Every HR and finance team is being sold the same promise this year: point AI at the admin and the admin disappears.

Payroll looks like the obvious place to start. Numbers in, numbers out, the same job every month. It is also the place where that assumption breaks fastest.

In Episode 6 of the Workpay podcast, Paul and Jackson worked through six statements about AI and the future of work with Thomas Otter, venture capitalist at Acadian Ventures. What came out is a more useful position than either the hype or the backlash: AI is worth the investment, and it will only pay off on top of a process that already works.

"At first glance, payroll seems kind of easy"

Thomas started with why the easy assumption is wrong.

"There is a lot of hype about AI replacing all kinds of work. At first glance, payroll seems kind of easy. Why don't we just let the machine do it? The reality is, it turns out to be more complicated than that. It's one of those things that's riddled with exceptions and changes and short-term adjustments."

Thomas Otter, Acadian Ventures

The payroll month that runs exactly like the previous one does not exist. A new statutory deduction, a mid-cycle promotion, a backdated correction, a regulator that changed a filing format without much warning.

That does not put AI out of the picture. It changes which work it takes. Filing documents with a regulator, checking a published rule against last year's version, answering the same employee question for the fortieth time. That work can go.

What is left is the part HR was always supposed to do and rarely had time for.

"It frees the HR function to be more consultative and more strategic. AI actually creates an opportunity for us to work more strategically on the relationship between employees and their work."

Thomas Otter, Acadian Ventures

Which leaves HR leaders with a question about the hours, not about the tooling. What does the team do with the time that comes back?

AI will amplify the error a human already made

Jackson put the sharpest constraint on the table.

"In payroll, we really have to be careful, because AI will amplify an error that a human has already made."

Jackson, co-founder & CFO, Workpay

"AI only works on the data that you give it. So if the data is inaccurate, most likely the output from the AI will be inaccurate. In payroll it's definitely important to ensure that the payroll foundation is correct first."

Jackson, co-founder & CFO, Workpay

This is where payroll differs from the rest of HR. Elsewhere, a bad output is embarrassing and recoverable. In payroll it lands in someone's bank account, and it lands there in every country at once.

The sequence matters more than the tooling. Fix the foundation, then automate it.

Where AI already earns its place

Once that foundation holds, Thomas named two applications that are working today.

The first is regulatory change.

"A government publishes a new regulation. You can use AI to interrogate that regulation, understand what's changed from the previous year, so that your experts can focus on those particular areas of change rather than manually analysing thousands of pages of documents. The AI can focus the experts on the three or four important changes."

Thomas Otter, Acadian Ventures

The second is the employee's own question.

"A conversational layer on top of payroll information gives employees the ability to intelligently question their pay information. The employee is paid less one month than the other month, and they can ask the AI: why was I paid less this month? And the AI can look at your payroll and say, well, last month you worked more overtime."

Thomas Otter, Acadian Ventures

Both are useful. Both are only as good as the payroll record behind them. Jackson added the obvious guardrail: an employee asking about their own payslip should never get an answer about a colleague's.

The circular saw

The cheaper AI makes software to build, the more tempting it becomes to build your own. Thomas was blunt about where that ends.

"Of all the systems in the world that you do not want to build yourself, payroll is the highest on that list."

Thomas Otter, Acadian Ventures

"The metaphor that I use is that AI vibe-coding tools are like the circular saw. The circular saw revolutionised the furniture industry, allowing people to build furniture far cheaper and far more accurately than they could in the past. It also enabled some hobby people to build great things at home. Most of us, though, who bought these tools used them once, built something that was not very useful, and then put the sheet over the saw and left it in the garage. Some poor unfortunate people cut their fingers off."

Thomas Otter, Acadian Ventures

The tools are genuinely good. That is what makes the temptation dangerous. Building your own payroll now looks possible in a way it did not two years ago, while the cost of getting it wrong has not moved at all.

Jackson gave the technical reason it goes wrong.

"A similar statement in the regulations of one country can be interpreted differently in another country. So if you instruct an AI to do the calculation for you, it might take the calculations of country A and make the same calculation for country B, while the experts in country B interpret that same statement differently."

Jackson, co-founder & CFO, Workpay

Generic software is getting cheaper to replicate. Knowing how a specific regulator reads a specific clause is not.

From automation to prediction

The more interesting shift is not in processing. It is in what payroll data lets you see forward.

"The CFO of the future needs systems that allow them not just to know what was my payroll cost last month. They need systems that can tell them what will be my cost in three months, six months from now."

Jackson, co-founder & CFO, Workpay

"They need a system that will tell them: if I'm hiring 20 employees in the next three months for engineering, three in support and the rest in the other departments, what would be that increase in cost, without doing a lot of work."

Jackson, co-founder & CFO, Workpay

A system that already holds salary bands, statutory costs per country and the hiring plan should be able to produce that number in minutes. In most companies it still takes a week of spreadsheet work and produces a figure nobody fully trusts.

Automation makes last month cheaper to process. Prediction changes what you can decide this month.

Answers before robots

On employee experience, Thomas made the case for a narrower and more honest ambition.

"Most of the time, employees aren't interacting with HR. It's not as if you wake up every morning and think, as an employee, how do I interact with HR today? But there are moments in your life when that interaction is crucially important. An example might be: you're about to have a child. And for you personally, that's the most important thing that you've ever done. And so you want in that moment great support from your employer."

Thomas Otter, Acadian Ventures

"This is when AI, done properly, can augment and support whatever HR is doing. I don't have to sit on the phone in a queue or go and make an appointment. I can get that information instantaneously from AI."

Thomas Otter, Acadian Ventures

And when the person does need a colleague, that colleague already has the context. Reliability on the handful of days that matter most to someone beats sophistication on the other three hundred.

The adoption problem nobody budgets for

The strongest line in the episode was also the least technical.

"If employees don't want AI to work, it won't work. It doesn't matter how clever AI gets, clever employees will figure out a way to make it not work. So you have to bring your people along with you. And if you keep standing up and saying, hey, we're going to deploy AI and it's going to mean that we will take away 60% of the jobs in the company, then guess what? Your AI project is going to fail."

Thomas Otter, Acadian Ventures

Jackson made the process version of the same argument.

"You really have to fix the fundamental process first before you can apply AI into it. If the process is not clear internally, AI will not come and make that process clearer."

Jackson, co-founder & CFO, Workpay

Buying an agent does not create clarity. It runs whatever you already have, faster.

What to do with this

Three things come out of the episode for anyone running HR or payroll across multiple markets.

Fix the data before you buy the tool. An AI layer on unreliable payroll data produces confident errors at scale.

Buy local depth rather than building it. The interpretation of a regulation in one country does not transfer to the next, and that is precisely the part a general-purpose model gets wrong.

Sell the redesign, not the reduction. If your AI story is about removing headcount, the people who have to make it work will make sure it does not.

Listen to the full conversation

This is the short version of a longer conversation with Thomas Otter on AI, payroll and what actually changes about work.

Listen to Episode 6 of the Workpay podcast

If you are weighing up what to automate in payroll and what to leave alone, talk to us.

‍