Work

A copy-analysis process living in spreadsheets that should have been software; a finalist vendor quoting ~$50K.

The fastest way to win the argument was to build the thing.

• Vibe-coded a working SaaS UI prototype in 3–4 days; 50 commits on Jon's org repo over five weeks for the next-gen UI.●On record · repo

• Hired instead of the vendor; paid about five times what I'd have asked. Jon: “I don't think there's many people that can even touch Andrew when it comes to AI.”●Client-stated

Build the prototype instead of the proposal.

An FTC earnings-claim detector (express and implied claims) giving inconsistent results; the internal team stuck.

Don't ask the model to do everything — regex pre-filter the obvious, send only uncertain chunks to the LLM, ship the stable version first.

• Detection architecture + prompts; four-month engagement.●On record · call 2024-05-28, Slack dates

• Solved free on the first call, then hired. Video testimonial: “by far the best AI expert I've had a chance to talk to.”●Client-stated
• Honest note: later edge cases exposed misses; the architecture, not the accuracy claim, is the lesson.

Give people the whole answer first.

A cold-email education company with a stalled growth engine and a refund problem.

The product was fine; the acquisition engine and the post-purchase experience weren't.

• Weekly partner masterclasses (~30 sessions, 100–130 live attendees against a 150K+ list); cohort onboarding I hosted weekly; a refund-intercept funnel that made the customer raise the refund first.

• ~$45K → ~$97K/month in ten weeks●Client-stated
• 49 of 50 refund requests recovered●Client-stated
• Longer arc ~$39K → $300K+/month over ten months (team of 11–16; no P&L published)○Self-reported

Acquisition is a system, not a tactic.

Six dead months of cold email; a CEO on the call saying he'd lost “tens of millions of valuation.”

The answer was in data they already had — 1,089 HubSpot deals and 617 recorded sales calls nobody had read as a set.

• Mirrored 617 Chorus calls to SQLite; deals → 931 accounts; caught a 35% churn overstatement (including in a second independent run); reversed my own thesis (Enterprise churns 32% vs SMB 14%); submitted the whole suite to two frontier models blind and published the disagreement table with 12 corrections; 14-slide brief.●On record

• The CEO went from furious to “what do you guys need from me?”●On record
• Campaign outcome: pending at time of writing.

Break your own conclusion before someone else does.

A 26-year broker floundering with ChatGPT for months; presentations eating 60 hours; underwriting eating days.

He didn't need a course; he needed the work taken off him — and one of his businesses shut down.

• Presentation stack (deck, survey, dashboard, landing page, email automation) in one session; underwriting skill; AI employee “Grace”; live-built an Argus-replacement underwriting skill on a real property in ~30 minutes; 12-week AI installation cohort co-built.

• 60 hrs → 15 min●Client-stated
• $120K copyright liability resolved in under a week, by him, with skills from our sessions●Client-stated
• Leads 3–4/month → 8–10/week●Client-stated
• 13 of 45 conference leads asked for an intro after Grace ran follow-up●On record
• Trent, one week in: “probably the best money I've ever spent.”●Client-stated

Install, don't inform.

Nobody knew where the spend was leaking.

• One prompt against seven days of ad data.

• 158 ads (75% of active) under 1.5 ROAS; $45K/week burning to return $36K; a $30K/week recovery path; a 37% blind spot. The agency owner: “That is a valuable tool… we could sell that.”●Client-stated

The diagnosis is the product.

Founders who'd refunded $80K of sales; activation and retention weren't connected.

Referrals, personalized onboarding, activation, and retention are one mechanism, not four features.

• The strategy, and the design for an automated referral-handoff (the app drafts a personalized intro from the referrer's number and onboards with referrer-supplied context).

• “You're our core user right now.” Founders committed to build the feature; became a revenue-sharing partner.●On record

The best consulting looks like being their most demanding user.