AI is already in your hiring, learning, analytics and planning. The capability a people team needs is judging where it helps, where a person must decide, and being able to show which is which.
Data & AI Skills for People & HR Teams
By James Cotton · Last updated · 7 min read
Part of our topic guides on AI Skills for Business and AI Governance & Data Strategy.
By James Cotton, Founder, iO-Sphere
Four places AI already sits in a people team
On an ordinary week, AI in a people function drafts the job advert and sifts the applications, suggests what someone should learn next, reads the free text in the engagement survey, and runs the headcount scenarios for next year. Most of it arrived inside software the team already had.
Those four places (hiring, learning and development, people analytics, workforce planning) cover most of what a people function does to and for the people it employs.
Hiring, learning, analytics, planning
Hiring. AI can draft the advert, strip the jargon out of it, sort applications against criteria you set, and summarise interviews.
The moment a tool ranks or filters candidates, it can act on a pattern from your past hires that you would never write into a policy. It does that silently, and to everyone at once. A person owns the shortlist; the tool prepares it.
Picture a people team that switches on the scoring feature in its applicant tracking system for a high volume role, and halves its sifting time. Three months later a rejected candidate asks how the decision was made, and nobody can say what the score weighed or who checked it.
Learning and development. AI can map the skills a role needs, suggest a learning path, draft and translate material, and mark practice exercises. A recommended module is help with somebody's week.
Once the output becomes a verdict on how far along someone is, it has become a decision about their career. Decisions about who is ready for promotion, whose development has stalled and who goes onto a formal plan stay with a person.
People analytics. AI can surface patterns in engagement, sentiment and turnover that a human would take weeks to find.
These models learn from what happened before, so they can repeat old bias, and a flight risk score or a performance prediction is an estimate built from that history. The pattern is a reason to go and look, not a reason to act on a person.
This is where the data half of the job does most of its work, and the skills involved are ordinary: knowing what a dashboard's number is a share of, telling a trend from noise, and knowing what your HR system's export does and does not contain.
Small numbers need particular care, because a team of nine produces percentages that swing on one leaver. A team without those habits reads its own dashboard wrong long before any AI is involved.
Workforce planning. AI can run the forecasts and the scenarios: the headcount maths, the skills supply models, the shape of a restructure on paper.
Whose role changes and where any cuts land stay with people who can be asked to account for them. A model shows you the options. It cannot carry the responsibility for the one you choose.
The law behind a decision about a person
Two bodies of law sit behind what your function decides about someone.
The Equality Act 2010 prohibits direct discrimination (section 13) and indirect discrimination (section 19), and indirect discrimination needs no intent. Section 39 applies those prohibitions to employment, covering recruitment, promotion and dismissal.
Data protection law covers the automated part. UK GDPR Article 22, as amended by the Data (Use and Access) Act 2025, deals with a decision based solely on automated processing that has a legal or similarly significant effect on a person.
The ICO's guidance on automated decision-making and profiling, updated on 31 March 2026, says such a decision is permitted only with safeguards. You have to tell the person, hear their representations, provide human intervention, and let them contest the decision.
Stricter limits apply where special category data is involved. The guidance is still marked as draft, and it is the ICO's current published position.
Amazon built an experimental CV screening tool and found it penalised CVs that contained the word "women's". Tools of that kind rarely explain their scores, the team could not be sure it would not find other ways to discriminate, and Amazon scrapped it (Reuters, Jeffrey Dastin, 10 October 2018).
What the capability looks like in a team
For any AI touching a decision about a person, someone can show three things: what the tool did, what a person decided, and how they would know if it were going wrong.
The third is where most teams are thin. For a screening tool, the check is whether it advances different groups through its stages at similar rates. Check the outcomes yourself, whatever the supplier's fairness claims say.
If one group gets through far less often and there is no job-related reason for it, the tool is having a different effect on different groups. Examine it before the next intake runs through. The same habit applies to an attrition model: what data trained it, and where it would mislead.
These are decisions you can be asked to justify, to an employee, an auditor or an employment tribunal, and the check you ran is the answer you give.
Who owns AI across the organisation as a whole is a separate question, and we cover it in who is responsible for AI governance.
How a people team builds this
A tool already in use is the urgent case, and team training reaches everyone who touches it within weeks.
Our Data & AI fluency team training runs from a half-day to multi-week, built around your team's own hiring, learning and analytics tasks, and priced per engagement. iO-Sphere's coaches deliver it. The wider menu sits on our business training page.
Depth is for the one or two people who will own the decisions the law covers: what a screening tool is allowed to do, when an automated decision needs a person inside it, and what gets recorded.
Our Data & AI Governance apprenticeship runs on the Level 4 Data Protection and Information Governance Practitioner standard (ST0967). Every one of those decisions is made on personal data, which is what the standard is about.
There is no apprenticeship standard called AI governance. This is the funded route into it, and several providers deliver the standard.
The whole programme runs about 18 months: roughly 15 months of training, then three months of end-point assessment. The standard itself sets a typical training period of about 18 months before that final assessment.
For most people teams the training fee is nothing or £500. It is nothing when the apprentice is aged 16 to 24 at the start, at any employer.
It is £500 (5% of the £10,000 funding band) when they are 25 or over and the employer's pay bill is under £3 million. A levy-paying employer uses its levy account, and pays £2,500 (25%) only if that account is empty.
How the money reaches an employer sets out the mechanics, and funded data governance training covers the route in more detail.
If you need one person brought up to speed on a single HR tool for an afternoon, a vendor session will do that better than any programme of ours.
If your people data is incomplete, scattered or not trusted, AI skill will not fix it. Get the data right first, then build the judgement on top.
iO-Sphere trains the people in your function, and does not run the function or audit your HR systems for you.
Frequently asked questions
Can we just ban AI in hiring?
You can, and some people teams draw the line at the sift. A blanket ban is simple to state and hard to hold, because it also covers work that carries very little exposure, such as drafting an advert, arranging interviews or tidying up notes a person wrote. Teams that make a ban stick tend to write down which steps AI may touch and which it may not, and give one person the job of keeping that list current as the software changes.
What about AI nobody in the people team chose, like a general assistant used for interview notes?
Treat it the same way. What matters is the decision the output feeds, and the route a tool took into the building makes no difference to that. Two things help. Ask your suppliers to tell you when a feature that scores, ranks or predicts something about a person goes live in a system you already run, because those arrive by software update and get switched on without anyone in the people team choosing them. Then ask your own team what they already use for job descriptions, notes and survey summaries.
We hire a handful of people a year. Is this worth building at our size?
The exposure comes from the decision, so it does not scale down with volume: one rejected candidate can ask the same questions as a hundred. What changes at small scale is the checking, because with a few dozen applications a year the pass rates will not tell you anything reliable. The check that works at that size is to run the tool across a set of past applications you judged yourself, and look closely at every case where it disagrees with you.
Who should hold the funded apprenticeship place in a people team?
The person who will be asked to approve tools and produce the evidence afterwards. That is often an HR operations or people analytics lead, and sometimes whoever already handles subject access requests and retention. The standard behind it is a data protection and information governance one, so it suits someone who will own records and decisions for the function. It is an apprenticeship, so it needs protected training time inside the working week, and someone who will still be in the role in eighteen months.
Upskilling a whole team?
Tailored data and AI training for organisations, from data literacy to technical upskilling, built around your team's real work.