Case Studies
Turned a Week of Manual Scheduling Into a Review
Plain English: A scheduler was spending 4 days per month making a manual Excel-based calendar for 100+ client visits. We built a simple, web-based tool the scheduler runs instead.
Tech Specs: We built a deterministic constraint solver first, covering drive radius, cadence, and PTO. Then, we layered in an LLM. The result is a browser tool with a drag-to-reschedule calendar.
400+ hours saved a year · about $26,000 at a $65/hr loaded cost
Built a Teams Bot for Accurate Sales Quoting
Plain English: An insurance brokerage relied on a PDF library, OneNote files, and unwritten rules to quote new business. We built a Teams bot that answers team member questions accurately, while citing its sources. The brokerage owns the system and its rule base, with no lock-in to us.
Tech Specs: We built a deterministic extraction pipeline that created over 7,000 rules surfaced from natural language interactions with a database.
7,000+ rules extracted · every answer cites its source
Made AI Spend Visible in Real Time
Plain English: A hung process quietly racked up thousands of dollars in AI API charges overnight before anyone noticed. With 18 AI services sharing a handful of API keys, finding the culprit meant hours of log spelunking. This is one of our own portfolio companies.
Tech Specs: We routed every AI call through a lightweight gateway that tags each request with the calling service. A cost spike is now a filtered dashboard view, not a log hunt.
Investigation time cut from 2 hours to under 2 minutes
Built 14 Watchdogs So Problems Get Caught Before Customers Do
Plain English: A small team running a legacy platform was purely reactive: problems surfaced only when a customer emailed angry. Underpayments and stuck jobs went unnoticed for days. This is one of our own portfolio companies.
Tech Specs: We built automated checks on cadences from 10 minutes to daily, each rated by severity, some able to self-repair. The rest raise an alert with full context so a human fixes only what actually needs a human.
14 automated checks running continuously · issues caught before the customer notices
Replaced a 20-Year Legacy Platform Without Losing a Customer
Plain English: A 15-year-old platform run by one irreplaceable engineer was shrinking as customers left for competitors. Outside estimates put a rebuild at $250,000 and 2 to 3 years. This is one of our own portfolio companies.
Tech Specs: We rebuilt in parallel, ran a beta with key accounts, then cut over every customer and shut off the old servers. AI-assisted development got two non-engineers shipping code within months.
Growth flipped from -15%/year to +50%/year · maintenance down to 2 hours/week
Turned Bug Reports Into Reviewable Fixes While the Team Slept
Plain English: A four-person team was stuck waiting on itself: a customer bug meant a support agent filed a ticket, then an engineer had to drop everything to reproduce, diagnose, and fix it, often over days. This is one of our own portfolio companies.
Tech Specs: We built a chain of agents that investigates a Linear ticket, proposes a fix, and opens a pull request, all gated by a human review before anything merges or deploys.
Turnaround cut from days to as little as 5 to 15 minutes before a human reviews
Turned a 10-Hour Sourcing Grind Into a 10-Minute Review
Plain English: One person was manually scanning six platforms for opportunities, vetting each one, and getting it in front of the right audience. This is one of our own portfolio companies.
Tech Specs: We built a scraper and extraction pipeline that cleans and structures each listing, resolves location with a confidence score, and drops anything uncertain into a one-at-a-time human review queue.
Review time down from 10 hours a week to about 10 minutes
Went From a 20-Deal Ceiling to Capacity for 250+
Plain English: Before a partner picked up the phone on a deal, someone had to read CIMs, teasers, and financial workbooks by hand and check them against the buy box. That capped the team at roughly 20 deals in flight at once. This is one of our own portfolio companies.
Tech Specs: We built a research agent that pulls every attachment off a Linear deal, scores it against a 12-line buy box, and posts a PASS / CONTINUE / needs-a-call memo with citations for a partner to review.
First-pass memo in about 15 minutes, up to an hour for a careful read · old cap was ~20 deals in flight, capacity now for 250+, though we haven't pushed that far yet
Who We Serve
- We partner best with firms with 30+ employees and more than $5 million in revenue.
- We work with businesses who see opportunities to deploy AI for good in their organization, no matter what stage of AI deployment they are in.
- If you are a non-profit looking for low bono or pro bono work, please contact us.
We are not a fit for sub-$5M firms still testing whether AI is real, or for teams that want AI to replace their people.
What We Believe
- AI should support people, not replace them. We build tools that help people do more work that matters. If a project's goal is cutting jobs, we're not the right firm.
- Effective AI needs shared buy-in. The most effective projects create shared excitement on your team.
- People in the loop. AI can save your team time, but important decisions remain theirs.
- Ownership stays with you. From training to data, we believe core competencies and operations should live with your organization, not ours.
- AI deployment must be thoughtful. Not all projects are a good fit for AI. We focus on the ones that are.
Where to Start
Most organizations move through four stages, and we meet you at whichever one you're in.
- Stage 1: Thought partner. Use AI to ask questions, not connected to any systems. Someone pastes data into ChatGPT and copies the answer back by hand.
- Stage 2: Assistant. Tools are wired into your internal systems and data, reading your source documents and learning your organization.
- Stage 3: Teammate. AI runs workflows on its own, reviewed by your team at checkpoints you define.
- Stage 4: System. High-impact, complex workflows run end to end and improve over time. Rare, and usually the last stage we reach with a client, not the first.
We don't charge for getting to know you. That would be ridiculous.
Who's Behind This
Colin Burns, public markets and capital allocation.
Justin Duke, software engineering and SaaS.
Myles Marino, early-stage operations.
Harrison Roday, business operations, finance, and deal execution.