- The situation
- A large-scale, global concrete floating dock manufacturer serving the superyacht market relied on four people to monitor public project sources every week.
- The problem
- The company believed relevant projects had been missed, worried the same leads could reach competitors, and had no direct control over coverage, timing, or qualification.
- What I built
- BidSig, a configurable lead-intelligence platform with individual and grouped scans, scheduling, duplicate prevention, Airtable, Apify-powered enrichment, JEV scoring, human review, and HubSpot handoff.
- The result
Baseline coverage is complete across 36 selected U.S. districts and territories. The system now maintains the pipeline incrementally by processing only newly discovered projects.
Independent Build — Content Production Tooling
Cutting Content Tooling Spend to Zero: A Custom-Built Video Pipeline
- Workflow Automation
- Content
- Operations
- The situation
- Producing polished video content from a mix of motion graphics and real screen capture typically means paying for two separate categories of software: a high-resolution screen recorder and a video editor to turn raw footage into publish-ready files.
- The problem
- Two source types, one output requirement: motion graphics built in HTML/CSS and real screen footage both needed to become clean, correctly-sized MP4 files, but no single existing tool handled both.
- What I built
- MP4K, an Electron-based desktop tool that renders HTML/CSS animations directly to production-ready H.264 MP4 and captures, trims, crops, and narrates high-resolution screen recordings in place.
- The result
Eliminated the two tool categories this workflow required, avoiding roughly $300+ per person, per year.
Independent Build - Productivity / AI Workflow
Replacing Typing With a Voice-First Workflow
- Workflow Automation
- Operations
- AI / Productivity
- The situation
- I was constantly bouncing between Codex, Claude, browsers, notes, messages, and other applications, and typing was becoming the bottleneck.
- The problem
- Built-in dictation helped, but every application handled voice differently. My vocabulary, recurring phrases, tone, and corrections didn't carry with me, so I was still stopping to fix transcripts and clean up what I had just said.
- What I built
- TypeNull, a native Mac voice-writing layer that lets me dictate into whatever application I'm already using, cleans up filler and false starts without rewriting my voice, remembers my vocabulary and snippets, and can edit existing text by voice.
- The result
37,825 words dictated across 23 recorded days for about $0.19 in estimated AI usage - and the workflow has evolved far enough that, with a Bluetooth mic and repurposed controller, I no longer need to stay tied to my keyboard or desk.
Internal Tooling
Catching an AI Usage Problem Before It Became a Habit
- Workflow Automation
- Data & Research
- The situation
- A friend who oversees three teams — sales, marketing, and operations, 15 people total — had just rolled out both GPT and Claude across the group.
- The problem
- Tracking usage meant logging into two separate platforms with no way to correlate them, no per-person visibility, and no early warning before someone blew through a cap.
- What I built
- A single live dashboard pulling usage data directly from OpenAI's and Anthropic's official APIs, filterable by role, team, or name, styled with a custom wave-progress visual system.
- The result
In week one, the dashboard caught a new team member blowing through their cap — leading to a fix that cut their usage for good, with a trend-tracking database already underway.
Enterprise Software
Building a Retention Function From Zero to 31% NRR Growth
- Retention
- Growth
- Leadership
- The situation
- A global software platform with 200,000+ customers across 190 countries had no dedicated retention function and was quietly leaking revenue.
- The problem
- Without anyone owning retention, churn wasn't being diagnosed or addressed — it was just absorbed as a cost of doing business.
- What I built
- Built a retention team from scratch — playbooks, an objection-handling framework, a commercial terms dictionary, and a redesigned customer lifecycle from onboarding through renewal.
- The result
A 31% increase in Net Revenue Retention within 9 months, with the team scaled to 17 people.
Enterprise SaaS
Turning Monday's Commitments Into Friday's Accountability
- Workflow Automation
- Growth
- The situation
- A Sales VP at a large SaaS organization ran a dozen-plus reps through a standard weekly rhythm — a Monday pipeline call to set commitments, a Friday check-out call to review them.
- The problem
- No one can actively listen, coach, and take accurate notes on a call at the same time. By Friday, the specifics had faded, and the check-out call had quietly turned into a status ritual instead of real accountability.
- What I built
- Two scheduled agents bookending the week — one that pulls Monday's call transcript and logs every rep's commitments, one that resurfaces that exact list Friday morning with coaching prompts already attached.
- The result
Friday stopped being a status update and became a real coaching conversation — no new software, no retraining, just closing a memory gap.
Content Skills
The Middle Step Nobody Wants to Do
- Skill Creation
- Content
- The situation
- Personal projects kept generating real insights I wanted to turn into LinkedIn posts, but getting from a rambling brain dump to something worth publishing took real time.
- The problem
- Handing it to a generic AI tool didn't save that time — I ended up re-explaining my tone and rewriting whatever it handed back, every single time.
- What I built
- A skill trained on a person's real writing that organizes, structures, and formats a brain dump, then writes it back in their voice — the same process every time, but a fresh build for each person.
- The result
Built for myself, then deployed for 5 others — a mix of executives, leadership, and ICs. For one, per-idea processing time dropped from 30-35 minutes to about 3.
Enterprise Networking
Owning One Customer Relationship Across Three Roles, Three Years
- Retention
- Growth
- Leadership
- The situation
- An enterprise network-visibility product needed someone who could grow with the customer relationship as it matured from onboarding into a strategic account.
- The problem
- Most account handoffs lose context at every transition — the person managing renewal isn't the person who ran onboarding, and the customer feels it.
- What I built
- Owned the full relationship arc directly — Customer Success Manager, then Global Account Management, then Sales Manager — across three years on the same product line.
- The result
Continuity the customer could actually feel, instead of getting handed off between strangers at every stage of the relationship.
Investor Research
Finding the Investors Standard Directories Miss
- Data & Research
- Growth
- The situation
- A team looking for family office investors was relying on the same directories and lists everyone else already knows about.
- The problem
- The obvious sources are crowded — directories and conference lists don't surface the family offices that haven't shown up everywhere else yet.
- What I built
- Three parallel research pipelines — directory scraping, conference attendee lists, and SEC EDGAR filings — feeding one combined lead list.
- The result
A lead list built from sources most people never check, not just the ones everyone already has.
Web3 Infrastructure
Scaling a Company From 8 to 120 During a Leadership Transition
- Growth
- GTM
- Leadership
- The situation
- An infrastructure company lost its CEO mid-growth, leaving sales, marketing, customer success, and demand-gen without a clear owner.
- The problem
- A company can't grow through a leadership gap — someone has to own the whole commercial engine, not just their own function, or growth stalls.
- What I built
- Stepped in to run sales, marketing, customer success, and demand-gen as the de facto executive, including new cloud partnerships that generated co-sell revenue.
- The result
The company grew from 8 to 120 people with 500% ARR growth in a single year.
Financial Services
From Voice Memo to Marketing Board, Automatically
- Workflow Automation
- Content
- The situation
- A marketing executive at a mid-market financial services company had good ideas surfacing constantly — almost always while she was in motion, walking between meetings or mid-conversation.
- The problem
- The only way to act on an idea was to stop, sit down, and type it all out — 30-35 minutes per idea, 2-4 times a day, just to capture and log it.
- What I built
- An agent that takes a quick voice-to-text drop in Slack, makes sense of the raw brain dump, and logs it directly into the team's Notion board in the correct tone, structure, and terminology — no sitting down required.
- The result
Reclaims 4+ hours a week that used to go to manually processing ideas, with per-idea time dropped from 30-35 minutes to about 3.
Decision Support
A Five-Person Advisory Board for Solo Founders
- Skill Creation
- Leadership
- The situation
- A solo founder making a high-stakes call has no exec team in the room to pressure-test the decision before it ships.
- The problem
- Big decisions made alone, without pushback from different angles, are exactly the decisions most likely to have a blind spot nobody catches until it's too late.
- What I built
- A skill that spins up five expert personas in parallel to weigh in on any decision or piece of content, then a "chairman" synthesizes it into one clear recommendation.
- The result
A founder gets a boardroom's worth of perspective on a decision, without needing to hire a boardroom.
Content Skills
Catching AI Tells Before Your Audience Does
- Skill Creation
- Content
- The situation
- People using AI to help draft content had no way to know how "AI" their finished writing actually sounded before publishing it.
- The problem
- Content that reads as obviously AI-generated undermines trust with an audience, even when the substance is solid.
- What I built
- A diagnostic skill that scans a piece of writing, flags the specific "AI tells," and returns a letter grade — no automatic rewriting, just a clear read.
- The result
Writers can check their own work before it goes out, instead of finding out from a comment section.
Content Skills
Turning Corporate Copy Into Something a Person Would Say
- Skill Creation
- Content
- The situation
- Content coming back from AI tools often reads technically correct but stiff — too corporate, too generic, not like an actual person wrote it.
- The problem
- Technically accurate writing that sounds like nobody in particular doesn't build trust or connection with a reader.
- What I built
- A skill purpose-built to take overly technical or corporate-sounding content and rewrite it to sound like an actual human said it.
- The result
Content that reads like it came from a person, not a policy document.
Meta / Process
Why a Personal Skill Library Compounds Over Time
- Skill Creation
- The situation
- Every new task meant re-explaining context, tone, and process to an AI tool from a standing start.
- The problem
- Re-explaining yourself every single time is a hidden tax on every AI-assisted task — it adds up fast and never gets easier.
- What I built
- A personal library of reusable skills — for voice, decision-making, content auditing, and more — each one teaching the tool something once instead of every time.
- The result
A toolkit that compounds: each new skill makes the next task faster, not just this one.
Field Services
From Field Notes to Branded Estimates in Minutes
- Workflow Automation
- Operations
- The situation
- A fence installation company's reps were writing up every job by hand, translating field notes into a customer-facing estimate after the fact.
- The problem
- Turning raw field notes into a branded, accurate estimate took hours per job and varied by which rep wrote it — slow quotes lose jobs.
- What I built
- An AI extraction tool that reads field notes and generates a branded estimate automatically, plus a full job-flow CRM prototype covering the pipeline from lead to install.
- The result
Field notes now become a client-ready estimate in minutes instead of days — the estimate goes out while the interest is still hot.
Trade Services
An AI Assistant That Actually Knows the Business
- Skill Creation
- Operations
- The situation
- A welding and fabrication shop wanted an AI tool of their own but had no discovery process, brief, or instructions to hand off to it.
- The problem
- Without a documented process, any AI tool built for the shop would only be as good as whatever the shop happened to remember to explain that day.
- What I built
- A ready-to-use Claude Project: a discovery checklist, project instructions, and a website brief packaged so the shop's knowledge is captured once and reused every time.
- The result
The shop now has a standing AI assistant that already knows their business, instead of starting from zero in every new conversation.
Watersports / Rentals
Turning Ad-Hoc Pricing Into a Repeatable System
- Workflow Automation
- Operations
- The situation
- A watersports rental company priced inventory manually, without a system to track what should change and when.
- The problem
- Pricing decisions lived in someone's head, not in a tool — no consistent way to test, adjust, or defend a price.
- What I built
- A pricing management prototype with an AI assistant, "Casti," to help reason through pricing changes and keep them consistent.
- The result
Pricing decisions are on paper for the first time — a working prototype ready to move from testing to daily use.
Data Tools
Query Your Own Data Without Writing SQL
- Workflow Automation
- Data & Research
- The situation
- A team had useful data sitting in raw files, but getting an answer out of it meant asking someone who knew SQL — or waiting.
- The problem
- Simple questions about a team's own data required a technical middleman, which slows down every decision that needs an answer today.
- What I built
- A tool that lets anyone drop in a dataset and query it directly, in plain language, no SQL or spreadsheet wrangling required.
- The result
Anyone on the team can get an answer from the data themselves, the moment they need it.
Client Deliverables
Same-Day Leave-Behinds From Raw Meeting Notes
- Workflow Automation
- Operations
- The situation
- After a discovery call, whoever ran the meeting had to sit down and manually turn scratch notes into something client-ready.
- The problem
- Good discovery notes are useless to a client on their own — they need a polished, professional leave-behind, and building one by hand eats the time you just saved by having a good conversation.
- What I built
- A tool that turns raw meeting or discovery notes directly into a polished, client-ready leave-behind document.
- The result
The leave-behind goes out the same day as the meeting, while the conversation is still fresh for the client.
Music / Entertainment
Rebuilding Seranation's Website to Match Who They've Become
- Growth
- GTM
- The situation
- Seranation is a reggae-rock band that hit a level most acts never reach — a #1 hit on iTunes Reggae and a #5 spot on the Billboard charts. Their web presence hadn't caught up to their music.
- The problem
- Their site was really just their Shopify store with a few ticket links bolted on — outdated, hard to update, and not converting, especially for a band that had just landed a Billboard hit.
- What I built
- A full site rebuild after a discovery process covering their goals, revenue sources, and brand — built with Claude, Lovable, and custom tooling, connected to Bandsintown, Shopify, YouTube, Spotify, and their socials. Includes a custom direct-link tour-date system maintained through a Telegram-based AI agent workflow.
- The result
A site built around their actual top revenue drivers — merch, streams, and video — that updates in seconds over a Telegram message instead of requiring dashboard or dev work.
Yours would be the next one.
Send your details and I'll get back to you.