What a Head of AI Actually Does
A Head of AI decides what AI the company builds and buys, kills the parts that do not work, and owns how AI is used across the business. What they do not do, mostly, is build the AI themselves. That single distinction clears up most of the confusion about the role, because the title makes it sound like a senior engineer, and the actual job is closer to a portfolio manager with a mandate. They are accountable for the return the company gets on AI, not for the model architecture.
The job description version of this role is usually a wish list: “drive AI innovation, foster a culture of experimentation, deliver transformative outcomes.” That tells you nothing about what the person does on a Tuesday. The concrete version is more useful, and less glamorous, and it is what separates a Head of AI who produces a return from one who produces slide decks.
What people think the job is
The common assumption is that a Head of AI is the person who builds the company’s AI: trains the models, writes the agents, runs the data pipelines. That is a data science or ML engineering job, and a Head of AI usually oversees those people rather than being one.
The second assumption is that they are the company’s AI tool-picker, the person who evaluates and buys the software. That is part of it, but a small part, and a Head of AI who spends most of their time in tool demos has misunderstood the role. The tools are the least durable decision they make.
What both assumptions miss is that the job is about decisions and ownership, not construction. The value a Head of AI creates is in choosing what is worth building, making sure it gets integrated and used, and owning the result, none of which is the building itself.
The actual work, week to week
Strip away the job-description language and the real work falls into a handful of recurring activities.
They decide where AI is worth the investment, and, just as importantly, where it is not, so the company places its bets deliberately rather than chasing whatever is loud that month. They consolidate the scattered pilots that departments started on their own into a coherent portfolio with shared standards. They move the initiatives that work from proof-of-concept into workflows people actually use, which is where most AI value is won or lost. They own governance, so the company can use AI without accumulating the liability that unmanaged AI creates. And they translate fast-moving AI developments into terms the CEO and the board can make decisions with.
None of that involves writing a model. All of it involves judgment about what matters, and the standing to make those calls stick across functions that do not report to them.
The most important thing they do
If the role came down to one activity, it would be this: a Head of AI kills bad AI projects. That sounds negative, and it is the single highest-value thing they do.
Most AI money is wasted through fragmentation, dozens of half-finished pilots, duplicated tools, experiments that impress in a demo and deliver nothing in production. A company without a Head of AI has no one whose job is to look across all of that and say “this one is working, scale it; these five are not, stop them.” So the pilots accumulate, the spend grows, and the return stays flat. A Head of AI who is willing to kill projects, and has the mandate to make it stick, is exactly what turns AI spend into AI return. The organisations that have this role see roughly 10% higher return on their AI investment (IBM), and a large part of that is simply someone with the authority to stop the waste.
What a Head of AI does not do
Naming the boundaries keeps the role from becoming a catch-all, which is how it gets diluted into ineffectiveness.
A Head of AI is not the company’s data scientist; they set direction for the technical team, not do its work. They are not the person who personally builds every automation. They are not a pure compliance officer, though they own governance; a Head of AI who spends all their time on policy has stopped creating value. And they are not a figurehead whose job is to signal to the market that the company “takes AI seriously,” which is the most common way the role fails, a grand title with no budget, no mandate, and no authority to stop a bad project.
The role only works when it has real ownership: of the strategy, the portfolio, and the governance, with a direct line to the CEO. Without that, it is a costume.
How to tell a real one from a figurehead
The test is where they spend their time and what they are allowed to decide. A real Head of AI spends most of their time on strategy and on the portfolio, deciding what to build and killing what does not work, and they can actually stop a project that another function wants to keep. A figurehead spends their time on governance paperwork and board presentations, and cannot override a department that wants to keep its pet AI tool.
Ask a simple question of anyone in the role, or anyone you are considering for it: “What AI project have you killed, and what happened?” A real Head of AI has a clear answer. A figurehead does not, because they were never given the authority to kill anything. The willingness and the standing to stop work is the clearest signal that the role is real. It is the same distinction between renting an answer and owning the decision: a Head of AI who cannot decide is not one.
This is the version of the role Thane Alaric provides, owning the AI decisions and the outcome rather than occupying the title, as part of the broader Head of AI role a company puts in place. What the role would own in your company depends on what your company already owns. Book a call and we can map the real scope rather than the generic one.
Frequently Asked Questions
What does a Head of AI actually do?
A Head of AI decides what AI the company builds and buys, moves the initiatives that work into real use, kills the ones that do not, and owns governance and vendor decisions. They are accountable for the return the company gets on AI, not for building the models themselves. The core of the job is judgment and ownership, not construction.
Does a Head of AI build the AI?
Usually not. Building models, writing agents, and running data pipelines is data science and ML engineering work; a Head of AI oversees those people rather than doing that work. The title suggests a senior engineer, but the actual role is closer to a portfolio manager with a company-wide mandate, accountable for outcomes rather than architecture.
What is the most important thing a Head of AI does?
Kill bad AI projects. Most AI money is wasted through fragmentation: too many half-finished pilots and duplicated tools that never reach production. A Head of AI with the mandate to look across everything and stop what is not working is what turns AI spend into AI return. Organisations with the role see roughly 10% higher return on AI investment (IBM), largely from ending the waste.
What does a Head of AI not do?
They are not the company’s data scientist, not the person personally building every automation, and not a pure compliance officer, though they own governance. Most importantly, a real Head of AI is not a figurehead whose job is to signal that the company takes AI seriously. A grand title with no budget, mandate, or authority to stop a bad project is the most common way the role fails.
How can I tell if a Head of AI is effective?
Look at where they spend their time and what they can decide. An effective one spends most of their time on strategy and portfolio and can actually stop a project another function wants to keep. Ask them what AI project they have killed and what happened; a real Head of AI has a clear answer, while a figurehead does not, because they were never given the authority to kill anything.
Is a Head of AI the same as a Chief AI Officer?
They describe the same job at different scales, and the naming matters less than the mandate. Both own AI as a company-wide function: strategy, portfolio, governance, and vendor decisions, reporting to the CEO. What matters is real ownership and the authority to make decisions stick across functions, not whether the title says Head of AI or Chief AI Officer.
About the Author
Daniel Förster is the Managing Partner of Thane Alaric. For over a decade he has built and run companies, served over 2,000 founders, and worked as the embedded operator (COO, CFO and CMO in function) inside founder-led businesses. He now leads Thane Alaric, where companies become AI-native the right way: own the engine, rent the models, keep the judgment that’s yours.
