An employee earns a promotion. The HR system logs it immediately. Payroll updates its records the following month. Workforce planning catches up a quarter later. Finance tracks headcount in a separate spreadsheet.
Every system is functioning exactly as it was designed to. Yet the company still lacks a single answer to a basic question: what does that employee actually cost?
Now add AI. The tool analyses workforce costs and instantly recommends a compensation band. The output is confident. The writing is polished. But the answer is completely wrong.
This is the problem I see increasingly as organisations adopt AI: they mistake activity for results. But if those insights are based on inaccurate, incomplete or disconnected information, what is the value?
The AI has done its job. It has taken the information available to it and reached a conclusion. The problem is not necessarily the model, or even how it is being used. The problem is the data. The model is interpreting broken information.
Adoption is the wrong metric
The question of whether firms should be using AI has largely been settled. McKinsey’s latest global study noted that 88 percent of businesses now cite frequent AI use in at least one corporate area, up from 78 percent the previous year.
More revealing is what comes next: only 39% report any clear impact on their bottom line.
That gap gets treated as a software problem or a talent challenge. In my experience, it is more fundamental than either. It is a data problem.
I have worked in fast-growing businesses, and I have seen the same cycle repeat. Software doesn’t naturally simplify an organisation as it grows. Instead, companies add tools as they expand. Each tool solves a real and distinct problem, until the set of fixes has quietly built a much bigger one.
HR operates on one system. Finance operates on another. Payroll, recruitment, workforce planning and compliance each have their own processes and data. Individually, they work. Together, they can make it remarkably difficult to answer even the most basic questions about the organisation.
AI does not automatically fix that problem. It amplifies it.
If your information is accurate, connected and well governed, AI can be incredibly powerful. It kills the ‘no joy’ tasks, identifies patterns people simply do not have the time to find, and gives employees better insights faster. If your data is fragmented, AI can produce answers that sound certain without being reliable. In payroll, compliance or workforce planning, that is not a minor inconvenience. It is a significant risk.
What good foundations actually look like
The answer is not to slow down AI adoption. It is to build the conditions that allow AI to work properly.
That begins with data moving between tools instead of being manually entered, reconciled and corrected by people who should be spending their time elsewhere. It means putting controls in place when data enters the organisation, rather than discovering months later that two systems tell different stories and someone has to work out which one to trust.
Then there is transparency, which I believe is one of the areas most organisations overlook. If a system informs a leader that attrition is rising, that leader ought to be able to find where the results came from. If AI recommends a compensation decision, they should be able to trace that recommendation back to underlying data. Without that visibility, you are not really making a decision. You are accepting one that has already been made for you.
The need to challenge
But explainability will only get you so far. Being able to trace an answer is worthless if nobody is willing to question it.
This is why AI adoption is as much a leadership problem as a technical one.
A polished AI-generated answer carries a strange kind of authority. It is fast, confident and presented without the uncertainty a human might naturally express. Challenging it requires someone to say, often in front of more senior colleagues, that the facts do not match what they are seeing on the ground.
Most organisations are not set up to make this easy. Junior employees are most likely the ones closest to the data and the least likely to speak up. If they have seen a previous challenge dismissed, they will not raise the next one. And the business loses the early warning it needed.
That is why the instruction I give my own teams is simple. If a result does not match what you are seeing in real life, say so before we act on it.
AI can surface the information. It cannot replace human judgement. Leaders, in particular, have to demonstrate that challenging an AI-generated answer is not failure. It is part of using the technology responsibly.
Where human judgement still matters
The best use of AI is to give people better information and more capacity to exercise their judgement.
That is especially true in HR. People don’t go into HR because they want to spend their days fixing spreadsheets, reconciling systems and searching for patterns across disconnected data. They go into the profession to develop people, build better organisations and create workplaces where employees can do their best work.
This is also why the language businesses use around AI matters. If the conversation is only about productivity, efficiency and headcount, employees will understandably see AI as a threat. And people who see it as a threat are less likely to tell you when it is wrong.
Change the conversation. Talk about removing the work nobody wanted to be doing in the first place. Talk about giving people better information, more time and greater capacity to focus on work that requires human judgement. You get a very different response.
The people behind the numbers
Every number produced by a workforce system represents a real person. A salary. A promotion. A restructuring decision. That is the standard we should hold our technology to.
AI can be remarkably powerful, but only when the foundations beneath it are reliable. Good data, connected systems and strong governance make better AI possible. A culture that encourages people to question its outputs makes better decisions possible.
Get the foundations right and AI can make people far better informed and far more effective.But the final decision still belongs where it always has: with the person.
