What does a Chief AI Officer do? The job description, the fractional version, and when a small company needs one
The Chief AI Officer owns how the company gets redesigned around AI, not which tools it buys. The job, the fractional version, three signals.
Every Chief AI Officer job description I have read is a CTO job description with the word "AI" pasted in.
That's the problem with the title. It is being hired faster than it is being defined, and when a role gets hired before it gets defined, the org chart fills in the blank with whatever it already knows. So the new Chief AI Officer inherits vendor evaluation, a pilot budget and an adoption dashboard... and eighteen months later the company has more AI than it had, and exactly the shape it always had.
So, what does a Chief AI Officer do? Here is the version I would write down. The Chief AI Officer is accountable for how the organisation gets redesigned around AI. Roles, processes, decision flow. Not for which tools it buys.
That one sentence separates the job from every other seat at the table, and it is the sentence the descriptions I have read leave out.
What follows is the job description I would hand a founder, the fractional version of it, because in a 20 to 200-person company the job is a season and not a seat, and the three signals that tell you the season has started.
The title everyone is hiring for, and the job almost nobody has written down
Titles get created when a category starts to matter and nobody knows where it lives yet.
A company at this stage has the licences. It may have the title. It has not written down what the title is for, and so the description gets borrowed from a seat that already exists. The ones I have read sort into three shapes. A rebadged CTO: systems, integrations, security, "AI infrastructure." A rebadged Head of Data: models, pipelines, a roadmap of use cases. Or an innovation role: pilots, demos, a lab, and no authority over anyone outside it.
None of the three has the authority to change who does what. And who does what is the only thing that decides whether AI changes the company.
MIT's Project NANDA looked at enterprise generative-AI initiatives in 2025 and found that roughly 95% of the organisations it studied showed no measurable return, and that the blocker was workflow integration, not model quality. BCG's 10-20-70 rule puts about 10% of AI value in the algorithms, 20% in the technology and data, and 70% in people and processes. Almost every job description I have read is written for the first 30%.
The job nobody has written down is the 70%. That is the Chief AI Officer's job.
What a Chief AI Officer actually does: five responsibilities
The job description, then. Five responsibilities, roughly in this order, and the tool arrives last on purpose.
The sequence: Map → Point → Sort → Redesign → Carry and hand over. Tool decision after all five.
1. Owns the map of the real organisation
Not the org chart. The actual information graph: who decides, who routes, where work stalls, and how much of each person's attention goes to coordination rather than creation. The output is two artefacts: a real-org diagram, distinct from the formal chart, and a per-role snapshot of attention spent on coordination versus creation. The chart says who reports to whom. The map says how work moves. The gap between the two is the redesign's risk register.
2. Decides where AI is pointed, and who has the decision rights
Somebody has to own the answer to "what outcome are we producing, and which parts of it are the machines doing?" Left unowned, that answer gets made by whoever is loudest, or by whichever team got the licences first. The Chief AI Officer makes it explicit: which work is routed to AI, which stays human, and who decides when that line moves. It is a leadership function, not a project-management one, and it is where AI strategy actually lives in a company this size.
3. Sorts every workflow into a zone, out loud
This is the 3-Zone Co-Pilot (Replace / Augment / Refuse). Every workflow in the company belongs in one of the three. Replace: the AI does it end to end. Augment: the AI prepares, a human decides. Refuse: the downside is asymmetric, keep the model out. The premise of the framework is that adoption fails when work is misfiled: a workflow that should have been replaced outright gets politely augmented instead. The Chief AI Officer makes the call, in writing, starting with the ten workflows that eat the most attention.
4. Redesigns roles, process and decision flow around the new shape
This is the responsibility the three before it exist to serve. The structural reason it is needed is a pattern I call the Hourglass Collapse: the narrow middle of an org chart exists to route information, AI routes it faster, and once the routing job is automated that layer has nothing load-bearing left to do. Somebody senior has to redraw the chart ahead of that, name who owns what, and make sure the functions every AI-augmented team needs are covered. I use the ORBIT Framework for this. It names the five functions a team has to have covered, and the first thing it does on any team is find the one nobody owns.
5. Carries the change at the founder's table, then hands it over
Reassigning people's jobs is not a memo. It is a series of identity conversations, held by someone with the standing to hold them and the distance to say the hard part plainly. That is why the seat sits with the founder, not under the CTO. The last part of the job is the handover: the map, the zone classification, the redrawn chart and the working loops stay with the company, with named owners, so the capability outlives the person.
Notice what is not on the list: choosing a vendor. It happens late, after everything upstream is settled, and by then it is easy.
The ten-minute test
Open the Chief AI Officer job description you have, or any posted one.
- Mark every line that is about tools, vendors, pilots, platforms or dashboards.
- Mark every line that is about who decides what, or whose role changes.
- Count each set.
If the second set is empty, you have written a procurement role with an executive title. If the first set is empty, you have written this page. Anything in between tells you which seat the description was borrowed from.
What a Chief AI Officer does not do
Three things get filed under this title that belong to other seats.
Tool selection. Which model, which platform, which licence tier. An engineer can answer this in an afternoon once the org has decided what the tool is for. It changes every quarter anyway. If a job description leads with "evaluate and select AI vendors," it is describing a procurement role with an executive salary. The same goes for running pilots: a pilot proves a tool works, and nobody is short of that proof. What is missing is a decision about what changes when it does.
Model building. Training, fine-tuning, research. If your company needs this, you need a machine-learning lead and probably a data team, and AI is very likely your product. That is a different problem and a different person.
IT. Identity, devices, security policy, the integrations that keep the stack running. The CTO or IT lead owns this and should keep owning it. The Chief AI Officer needs them, works alongside them, and does not replace them.
Chief AI Officer vs CTO vs Head of Data vs AI consultant
The four get used interchangeably. They are four different purchases.
| Chief AI Officer | CTO | Head of Data | AI consultant | |
|---|---|---|---|---|
| Accountable for | How the org is redesigned around AI: roles, process, decision flow | The systems: architecture, security, engineering delivery | The data: pipelines, quality, reporting, the models that sit on them | The deliverable: a recommendation |
| Starts with | The org's actual information graph | The stack | The warehouse | Tool selection, use-case list |
| Position | Inside, at the founder's table | Inside, runs engineering | Inside, usually under the CTO or COO | Outside, advisory |
| Can change who does what | Yes, that is the job | Inside engineering | Inside the data team | No |
| Ends | At handover, if fractional. Never, if full-time | Doesn't | Doesn't | At the report |
The CTO and the Head of Data own what they own, and should. Neither has the mandate to tell the operations team that a coordination role no longer exists, and neither should be asked to. The consultant is accountable for the deliverable, which is fine when you have a defined technical question and wrong when your org shape is the constraint.
Then there is the full-time Chief AI Officer: a permanent executive, on payroll, with a team under them. The right hire when AI is what you sell. For a founder-led B2B company it is a permanent answer to a temporary question, and a seat only a large company can justify carrying.
What "fractional CAIO" means
The short version: the same accountability, held part-time, for a fixed term, by someone who reports to the founder and is built to leave. I define the role in full on the Fractional CAIO page; this page is about the job.
"Fractional" describes the time and the term. It does not describe the seniority. Three properties make the shape work in a 20 to 200-person company:
- Senior enough to redraw the chart. The job moves people between roles. The founder has to treat the person doing it as a peer, or nothing moves.
- External enough to say what insiders cannot. Nobody inside a layer can be the one to conclude that the layer has no load-bearing job left. Someone from outside, holding a seat, can.
- Temporary by design. The redesign takes months. The seat should end when the redesign does.
It is not a consultant with a better title. The consultant answers for the report. The Fractional CAIO answers for what changed: the roles, the workflows, the capability still there after handover.
The tooling stays. The executive leaves.
When a 20 to 200-person company needs one
Three signals. Any one of them is enough.
Signal 1: you already bought the AI and it went cold. Licences went out company-wide, a mandate came down, and six months later the dashboard reads "last opened: 47 days ago" while a few curious employees do real work in personal accounts nobody can see. That is not a tooling failure. It is the structure answering truthfully. I have written up the full diagnosis in why AI rollouts fail.
Signal 2: AI decisions are landing on the wrong seat. The CTO is choosing the tools because that is where technology decisions live. The ops lead is being asked to "drive adoption" of something they did not choose. Whoever is loudest is deciding what gets automated. The seat that should own the question does not exist, so the org chart is routing it to the nearest one.
Signal 3: competitors are shipping the same output with fewer people, and the founder is the only person who could redraw the chart. No calendar to design it, no bench to delegate it to, and no appetite to hire a permanent executive for a problem with an end date.
One disqualifier, and it outweighs the three signals combined: the founder has to be curious about how the company itself should change. This work reassigns people's jobs. It does not survive a founder who wants the outcome without the restructuring. And below about 20 people there is not enough org to redesign yet.
What a Chief AI Officer costs, and why the salary question is the wrong one
People search "chief AI officer salary" because it is the natural next question. I do not publish a salary benchmark on this site, and I would not trust one for a 20 to 200-person company if I did.
The benchmarks I have seen are drawn from companies where the role is a permanent seat with a team underneath it, equity on top, and a mandate that never ends. That is a different job. Quoting it to a 60-person agency or a 120-person SaaS company answers a question the company should not be asking. The useful question is not "what does this person cost per year?" It is "what does the redesign cost, and when does it end?"
A salary pays for a year of someone's presence. This job has an end date, so it should be priced like something that finishes. That is the reasoning behind the fractional model, and the number is published where it belongs, on the page for what a Fractional CAIO costs.
The binary
There are two versions of the next twelve months.
In the first, you hire the title, or leave the seat empty and let the CTO absorb it. Either way the job reads systems and vendors, the chart stays as drawn, and the AI budget goes to making the old shape slightly faster.
In the second, you write the job down as accountability for how the company gets redesigned, put someone at the founder's table for a season, and let the tools arrive last.
One of those is happening at your company this year. The only thing you choose is which.
FAQ
What does a Chief AI Officer do?
A Chief AI Officer is accountable for how the organisation gets redesigned around AI: its roles, its processes and its decision flow. The job is to map the real org, decide where AI is pointed and who has decision rights, sort every workflow into Replace, Augment or Refuse, redraw roles and process around the new shape, and carry that change at the founder's table. Tool selection happens last and is not the job.
What is a fractional CAIO?
A fractional CAIO holds the Chief AI Officer's accountability part-time and for a fixed term: the redesign of roles, process and decision flow, not the tools. The person reports to the founder and leaves at handover. "Fractional" describes the time and the term, not the seniority. The full definition, the engagement shape and the cost are on the Fractional CAIO page.
Does a small company need a Chief AI Officer?
A company of 20 to 200 people needs the function for a season, not the permanent seat. The signals are a top-down AI push that went cold, AI decisions defaulting to the CTO or whoever is loudest, and competitors shipping the same output with fewer people while the founder has no calendar to redraw the chart. Below about 20 people there is not enough org to redesign yet.
What is the difference between a Chief AI Officer and a CTO?
The CTO is accountable for the systems: architecture, security, engineering delivery. The Chief AI Officer is accountable for the organisation: who does what once AI is doing part of the work. A CTO can change roles inside engineering. A Chief AI Officer changes roles across the company, which is why the seat sits with the founder rather than under the CTO. The two work alongside each other; neither replaces the other.
What is a Chief AI Officer's salary?
This site does not publish salary benchmarks. Published figures come from large companies where the role is a permanent seat with a team and equity attached, which is a different job from the one a 20 to 200-person company needs. The better question is what the redesign costs and when it ends. The fractional model is priced per engagement rather than per year; the number is on the Fractional CAIO page linked above.
Read this next: The Hourglass Collapse... the structural reason the seat exists, and why the middle of the chart goes first.
Sources
- MIT Project NANDA: The GenAI Divide: State of AI in Business 2025 (July 2025). 300+ initiative reviews, 52 interviews, 153 survey responses; finding that ~95% of organizations studied saw no measurable P&L return, with the blocker identified as workflow integration rather than model quality. Widely mirrored PDF: https://cloudelligent.com/wp-content/uploads/2026/02/v0.1_State_of_AI_in_Business_2025_Report.pdf
- BCG: the 10-20-70 rule (10% algorithms / 20% technology and data / 70% people and processes) and the 2024 finding that ~74% of companies struggle to achieve and scale AI value. See From Potential to Profit: Closing the AI Impact Gap: https://www.bcg.com/publications/2025/closing-the-ai-impact-gap and The Leader's Guide to Transforming with AI: https://www.bcg.com/featured-insights/the-leaders-guide-to-transforming-with-ai
Frameworks referenced are my own: the Hourglass Collapse, the ORBIT Framework, and the 3-Zone Co-Pilot (Replace / Augment / Refuse).
The Chief AI Officer job description, one responsibility per issue, in your inbox
The frameworks behind this page go out to the list, one per issue.
Or if you already know this is your situation: book a 30-minute scoping call