← Back to home Specialties

I don't use AI. I wire it into the business.

Anyone can open a chat window. The work I do is deciding what data moves, where it goes, and what happens to it on the way. Six systems below, each running in production today, each with the architecture that makes it run.

Call intelligence Lean Marketing Plan generation 1PMP Architect Second brain Obsidian + Claude GPT library Client facing Grounded writer Lean Intelligence Agent pipeline DropChain

What the systems moved

1,200

Opt-Ins in One Week

From one tool that turns client intake into a finished marketing plan.

20% → 70%

Renewal Rate

After AI reporting and health scoring gave advisors the data before every renewal.

10+

Hours Reclaimed Weekly

Per person, recurring, with no manual upkeep to keep it running.

200+

Clients on the Platform

At 60%+ monthly active usage, more than double the typical B2B rate.

The Problem

Most AI spending buys nothing.

A 2025 MIT study on enterprise AI adoption found that 95% of companies investing in generative AI saw no return. The technology wasn't the problem. Researchers traced the failure to how the projects got built. Teams built their own tools instead of buying proven ones, then pointed them at departments like marketing and sales instead of the back-office workflows where AI reliably takes cost out.

That failure has a shape, and it shows up the same way every time. Someone decides the company should be using AI, then goes looking for somewhere to put it. The tool arrives before the problem does.

Every system on this page started from the opposite end. Find the number that's stuck, then build the smallest system that moves it.

Source: MIT Project NANDA, "The GenAI Divide: State of AI in Business 2025."

My Approach

Four rules. No exceptions.

Diagnose First

I don't open a tool until I know what's broken. A project that starts with "let's use AI for this" already has it backward. Start with the constraint. AI is just the fastest way to remove it.

Build for Daily Use

The bar for shipping is whether the team notices if it disappears tomorrow. If a system isn't opened every day, it isn't finished.

Design for Autonomy

Most "AI tools" are a chat window bolted onto a prompt. I build agents that act inside production systems, with enough context to make a decision without someone reviewing every output.

Measure the Outcome

Every system here was built against one number. If that number didn't move, the project isn't done, no matter how good the demo looked.

OPERATIONS / Lean Marketing

Every coaching call ends in four places at once.

A good coaching call used to die in someone's memory. The advisor knew what happened, nobody else did, and the client's win never made it anywhere it could be used. I built the system that reads every call and writes the result into the four places that need it, without anyone typing a note.

Coaching call RAW AUDIO Transcribe + extract AI PROCESSING Client win log EVIDENCE LIBRARY CS handoff RETENTION RISK Knowledge base TEAM TRANSFER Weekly report COACH QUALITY
One input, four destinations. The routing is the point. Each destination gets only the part of the call it can act on, so nobody reads a transcript to find their piece.
10+ hrs reclaimed per week, per person, across a team of seven.
DEMAND GENERATION / 1PMP Architect

A full marketing plan, built from a client's own answers.

The strategy session was the bottleneck. Clients waited days for a plan I could have walked them through in an hour. So I built the agent that runs the same diagnostic I would, stage by stage, and hands back a finished One-Page Marketing Plan while the client is still at their desk.

n8n  ·  1PMP ARCHITECT
The 1PMP Architect workflow in n8n, showing a HubSpot trigger feeding three colour-coded agent stages that generate the plan, then publish it and notify the team.
The live workflow. A HubSpot form fires the trigger, then three colour-coded stages run in sequence: Before builds target market, messaging, and media; During builds lead capture, nurture, and conversion; After builds experience, lifetime value, and referrals. The finished plan is published to the CMS, then the client and the team are notified.
1,200 opt-ins in a single week, the highest-converting asset the Accelerator has shipped.
PERSONAL INFRASTRUCTURE / Second brain

One system that holds everything, and acts on it.

I run my whole operation out of a single vault. Obsidian holds the knowledge, Claude Code does the reasoning against it, and Telegram is how I talk to the thing from my phone. It has enough context to take action without being walked through every step, and it stays simple enough that asking it a question feels like asking a person.

OBSIDIAN  ·  GRAPH VIEW
Graph view of the Obsidian vault, showing dense clusters of linked notes with hub notes at the centre of each cluster.
The vault as it stands. The dense clusters on the left are where the linking has been done, so context carries between notes. The loose points on the right are captured but not yet connected.
Calls + notes CAPTURE Obsidian vault SINGLE SOURCE READ / WRITE Claude Code REASONING Telegram INTERFACE
Three surfaces, one vault. Capture and interface are separate on purpose, so getting something in never depends on being at a desk.
CLIENT ENABLEMENT / GPT library

Twenty tools so the work doesn't stop between sessions.

Coaching only moves a business if the work continues after the call. Clients would hit a wall on a Tuesday and sit on it until the next session. So instead of one general assistant, I built a library of narrow tools, each trained on one piece of the marketing process, each doing a job a client would otherwise wait on me for.

Website copywriter Turns a client's positioning into page copy, one section at a time.
Messaging analyzer Audits existing copy against the positioning and flags where it drifts.
Content pillar builder Sets the core themes so every post ties back to one strategy.
+17 more in the library
20+ GPTs in production, each scoped to one job rather than one assistant trying to do all of them.
PRODUCT / Lean Intelligence

An AI writer trained on the business, not the internet.

Generic AI copy fails for one reason. It doesn't know the business. Lean Intelligence reads each client's foundational documents first, the offer, the ICP, the messaging, the tone of voice, so the model is grounded in that client before it writes a word. I owned this one end to end, from the interface through the roadmap.

FOUNDATIONAL DOCS Offer Ideal customer Messaging Tone of voice Grounded writer TRAINED PER CLIENT On-brand copy OUTPUT
Grounding, not prompting. The documents load before the request, so the model starts every job already knowing who it's writing as.
200+ clients at 60%+ monthly active usage, more than double the typical rate for B2B software.
MULTI-AGENT / DropChain

A marketing team, split into four agents.

One prompt asked to do a whole marketing strategy gives you shallow work in four directions. So I split it into four agents, each with one job, running in sequence. The important part is underneath: every stage writes what it learned into a shared context, so the SEO agent knows what the messaging agent decided.

Research AGENT 01 ICP + messaging AGENT 02 SEO strategy AGENT 03 Content execution AGENT 04 SHARED CONTEXT Each stage writes its findings back down, and every stage after it reads from here.
Sequence with memory. The chain is what stops stage four from writing content that contradicts the positioning set in stage two.
See the full breakdown on DropChain →

Ready to put AI to work?

Tell me where your team is losing the most time. There's almost always a number waiting to be moved.

Get In Touch