Results, not decks.

The legacy vendor was cutting off access with a take-it-or-leave-it exit and refused a data sample. We verified the export before the deadline, then rebuilt the whole library inside Salesforce, the CRM the sales team already lives in. It cleared the client's architecture review board, code review, and go-live on the first pass.

The friction

A reference library used by sales since 2017 lived on a vendor platform with access ending 12/31. The data came back as schema-less exports: 3,086 project lines, 38,737 tag lines, 1,884 image lines. And the old tool sat outside the CRM where sales actually works.

How we found the truth

Reviewed the exit agreement and deliverables list, validated the export before the deadline, and gave the agency a stats app to spot-check it. Wrote the high- and low-level design, signed off by the client's business owner.

What we built

A "Reference Finder" inside Salesforce: 1,999 project records, a 904-tag three-level hierarchy, 38,071 project-tag links, weighted search with relevance bars, PDF export, draft/publish approval and three permission tiers. Surfaced in four internal apps. Built with AI-accelerated engineering.

Results

Live in production in 3 weeks. 131 summer interns given edit access to grow the library. The client is funding next year's program.

A 17-year healthcare executive had a clear billing insight and no way to build it. There was no spec. We shaped the business model together, then built OCR bill reading, a scoring engine and dashboards, live in five weeks.

The friction

Deep domain expertise and a real market opening, but no startup team and no build capability.

How we found the truth

Worked alongside the founder to shape the business model and translate billing expertise into a workflow a machine could run.

What we built

OCR-driven bill extraction, a 0-100 scoring engine, browse and dashboard views, and in Phase 2 an admin panel and plain-English bill summaries. The founder queries live data through Claude, with no analyst or SQL.

Results

Zero to live MVP in five weeks, designed to scale with volume rather than headcount.

A lean, PE-backed genomics team had patient data in every system and inbox, and systems that don't talk to each other. We settled how AI can safely touch patient data first, then picked a pilot in two weeks. The first fix we recommended isn't AI at all.

The friction

Leadership wanted AI tools available to staff who work with patient data. The CRM and the clinical system don't sync, and the team is two developers and one admin.

How we found the truth

The CEO rejected our pre-built use-case list in discovery, which reset the scope. We mapped the COO's ranked list of 33 use cases to real systems, held working sessions with each leader, and walked the operations floor.

What we're delivering

A systems inventory, a patient-data approach, and a two-workflow pilot. Step one: fix the CRM triggers so the follow-up tasks nobody needs are never created. Step two: AI-drafted task summaries with a human in the loop.

Where it's headed

One workflow live with a before-and-after time measure, a readiness gap assessment and a 60-day plan by Day 30.

A stealth fintech was emailing its SAFE, pro forma and pitch video to investors. We built one controlled place for all of it: allowlisted access, documents watermarked to each investor, and a founder dashboard for invites. Two security audits before launch.

The friction

Sensitive fundraising materials were going out by email and file shares while the company was in stealth, with no control over who saw what.

How we found the truth

Four recorded discovery calls, a founder homework doc covering audience and legal minimums, wireframes, and a line-by-line proposal revision.

What we built

A public access gate with an approval queue, an allowlisted investor portal with signed sessions, embedded video, podcast and document viewer, server-side PDF watermarking, and an admin dashboard with one-click invites and export.

Results

Working portal about four weeks from the first call. The second security audit caught and fixed a high-severity information leak before launch.

The firm had no shared, evidence-based way to judge where AI would pay off across its portfolio. We scored every company the same way, then defended each call live with the partners. The ranking held, and the top pick became a paid engagement.

The friction

Nine portfolio companies, and no common way to tell where AI moves the needle and where it burns time and capital.

How we found the truth

A seven-dimension readiness assessment for every company, overlaid with effort versus impact and four leadership filters. Readouts with each company's leadership, then with the partners.

What we delivered

A uniform scorecard across the portfolio and a short list of front-runners, each with a clear first move.

Results

The partners challenged every call live and the ranking held. The top-ranked company became a paid 30-day assessment.

A 25-person board had been through three moderators and a messaging debate open since 2018. We mined the association's own surveys, found two buyer groups the strategy ignored, and left with a signed-off plan: five methods, each with a named owner.

The friction

Industry volume down 13% and revenue down 20% over four years, a slow-moving board, and no shared plan.

How we found the truth

Mined five of the association's own sources, including a survey of 112 architects and builders. Found two ignored segments: code officials, and builders who had used the product and stopped. Taught the board a new planning frame in 45 minutes before the session.

What we delivered

Two days of in-person facilitation and a finished strategy for next year: vision, values, five methods with owners, obstacles and measures, including a new code-official effort and practical AI uses. Sent the evening after.

Results

Five owned methods from two days in the room. The board asked for the materials to keep using them.