The team already drafts and sorts with AI. More of what reaches its manager for approval was shaped partly by a tool, and it still goes out on the manager's authority.
AI Training for Managers: Answering for Your Team's AI Work
By James Cotton · Last updated · Sources checked · 9 min read
Part of our topic guide on AI Skills for Business.
By James Cotton, Founder, iO-Sphere
How managers rate the training they get
Of 823 managers whose organisations use or plan to use AI, 25% rated their organisation's "management training on AI adoption and driving AI-enabled change" as "good" or "very good" (19% good, 6% very good), in a poll by the Chartered Management Institute (CMI), which offers management qualifications, its own AI ones among them. Only two of the eleven items, on what their organisation has in place for AI, scored lower.
That 25% is managers' own rating of the training their organisation gives them; it does not measure how much training there is. CMI's June 2026 report, Artificial Intelligence; Real Leadership, rests on a CMI poll of 1,019 UK managers carried out between 20 February and 9 March 2026, plus 13 expert interviews; the 823 are the managers in that poll whose organisations use AI or plan to.
Answering for AI work assumes that someone owns it. On the same chart and the same base of 823, 29% rated "defined ownership or accountability for AI initiatives" in their organisation as good or very good.
The report's text adds two findings about managers themselves, with no base printed for either. In that text, 12% of managers rate themselves "very confident" in their ability to support their teams' AI adoption. And when the report's text ranks what holds managers back in their own use of AI, training comes first: "a systemic lack of training (38%, rising to 40% among managers aged 35 or over)".
The same tools, a different job
A team member who asks an AI tool for a first draft has a contained job with it: check the draft, correct it and stand behind it before anyone else relies on it. Their manager uses the same tools for another job. The manager answers for what the whole team sends out, most of which they did not write, including work done with AI agents.
Zoe Olivia Bradley, a Senior Programme Manager at NHS England quoted in CMI's report, puts the manager's side in one line: "AI can accelerate work, but it does not remove accountability." Her team uses AI to draft briefings and summarise complex information, with its outputs "always reviewed, challenged and refined".
CMI argues that "Accountability for AI risk is shifting from IT to line-of-business managers". The checking that comes with it is uneven: in the report's text, 24% of managers do not trust AI outputs and 19% do not know how to validate the results. Framing the question and challenging what comes back is the first half of a manager's skill with AI output.
Far fewer managers are confident with the tools that act on their own. In CMI's poll of 1,019 managers, 26% were "very confident" using generative AI themselves, and 10% using agentic AI, the "autonomous AI agents" that CMI pictures managers orchestrating alongside human staff. Both figures are about using each kind of AI personally.
Our AI Capability Framework orders its levels by who depends on the work. It draws the line between level 2, independent, where a person does it reliably on their own work, and level 3, accountable, where they do it on work other people depend on and are answerable for the result, which is where a manager's AI work sits: the team relies on what the manager approves.
When a tool helps decide something about a person
Drafting a report with AI seldom means deciding anything about a colleague. A manager's job does, from a warning to a performance rating, and automated tools can support those decisions. The rules on automated decisions about people, UK GDPR Articles 22A to 22D, replaced Article 22 for decisions taken from 5 February 2026.
The regulator, the Information Commission (the ICO), covers a tool that only supports a decision in its guidance on monitoring workers. That guidance is now under review. Where an automated system is used as a decision-supporting tool, the people making the decision should check its recommendation and "should not just routinely apply" it to workers.
They should be actively involved, with "meaningful" influence on the decision, including the "authority and competence" to go against the recommendation. They should also weigh up and interpret it, considering all the input information and taking other factors into account.
The guidance's example is a courier warned for late deliveries. The courier service's HR manager checked the complaints, reviewed the vehicle's tracking data and spoke to the worker before deciding to issue the warning: a decision with human involvement. The same chapter adds two things the employer "must" do: check regularly that the system is working as intended, and not disadvantage workers who ask for human intervention.
The regulator's draft guidance on automated decision-making, last updated on 31 March 2026 and now out for consultation, adds a training point: the human involved should be "suitably trained and qualified to understand the system's logic, outputs, limitations, and risks".
A course on using the tool teaches what to ask it and how to read the answer. For this decision the manager needs two things such a course leaves out: what the system weighs and what it cannot see, so that checking its recommendation means testing it, and the standing in their organisation to overrule it when the other information points elsewhere.
Looking at how the team uses AI
Each team member's use of an AI tool can show up in a usage report. Our reading is that a manager who opens that report is in the territory of the same monitoring guidance, still under review, even though the guidance does not mention dashboards of staff AI use.
The guidance's chapter on data protection and monitoring workers defines monitoring workers as "any form of monitoring of people who carry out work on your behalf", with analytics that infer performance and "productivity tools which log how workers spend their time" among its examples.
The employer must be clear about its purpose, select the least intrusive means and document why it is monitoring; it should not monitor workers "just in case". Before monitoring likely to cause high risk, it must carry out a DPIA, a data protection impact assessment, and the examples of high risk include "monitoring that may result in financial loss (such as performance management)". Workers must be told before monitoring begins.
Overall responsibility for monitoring should rest "at the highest senior management level", and anyone running it should be briefed on data protection law. The employer should seek and document the views of workers or their representatives first; covert monitoring is unlikely to be justified in most usual circumstances, and consent "is not usually appropriate in the employment context".
So before opening the first usage report, a manager needs to know what it is for, whether a lighter check would answer the same question, whether a DPIA was done, whether the team was asked and told, and who at senior level owns the monitoring.
Whether the team takes up AI at all
For a team member, whether to try AI is a choice made task by task. In large surveys, use is far higher where the employer or manager encourages it, provides the tool and trains people. Every figure below is an association.
Among US workers surveyed in 2024, Bick, Blandin and Deming (Management Science, 2026) found that 86% of those whose employer encouraged generative AI had adopted it, against 13% where the employer did not. Their working paper calls the estimate "not necessarily causal".
Humlum and Vestergaard's NBER working paper, revised in March 2026 and not peer-reviewed, covers Danish workers in late 2024, with employer initiatives as the workers perceived them. Where the employer neither encouraged AI chatbots, provided one nor trained people, about 41% had still used them at work.
Implemented individually, encouragement went with take-up of 76% and training with 69%. The figure for an enterprise chatbot was 57%. With all three in place, 93% had used chatbots at work and 28% used them daily.
Gallup, a US polling firm, surveyed 19,043 US employees in May 2025. In organisations implementing AI, 28% strongly agreed that their manager actively supports the team's use of it. Those who did were 2.1 times as likely to use AI a few times a week or more and 6.5 times as likely to strongly agree the tools are useful for their work.
In the UK, CMI's report finds managers in organisations further on with AI more likely to agree that employees feel supported by their manager to use AI tools (66%, against 47% in early-stage organisations). CMI calls it a correlation; these are managers' reports about employees, and subgroup bases are not printed.
Of the 717 managers in CMI's poll whose organisations use AI, 64% agreed that their senior leadership encourages staff to experiment with AI tools, and 13% strongly agreed that their senior leaders actively experiment with the tools themselves.
The gov.uk employer guide What works for AI upskilling in the UK, from the Department for Work and Pensions and Skills England and updated on 27 July 2026, gives its author's opinions. It lists the pitfall "Leadership support for AI training is weak or inconsistent". Training, it says, is less effective when leaders do not prioritise it, link it to organisational goals, "or model its use".
Bradley, in CMI's report, draws the consequence for managers: "Managers need to be confident using it themselves, otherwise people will use it informally and without the right safeguards."
What a manager's own training is built on
We recommend that a manager learns the AI tools alongside the team, and also has training of their own, built on the team's live work and the decisions the manager takes about it. A manager whose team drafts client reports with AI learns to test those reports; one whose team uses a tool to sort cases learns what that tool weighs.
Managers in CMI's report ask for both kinds. Asked what would help them lead their teams through AI, 52% cited a need for foundational technical AI skills training, and 47% wanted AI management and leadership training covering strategy, change management, culture and decision-making, the report's text says. And 37% said they need "guidance on how to upskill their teams and address capability gaps".
Our team training for a manager and their team is built on the team's own work and decisions, so the manager learns the tools on the same material as the team and practises the decisions that are theirs alone.
The measure of a manager's training is the team's work: level 4 in our framework, multiplying, where other people's work is better because of them. What happens around a course, the manager's own part included, decides whether training reaches that work.
The piece of work you would have to explain
Everyone on the team can open the same tool. What differs is what each person answers for: the team member for their own draft, the manager for what the team sends out, for the decisions a tool helped shape and for the reports they chose to read.
Pick one piece of work your team produced with AI's help, or one decision about a person that a tool helped you make: could you answer for it today, saying how it was produced, what the tool weighed and why you let it stand?
Frequently asked questions
Can the levy pay for a manager's AI training?
The three general management apprenticeships, Team leader (Level 3), Operations manager (Level 5) and Chartered manager (degree) (Level 6), are not government funded for new starts from 1 September 2026. Where leading AI change is the manager's job, AI Transformation (Level 4) can be funded if that existing job matches its standard; who pays turns on age and whether the employer pays the Growth and Skills Levy, which funds apprenticeships.
Is there a recognised qualification in managing AI?
CMI's Leadership for AI qualifications sit at Levels 3 ("Managing AI Adoption"), 5 ("Leading AI Transformation") and 7 ("Strategic Leadership of AI"), with a syllabus CMI's report puts at 80% management and leadership, 20% technical. The Ofqual register lists the Level 3 Diploma (610/7233/3) at 158 guided learning hours and 400 hours' total qualification time, examined in writing. The choice turns on whether the employer wants a regulated credential.
Is there an official code on using AI to manage staff?
Acas's consultation on a draft disciplinary and grievance Code (30 July to 23 September 2026) asked whether AI in those processes needs addressing; responses are under review. Tribunals must consider the Code and, in some claims, can adjust an award by up to 25% for unreasonable failure to follow it. Regulations in force since 12 May 2026 require a statutory code on AI and automated decision-making, with no deadline.
Upskilling a whole team?
Tailored data and AI training for organisations, from data literacy to technical upskilling, built around your team's real work.