Services What I do

Three layers, seven plays, one operator.

Every engagement is one of three shapes: a fixed-scope entry, an operated subscription, or a retainer for the clients who need a named person accountable for AI. Underneath are seven things I actually do. Most clients buy two or three, and the entry engagement tells us which.

01 The three layers

Entry, core, wrapper.

Migration and vendor exit attach to any of them when a client event triggers one. Nothing is billed by the hour, and nothing is sold as a build with no one to run it.

Entry fees are credited against the subscription when you start within 90 days. The wrapper is sold only to clients already on the subscription.
Layer	What it is	What you get	How it ends
Entry: a fixed-scope first engagement project · fixed scope · one to three weeks	Paid discovery or a point-in-time verification	A decision memo with kill criteria, the specification, the evaluation set the system must pass; or a verification report with severities, reproductions and a fix order	Credited against the subscription if you start within 90 days. Or it ends there, and you keep the documents.
Core: the operated subscription subscription · monthly · written response windows	Operated AI systems, context stewardship and adoption coaching, under written service levels	A system that is run, measured and answered for. One monthly report, same metrics every month. Disclosure, consent evidence and registrations kept current.	Any month, with handoff and a fix-it guide. Nothing stops working when I leave.
Wrapper: a fractional AI operations lead retainer · for subscription clients only	The accountable person for AI across the business	Quarterly roadmap, a decision log, vendor and contract risk reviews, reporting to the owner, a written incident response window	Steps back down to the subscription whenever you want. Deliberately few seats, so each gets attention.

02 The seven plays

What I actually do, and which layer sells it.

01 Core

Operated AI systems.

An AI system that does a job in your business, run by me on your accounts. The agent that answers the phone. The assistant that answers from your documents. The automations that move a record from intake to work to money. I build it, or take over one that exists, write the response windows, and run it under a monthly report. The bot disclosure, the consent script, the consent evidence and the sender registrations are part of the operation, not an add-on.

Typical stack: Twilio and the voice platforms built on it; Claude, ChatGPT or Microsoft Copilot; n8n, Zapier and Make; and the CRM, work-management and accounting tools you already pay for.

You get

A system on your accounts, under your logins
Written business-hours response windows, with your fallback documented
One monthly report, the same metric set every month
Disclosure, consent scripts and consent evidence kept current
The exit option: handoff plus fix-it guide, any month

Good fit when

A vendor sold you an agent and nobody owns it
Calls, forms or documents are handled by a system no one on staff can explain
You want one accountable person, not a support queue

Not this. I do not promise uptime I do not control. Every response-time commitment is backed onto the upstream vendor’s published service level, and the monthly report shows theirs next to mine.

Proof Lead bridge Permit signal Document search Warehouse assistant Design-build contractor

02 Entry

Verification and audit.

A point-in-time verification of an AI system someone else built or runs. What it does with a hostile input. What it leaks. What it costs under load. Whether its answers are grounded in your documents. Whether its disclosures and consent evidence would survive a complaint. Framed on the OWASP Top 10 for LLM Applications 2025, the OWASP Top 10 for Agentic Applications 2026 and the NIST AI RMF 1.0 (Jan 2023) and NIST AI 600-1 (Jul 2024). Written findings with a severity, a reproduction and a fix order for each.

You get

A findings report: severity, reproduction, fix order
The test set, so the fixes can be re-checked by anyone
A one-page summary written for your insurer, counsel or customer
A re-test of the fixes, if you want it

Good fit when

A vendor or agency delivered an AI system and you have only their word
An insurer, counsel or a large customer is asking questions you cannot answer
A contract is up for renewal and you want the system tested first

Not this. I never verify a system I operate. Independence is the product. A system I run is checked by someone else, and I will name one.

How a verification runs, and what it tests

Proof Warehouse assistant

03 Entry

Specification and discovery.

One to three weeks on your systems and your people, ending in a decision memo: build, buy or skip, with the kill criteria written down. If the answer is build or buy, you also get the specification and the evaluation set the system has to pass before it goes live. Drafting a specification is cheap now. Deciding what not to build is the part worth paying for. You keep everything whether or not we go further, and the fee is credited against the subscription if you start within 90 days.

You get

A build, buy or skip memo with kill criteria
The specification, if the answer is build or buy
The evaluation set: real questions, expected answers, pass marks
A written note on what it would cost to operate, in hours and vendor spend

Good fit when

Three vendors have pitched you and they all sound right
You have a system in mind and no way to tell whether it worked
You want to buy once, not twice

Not this. Not a document pile. If the memo says skip, that is the deliverable, and it is usually the cheapest one.

Proof Design-build contractor

04 Core

Context stewardship.

The facts your systems answer from, kept as an asset: pricing rules, process steps, templates, the golden-question set and its expected answers, the consent ledger, the vendor ledger. Versioned, tested every month against the evaluation set, and reported on, so the assistant answers from this quarter’s price list and not last year’s. This is the part that cannot be regenerated by switching vendors, and it is yours.

You get

A knowledge base with owners and review dates
A golden-question set run against it every month, with the score in the report
A consent ledger and a vendor ledger you can hand to counsel or an auditor
Every change logged with the evaluation delta it caused

Good fit when

An assistant confidently answers with last year’s facts
The same questions get asked in the group chat every week
Process knowledge walks out the door when a person does

Not this. Built for your team first. A customer-facing assistant is a separate conversation, and I will tell you honestly whether the context is ready for one.

Proof Document search Design-build contractor

05 Attachment

Migration and vendor exit.

Moving a system off a vendor, a model or a platform that stopped fitting, with reconciliation: the evaluation set passes on the new stack before the old one is switched off, records are counted on both sides, and there is a way back. Triggered by a client event, an entry engagement or an exit. It is never sold on its own, because the scripts are cheap and the accountability for the reconciliation is the only part worth paying for.

You get

A cutover plan with a way back
Evaluation parity: the same test set, passed on both stacks, before the switch
A reconciliation count you can check yourself
The old vendor’s data, exported and in your hands

Good fit when

A vendor raised its price, changed its terms or stopped answering
A model swap broke answers nobody had a test for
You are leaving an operator and want the system to survive the move

Not this. No big-bang cutovers. If the evaluation set does not pass on the new stack, the old one stays on.

Proof Lead bridge

06 Core

Adoption and training.

The people who use the system learn it inside their own work, not in a slide deck: a recorded session per role, a one-page guide per process, and fifteen minutes a month on what changed and why. Part of every subscription. Also sold as a paid workshop for a team that is not ready for an operated system yet. In Thryv’s April 2026 survey of 561 small-business decision-makers, 70% said they need more training to use AI effectively (Thryv, 2026 AI and Small Business Adoption Survey, Apr 2026, n=561).

You get

A recorded session per role, so the next hire gets the same one
A one-page guide per process, matched to the system
A monthly fifteen minutes on what changed
For a workshop: the recording, the guides and the question list, yours to keep

Good fit when

The tool is paid for and nobody opens it
Quality depends on who is on shift
Onboarding takes months because everything is verbal

Not this. No 400-page binder. If a page is not used within a month, it gets cut.

Proof Design-build contractor

07 Wrapper

Fractional AI operations lead.

For clients already on the subscription who need one named person accountable for AI across the business: a quarterly roadmap, a decision log, vendor and contract risk reviews, reporting to the owner, and a written incident response window. Deliberately few seats at a time, so each one gets attention. If a fourth client asks, the answer is a reporting pack, not more hours.

You get

A quarterly roadmap and a running decision log
Vendor and contract risk reviews before each renewal
Owner reporting in the owner’s terms, not the vendor’s
A written incident response window

Good fit when

AI decisions are being made in three departments by nobody in particular
A vendor contract is up and you want it read by someone who runs these systems
You want an accountable name, not a committee

Not this. Not a standing on-call. The response window is written, business hours, with your fallback documented.

The service levels and the monthly report, in full How an engagement runs, step by step All selected work

Scope Stated plainly

What I don’t do.

So you are not guessing and neither am I.

Open-ended hourly work
There is no hourly meter anywhere. Entry engagements are fixed scope, operation is a subscription with written service levels, and the fractional lead is a retainer with a defined window.
Verify a system I operate
Independence is what makes a verification worth anything. A system I run is checked by someone else, and I will name one.
Promise uptime I do not control
I commit to response windows, in business hours, backed onto the upstream vendor’s published service level. I do not sell 24/7 on-call, and I do not sell uptime.
Guarantee lead volume
I run the system that answers and routes the leads you get. I do not promise how many there will be, and I do not run advertising.
Legal, tax or employment advice
Consent scripts, disclosure wording and worker-classification questions are framed as evidence packs for your counsel. Counsel owns the conclusions.
IT support
Hardware, networks, email hosting and device setup belong to your IT provider or MSP. Several MSPs refer clients here for the AI layer, and I refer back for everything else.
Work on accounts I own
Everything is set up under your logins from day one. If we never speak again, it keeps working.
Industry specialisation
The function is the same everywhere: intake, work, money. Surveying, landscape architecture, construction and the rest of any one trade stay with the people who do them.

Before you ask Common questions

Questions people ask before the first call.

What does “operated AI systems” mean?

A voice agent, a document assistant or an automation in your business that I run on your accounts under written business-hours response windows. The context it answers from is kept current. A monthly report says what ran, what broke, what changed and what it cost. You can leave any month with a handoff and a fix-it guide.

Do you build, or do you only operate?

Both, in that order. The entry engagement decides build, buy or skip. If the answer is build or buy, I build it or take over the vendor’s system, then operate it. I do not sell a build with no operated option, because a system nobody runs is the thing that fails at 9:40 on a Tuesday.

Can you audit a system you built?

No. Independence is what makes a verification worth anything. I verify systems that someone else built or runs. A system I operate is checked by someone else, and I will name one.

How much does it cost?

There is no price list and no hourly rate. The entry engagement is fixed scope and credited against the subscription if you start within 90 days. The subscription is monthly, with the scope, the response windows and the report’s metric set agreed in writing first. The figures are in the proposal.

Which industries?

Any. The function is the same everywhere: intake, work, money. Most clients are owner-run businesses of a few to a few hundred people, and a growing share arrive through the MSP, accountant or agency that already serves them.

Do you work remotely, and where?

Everything is remote. I am in San Marcos, CA on Pacific time and work with businesses anywhere in the US.

What happens if I stop the subscription?

The system keeps working, because it was never on my accounts. You get a handoff with training for whoever runs it next and a written fix-it guide. Anyone you hire can change it.

Not sure which layer you need?

That is what the intro call is for. Thirty minutes, no prep. Bring the system nobody is accountable for.

Book an intro call