A fractional CAIO, so someone owns whether AI earns its keep.
Most small businesses looking for an AI consultant do not need another AI tool. They need one named person who owns the question of whether any of this is worth the money, and who has the standing to act on the answer.
Free and written. I read every one myself and reply within two business days.
That seat is empty almost everywhere. Subscriptions get bought and abandoned, a chatbot runs with nobody responsible for it, two or three pilots start in parallel and none of them finish, and staff quietly paste customer records into whatever is free.
A fractional Chief AI Officer holds that seat part-time. Not a project with an end date and a deliverable that goes stale, but a function: the roadmap, the vendor decisions, the pilots, the governance, the training, and the report that says in hours and dollars what changed. I run my own companies on systems built this way, so the judgment comes from operating them, not from watching a category.
The work.
- →The AI readiness assessment: how work flows, where data lives, what is already running, and who would own it
- →A ninety-day roadmap in sequence, with the first work chosen for low blast radius
- →Build, buy or wait decisions, and standing between the company and its vendors
- →One process at a time: baseline it, pilot it beside the old way, measure it, hand it off with a runbook
- →Workflow automation and agent systems in production, with approval gates and audit logs
- →The AI register: one table of every system running, what it touches, who owns it and when it was last reviewed
- →An AI use policy sized to the company, from a one-page starter to a regulated-industry version
- →What may and may not be fed to a model, on business accounts rather than personal ones
- →A written monthly report in hours and dollars, and a ninety-day readout that leads with the misses
Build, buy, or wait: how the call gets made
Every AI decision goes through the same questions before any money is spent.
- Buy when an off-the-shelf tool already does the job, the data it needs can safely leave the company, and the vendor will still be there next year
- Build when the work is specific to how this company runs, the data cannot leave, or no product fits without bending the process around it
- Wait when the process is not written down yet, nobody would own the result, or the gain is smaller than the cost of running and checking it
- Either way, the baseline is measured first, so the result is judged in hours and dollars rather than impressions
Governance that runs with the system, not after it
Every system I put into production carries the same four controls. They are what an insurer, a regulator or a large customer will ask about, and they are cheaper to build in than to bolt on.
- Approval gates: anything that sends, pays, deletes or commits on the company’s behalf waits for a person until it has earned more autonomy
- Evaluation: each system is tested against real examples before launch and re-checked on a schedule, with agreed conditions for switching it off
- Logging: prompts, outputs and actions are recorded, so any result can be traced to what produced it
- Named owners: every system in the AI register has one person inside the company responsible for it, and a runbook they can follow
Fractional CAIO vs AI consultant vs AI agency
All three can help a small business with AI. They are accountable for different things, and that difference decides whether the work survives its first quarter.
| Fractional CAIO | AI consultant | AI agency | |
|---|---|---|---|
| What they own | The AI function: roadmap, vendors, governance and results | A recommendation or a strategy document | The build they were hired to deliver |
| Time commitment | Part-time, ongoing | A defined project | A defined project or a support contract |
| Cost shape | Monthly retainer | Project fee | Project fee, often with ongoing license or support costs |
| Build or buy | Decided against a measured baseline | Advises; the company decides | Usually build, because building is what it sells |
| After launch | Owns the result and reports on it monthly | Gone once the deliverable is in | Supports what it built, if under contract |
| Best when | AI decisions keep coming and nobody owns them | One question needs an outside answer | The system is defined and someone inside can own it |
Free, written, reviewed by me.
With the free AI readiness assessment. You describe how the work flows today and where the data lives, I review it by hand, and you get a written note on the two or three places AI would pay off first, what would have to be true for each to run reliably, and what I would do first if I held the seat.
What you have at the end: One person accountable for AI in the business, a sequence instead of a pile of pilots, working systems with named owners, and a number each month that says whether it is working.
Who this is for, and who it isn't.
- →Owner-led companies where the leaks are big enough to matter and the owner still decides alone.
- →Service businesses with repetitive, document-heavy or communication-heavy work that eats staff time.
- →Leadership teams that have watched internal AI efforts stall and know nobody on the payroll can lead this.
- →Boards and owners who have realised nobody can answer who approved the tools already in use.
- →Companies whose insurer, regulator or biggest customer has started asking what AI they run.
- ×Companies looking for free advice or a quote without an assessment.
- ×Anyone who wants a tool recommended rather than a problem solved.
- ×Engagements where leadership will not make decisions or act on findings.
Assess, scope, execute.
Engagements are scoped as fixed projects or monthly retainers. Scope, deliverables, and the engagement range are defined at intake based on the company’s size and what it needs.
Assessment
Every engagement begins with a written assessment: the free snapshot first, then a full assessment if it is warranted. Current state, gaps, and a prioritized plan.
Scope
A fixed project to close specific gaps, or a monthly retainer for ongoing leadership. Deliverables, cadence, and the engagement range are set at intake.
Execution
I do the work directly or direct the people who do: your IT provider, developers, and vendors. Weekly working sessions and written progress.
Proof, not promises.
- A8L OS and the operator agentsAutonomous agents running day-to-day work for my own company, with risk tiers and approval gates.
- MDRN AI CoachAn evidence-grounded call-scoring platform built for a financial advisory firm.
- Agency HQThirty-two role-specific agents across seven departments, with a human-in-the-loop approval queue.
Asked before the first call.
What is a fractional CAIO?
A Chief AI Officer is the one person accountable for whether AI makes the company money, with the standing to decide what gets built, bought or stopped. Fractional means holding that seat part-time for a company that needs the function but cannot justify a full-time executive. It is a leadership role rather than an engineering one: the raw material is the business, not the model.
Do you work as an AI consultant for small businesses?
Yes, as the person who owns the result rather than a consultant who hands over a report. Most of the companies I work with are small or mid-sized, and the work starts with the free AI readiness assessment, which says where AI would pay off first before you spend anything.
How is this different from hiring an AI consultant or an agency?
A consulting project ends with a deliverable, and the deliverable usually dies because nobody owns it afterwards. An agency wins the work and subcontracts the building. A fractional CAIO owns the function and the number that goes with it: the roadmap, the vendors, the governance, and the report each month saying what actually changed. When the AI result is bad, it is my bad result.
What does the engagement look like week to week?
A standing weekly meeting with you and the managers whose departments have live work, covering what shipped, what the numbers say, what is blocked and what is next. A short written memo in the same shape every week. A monthly report in hours and dollars. A readout at ninety days measured against what was forecast. The rhythm is fixed and written into the engagement, so you always know what you are getting.
Do you build the systems or only advise?
Both, and the order matters. The assessment and the build-or-buy work are advisory. Where building is the right answer I build it, run it beside the existing process with agreed conditions for killing it, measure it against a baseline taken first, and hand it to a named owner with a runbook. Nothing is handed over without someone inside the company responsible for it.
Our insurer sent a questionnaire asking what AI we use. Can you help with that?
That is usually the moment this work starts. The answer begins with a register: one table listing every AI system running in the business, what it does, what data it touches, who owns it, how much it is allowed to do without a person, and when it was last reviewed. Most companies cannot produce that, which is why the questionnaire stalls. A first pass is quick once someone goes and looks instead of asking around, and the longer part is finding the tools nobody mentioned and confirming what each one actually touches. It is the same document a regulator, an auditor or a large customer will ask for next, and the policy and the sign-off rules follow from it.
How do you handle data, privacy and the rules we have to follow?
As a security problem first, because that is the other practice. Data gets sorted into what may be sent to a model and what may not, business use runs on business-tier or API accounts rather than personal logins, access is scoped, prompts and outputs are logged, and the worst case is capped before anything is switched on. Where regulation applies, the design starts from the rule rather than meeting it afterwards.
Which models and tools do you use?
Whichever fit the job and the data-handling requirements. Recent work uses Claude, OpenAI models, open-weight models where data cannot leave the company, Supabase and Postgres for data, and Next.js and TypeScript for the applications around them. The tool choice is a consequence of the decision, not the starting point.