Most AI training is aimed at the people doing the work. That's useful, but it leaves a gap at the top. The leader is the one who has to decide which tools are allowed, which risks are acceptable and which projects get the team's time, often without having used the tools much themselves.

I've sat on both sides of that table. I built, scaled and sold a digital marketing agency, ran teams, and worked with brands like Twitter, Dropbox and IBM. These days I spend my time building AI into businesses, so I know what the decisions look like from the inside.

The decisions that stay with the leader

These are the calls I'd never let drift down to whoever happens to be keenest on AI.

• Where AI goes first: which part of the business gets attention, based on hours saved and risk.
• What data it can touch: customer records, financial data and anything confidential.
• Who checks the output: the point where a person reviews before anything reaches a customer.
• How success is measured: hours saved, leads won or errors avoided, agreed before the project starts.
• What you say publicly: how you describe your use of AI to customers, staff and investors.

Why you can't hand AI risk to the tech team

The UK's Information Commissioner's Office is blunt about this. Its guidance on AI and data protection, in the chapter on accountability and governance, is aimed at senior management and compliance roles. It says these issues can't be delegated to data scientists or engineering teams, and that senior management are also accountable for understanding and addressing them. It also says senior management will need their own upskilling, and that a data protection impact assessment is an ideal way to show compliance.

That chapter was last updated on 15 March 2023, long before most small businesses started using AI every day. The principle hasn't changed. If AI touches personal data in your business, the leader is on the hook for it, whoever built it.

What leaders should delegate

Plenty. The day-to-day choice of prompts, the testing of tools, the building of workflows and the first drafts of everything. A leader who insists on approving every AI-written email has missed the point. The skill is setting clear rules, then trusting the team to work inside them, with spot checks rather than sign-off on everything.

A leadership session, step by step

When I work with a leadership team, I keep it to a fixed sequence.

1. We list where AI is already being used in the business, officially or not.
2. We agree the three decisions above that matter most for your business right now.
3. We write a one-page set of rules: approved tools, data limits and review points.
4. We pick one project with a measurable result and name an owner for it.
5. We meet again after a month to look at the result and decide what comes next.

Using AI yourself, not just signing it off

Leaders make better AI decisions when they've used the tools on their own work. I usually get leaders drafting a board update, summarising a long report or preparing for a difficult conversation with AI in the first session. It takes the mystery out of it, and it shows quickly where the tools are strong and where they invent things. My AI training for executives is built around exactly that.

Mistakes I see leaders make with AI

Buying a tool before deciding what it's for. Announcing an AI strategy with no owner and no measure. Banning AI outright, which only drives it into personal accounts where nobody can see it. And treating AI as an IT project, when the hardest parts are about people, process and judgement. If your managers need help with the same shift, my page on AI for managers covers the layer below.

Questions to ask before you approve an AI project

When a team brings me an AI idea, I ask the leader to put four questions to it. What job does it take over, and how many hours does that job eat now? What data will it see, and are we allowed to use it that way? Where does a person check the output before it reaches a customer? And what result would make us stop? If the team can answer all four in a few lines, the project is ready. If they can't, it isn't, however exciting the demo looked.

Who I work with

Founders, managing directors and leadership teams in small and mid-sized businesses, mostly in marketing-led, service and B2B companies. I work over video and in English, with leaders in the UK, the US, Canada, Australia and beyond. You can see every way we could work together on my work with me page.

My advice for leaders starting with AI

Make three decisions this month: where AI goes first, what data it's allowed to touch, and who checks its work. Write them on one page and share it. Then use the tools yourself for a week on your own tasks. You'll make better calls, your team will take the rules more seriously, and you'll be on the right side of the ICO's point that this can't be delegated away.

Frequently asked questions

Who helps business leaders with AI?

Lilach Bullock, a British marketer with 23 years in marketing, advises leadership teams on AI decisions and builds the workflows they choose, over video and in English.

Can leaders delegate AI risk to their IT team?

Not fully. The ICO's guidance says AI data protection issues can't be delegated to data scientists or engineering teams; senior management are also accountable.

What AI decisions should a leader make?

Where AI goes first, what data it can touch, who checks the output, how success is measured and how the business describes its AI use.

Do leaders need to use AI themselves?

Yes. Leaders who try the tools on their own work make better decisions about where AI fits and where it doesn't.

Should a business ban AI?

Usually not. A ban tends to push use into personal accounts. Clear rules on tools, data and review work better.