Where AI can help across the employee lifecycle, and what to weigh at each stage.
Where AI shows up in HR, from planning a role to saying goodbye. For each stage: what generative AI and AI agents can do, how much value and risk each brings, where a person must stay in charge, and who should own it.
Where AI can help: Generative AI: Summarise headcount, turnover and absence reports into a one-page brief for a planning meeting.; Generative AI: Draft and tidy role profiles and job descriptions so they're consistent across teams.; Generative AI: Write "what if" scenario narratives (for example growth, a new site or a hiring freeze) from figures you provide.; Generative AI: Pull skills themes out of job descriptions to show where skills gaps might be.; AI agent: Gather headcount, vacancy and budget figures from HR and finance systems into a planning pack.; AI agent: Flag roles that have been vacant longer than an agreed time, or teams close to budget limits.
GenAI: Lower risk; GenAI: Lower risk; GenAI: Take care; GenAI: Take care; AI agent: Take care; AI agent: Take care; Both: Higher risk; Both: Higher risk
Value: Medium
Watch for: scenarios that quietly become decisions about real jobs; figures nobody has checked.
Start here: summarising the data a planning meeting already uses.
Keep a person in charge of: any decision about roles, structures or headcount.
Owner: HR director or workforce planning lead, with finance.
View as: Manager
Here are my notes on my team's workload and the changes coming next year, with no names: [paste]. Turn them into a half-page summary for a planning conversation with HR: what's changing, what it means for the team's capacity, and three options with the pros and cons of each. Don't invent any figures.
View as: Employee
Here's the announcement about changes in our department: [paste]. Explain in plain English what's changing and when, then suggest five questions I could ask my manager or HR. Don't guess at anything the announcement doesn't say.
In practice
Illustrative example · fictional Leyfield Group
THE USE
Generative AI: Summarise headcount, turnover and absence reports into a one-page brief for a planning meeting.
WHAT HAPPENED
The planning team asked AI to summarise quarterly turnover by department for the leadership meeting. It merged two departments with similar names, Customer Service and Customer Success, so Customer Service's turnover looked about twice as high as it really was.
WHAT THEY CHANGED
Department names now go in as a fixed list alongside the data, and every figure in the brief is checked against the source report before it's shared.
RED FLAGS
AI scenarios shared without saying they're scenarios; nobody can say where a figure came from.
WHAT TO TELL YOUR PEOPLE
“We use AI to help summarise workforce data for planning. Decisions about roles are always made by people.”
Before you switch it on
Where does the data come from, and who keeps it accurate?
Can we see how each figure was produced?
Who reviews outputs before they reach decision-makers?
Has a data protection impact assessment been done if personal data is used?
How will scenarios be labelled so nobody mistakes them for decisions?
Where AI can help: Generative AI: Write clearer, more inclusive job adverts, and check existing ones for jargon or biased wording.; Generative AI: Draft structured interview questions and scoring guides linked to the role's requirements.; Generative AI: Summarise an application for the recruiter, who still reads it in full.; Generative AI: Draft candidate emails: invitations, updates and kind, clear rejections.; AI agent: Find and pull together a candidate's application, interview notes and scores from the applicant system for the hiring manager.; AI agent: Schedule interviews across diaries and send confirmations and reminders.; AI agent: Answer candidates' common questions about the process and the role.; AI agent: Move applications between stages once a recruiter has made the decision.
GenAI: Lower risk; GenAI: Lower risk; GenAI: Lower risk; AI agent: Lower risk; GenAI: Take care; AI agent: Take care; AI agent: Take care; AI: Higher risk; AI: Higher risk; AI: Higher risk; AI: Avoid; AI: Avoid
Value: High
Watch for: bias in screening; AI ranking or rejecting people; candidates not knowing AI is involved.
Start here: clearer, more inclusive job adverts.
Keep a person in charge of: shortlisting and every decision about a candidate.
Owner: talent acquisition lead, with HR and data protection.
View as: Manager
I'm hiring a [role] for my team. Here are the main tasks and the skills that matter most: [paste]. Write eight structured interview questions, two for each key skill, with what a strong, an adequate and a weak answer would include.
View as: Employee
Here's a job advert: [paste]. Ask me six likely interview questions one at a time. After each answer I give, tell me what was strong and what I could make clearer. Don't write my answers for me.
In practice
Illustrative example · fictional Leyfield Group
THE USE
Generative AI: Summarise an application for the recruiter, who still reads it in full.
WHAT HAPPENED
Summaries of applications for a project manager role kept leaving out career breaks and returner programmes, so strong returners looked less experienced than they were.
WHAT THEY CHANGED
The prompt now asks the summary to list every gap and what the candidate said about it. Recruiters read the full application for anyone with a gap, and summaries are never used to sort or reject anyone.
RED FLAGS
recruiters only reading AI summaries; "the system rejected them" being said out loud.
WHAT TO TELL YOUR PEOPLE
“We use AI to help draft our job adverts and schedule interviews. Every application is reviewed by a person.”
Before you switch it on
Can we see why it summarised or ranked a candidate the way it did?
Has it been tested for bias with our own roles and applicants?
Can it reject or progress anyone without a person approving?
How will candidates be told, and how can they ask for a person?
Has a data protection impact assessment been completed?
Where AI can help: Generative AI: Draft welcome guides and first-week plans from your current policies.; Generative AI: Answer new starters' common questions using only approved documents.; Generative AI: Turn long policies into short, plain-English summaries for induction.; AI agent: Find the right policy, form or contact for a new starter's question and point them to it.; AI agent: Send reminders before day one and chase missing forms or documents.; AI agent: Set up onboarding tasks across HR, IT and facilities, such as equipment, system access and induction sessions.
GenAI: Lower risk; GenAI: Lower risk; GenAI: Lower risk; AI agent: Lower risk; AI agent: Take care; AI agent: Take care; AI: Avoid
Value: High
Watch for: out-of-date policy answers; personal data in the wrong tool; agents acting on incomplete records.
Start here: answering common new-starter questions from approved policies.
Keep a person in charge of: right-to-work checks, contracts and anything sensitive.
Owner: HR operations, with the hiring manager.
View as: Manager
I have a new [role] starting in my team on [date]. Here's what our team does and who they'll work with: [paste, no personal details]. Draft a first-week plan: a short list for each day, including who they should meet, and a check-in with me each day.
View as: Employee
I'm starting as a [role] next week. Suggest ten questions to ask my manager in my first week, grouped into: the work, the team, how I'll be supported, and what good looks like after three months.
In practice
Illustrative example · fictional Leyfield Group
THE USE
Generative AI: Draft welcome guides and first-week plans from your current policies.
WHAT HAPPENED
A welcome guide for new warehouse starters quoted last year's holiday allowance, because an old policy was still sitting in the shared folder used as the source.
WHAT THEY CHANGED
Guides are now drafted only from the current policy library, each guide shows a "policies checked on" date, and HR operations signs off every new version.
RED FLAGS
new starters getting different answers from the guide and their manager.
WHAT TO TELL YOUR PEOPLE
“Some answers in your welcome guide are drafted with AI from our current policies. If anything looks wrong, ask your HR contact.”
Before you switch it on
Which documents will it answer from, and who keeps them current?
What personal data will it see, and is the tool approved for it?
If an agent chases documents, what happens when someone doesn't reply?
Who signs off guides before new starters see them?
How will new starters know which answers come from AI?
Where AI can help: Generative AI: Answer routine policy questions (holiday, expenses, hybrid working) from your published policies.; Generative AI: Draft replies to HR enquiries for an adviser to check and send.; Generative AI: Summarise a long enquiry history so an adviser can pick it up quickly (never for employee relations cases).; Generative AI: Keep FAQs and help articles up to date when a policy changes.; AI agent: Find the relevant policy or document for a question and answer from it, with a link to the source.; AI agent: Look up an employee's own information when they ask, such as their leave balance or payslip dates.; AI agent: Route enquiries to the right team and log routine requests and personal data changes.
GenAI: Lower risk; GenAI: Lower risk; GenAI: Lower risk; AI agent: Lower risk; AI agent: Take care; AI agent: Take care; GenAI: Take care; GenAI: Take care; Both: Higher risk; AI: Avoid; AI: Avoid
Value: High
Watch for: confident wrong answers; one person seeing another's data; sensitive cases handled by a bot; AI summarising or drafting anything in an employee relations case; a chatbot answering questions about an open grievance.
Start here: drafts an adviser checks before sending.
Keep a person in charge of: grievances, disciplinary matters and investigations, health, wellbeing and anything sensitive.
Owner: HR service delivery lead.
View as: Manager
Here's our [policy name] policy: [paste]. Summarise it for a manager in under 200 words: what employees are entitled to, what managers must do, and the three questions people ask most. Quote the policy section for each point.
View as: Employee
I want to ask my manager about [topic, for example flexible working]. Here's the policy: [paste]. Help me prepare: what the policy says, what I should include in my request, and three questions I might be asked.
In practice
Illustrative example · fictional Leyfield Group
THE USE
Generative AI: Draft replies to HR enquiries for an adviser to check and send.
WHAT HAPPENED
Advisers got through routine questions noticeably faster. Then one draft about hybrid working described a manager's informal practice as if it were the policy, and it nearly went out.
WHAT THEY CHANGED
The prompt now says "use only the published policy wording and quote the section", and advisers check the quote against the policy before sending.
RED FLAGS
advisers sending drafts unread; employees asking sensitive questions to a bot because it's quicker.
WHAT TO TELL YOUR PEOPLE
“We use AI to help draft replies to your HR questions. An adviser checks every reply before it's sent.”
Before you switch it on
How does it know when a question is sensitive and needs a person?
Can it only ever see the asking employee's own information?
What's the fallback when it doesn't know?
How will employees know they're using an AI assistant, and how to reach a person?
Where AI can help: Generative AI: Turn existing training material into summaries, quizzes and practice scenarios.; Generative AI: Draft a development plan from notes of a conversation between a manager and an employee.; Generative AI: Suggest learning that fits someone's goal or a skills gap.; Generative AI: Describe career paths, and what each next role involves, in plain English.; AI agent: Search the learning catalogue for courses that match a goal or skills gap.; AI agent: Build learning plans, enrol people on courses and send reminders.; AI agent: Suggest internal roles or projects that match someone's skills profile.
GenAI: Lower risk; GenAI: Lower risk; AI agent: Lower risk; GenAI: Take care; GenAI: Take care; AI agent: Take care; AI agent: Take care; AI: Higher risk; AI: Higher risk
Value: Medium
Watch for: learning that looks complete but isn't; skills data reused for selection; the same people always being surfaced for opportunities.
Start here: practice material and plain-English summaries.
Keep a person in charge of: promotion, succession and selection decisions.
Owner: learning and development lead, with talent management.
View as: Manager
Here are my notes from a development conversation with a team member, with no names: [paste]. Turn them into a development plan: two or three goals, the actions for each, the support I'll give, and a date to review progress.
View as: Employee
I'm a [current role] interested in moving towards [target role]. Explain what that role typically involves, the skills I'd need to build, and three practical steps I could take in the next six months.
In practice
Illustrative example · fictional Leyfield Group
THE USE
AI agent: Suggest internal roles or projects that match someone's skills profile.
WHAT HAPPENED
Matching kept suggesting the same small group of people, because fully completed skills profiles scored higher. Part-time and frontline staff, who had less time to fill in profiles, rarely appeared.
WHAT THEY CHANGED
Every internal role is now also advertised openly, matches are shown to employees as suggestions they can choose to follow up, and the team checks each quarter who is being suggested.
RED FLAGS
the system didn't suggest you" used as a reason; skills profiles becoming a test nobody agreed to.
WHAT TO TELL YOUR PEOPLE
“We use AI to suggest learning and roles you might like. It's a suggestion, not a selection, and every opportunity is open to everyone.”
Before you switch it on
What data does it use to suggest people, and could that favour some groups?
Will skills data be used for anything beyond suggestions?
Can employees see and correct their own skills profile?
How will internal roles still reach people the system doesn't suggest?
Where AI can help: Generative AI: Help a manager organise their own notes into themes before they write a review.; Generative AI: Suggest clearer, measurable wording for goals the manager and employee have already agreed.; Generative AI: Draft agendas and questions for check-ins.; Generative AI: Help a manager prepare for a difficult feedback conversation.; AI agent: Pull together the goals and check-in notes already recorded in the system, ready for a manager's review conversation.; AI agent: Send check-in and review reminders, and gather the feedback requests a manager asks for.
Watch for: biased or invented summaries; reviews that sound fair but aren't evidenced.
Start here: helping a manager organise their own notes before writing.
Keep a person in charge of: writing, scoring and deciding any review or rating.
Owner: line managers, with HR business partners.
View as: Manager
I need to give feedback to a team member about [issue, described without names]. Help me plan the conversation: how to open it, the specific examples to mention, questions to ask them, and how to agree next steps. Keep it fair and focused on behaviour, not personality.
View as: Employee
Here are my notes on what I worked on this year: [paste]. Turn them into a short summary for my review: my main achievements with evidence, what I learned, and two goals for next year. Use only what's in my notes.
In practice
Illustrative example · fictional Leyfield Group
THE USE
Generative AI: Help a manager organise their own notes into themes before they write a review.
WHAT HAPPENED
A manager pasted a year of one-to-one notes and asked for themes. The output included "time-keeping concerns", based on a single note about one late meeting, and the manager nearly carried it into the review.
WHAT THEY CHANGED
The prompt now asks for each theme to list the notes it came from, any theme based on one note is flagged, and the manager writes the review themselves.
RED FLAGS
reviews in identical wording across a team; managers pasting AI summaries into reviews unedited.
WHAT TO TELL YOUR PEOPLE
“Managers may use AI to organise their own notes. Your review is written and decided by your manager.”
Before you switch it on
Will any AI output appear in someone's review, and how will it be checked?
Can the person being reviewed see what was used?
Who signs off that it's being used as agreed?
Has a data protection impact assessment been done?
Where AI can help: Generative AI: Explain reward statements and benefits in plain English.; Generative AI: Draft communications about pay reviews and benefit changes.; Generative AI: Answer common questions about how pay and benefits work, from published policies.; Generative AI: Summarise pay market or survey reports for the reward team.; AI agent: Find the relevant pay or benefits policy and answer an employee's question from it, with a link.; AI agent: Prepare and check data for a pay review, flagging gaps and anomalies for a person to look at.; AI agent: Answer payroll questions, such as payment dates, from the employee's own records.
GenAI: Lower risk; AI agent: Lower risk; GenAI: Lower risk; GenAI: Take care; AI agent: Take care; Both: Take care; AI agent: Take care; AI: Higher risk; AI: Higher risk
Value: Medium
Watch for: errors in figures; unequal outcomes; decisions nobody can explain.
Start here: explaining a reward statement in plain English.
Keep a person in charge of: setting or recommending anyone's pay.
Owner: reward lead, with finance.
View as: Manager
I need to explain our pay review process to my team. Here's the policy and timetable: [paste]. Draft a short, clear message: what happens when, how decisions are made, and who to ask. Don't include anyone's pay.
View as: Employee
Explain how these employee benefits work, in plain English, and what I'd need to do to use each one: [paste the benefits summary, with no personal details].
In practice
Illustrative example · fictional Leyfield Group
THE USE
Generative AI: Explain reward statements and benefits in plain English.
WHAT HAPPENED
Plain-English explanations cut the number of reward queries, until one rounded a bonus figure up by £50 and the employee queried the difference.
WHAT THEY CHANGED
Explanations now quote figures exactly as they appear on the statement and never calculate anything, and finance checks a sample every pay cycle.
RED FLAGS
anyone saying "the model recommended it" about pay; figures that can't be traced.
WHAT TO TELL YOUR PEOPLE
“Your reward statement explanation was drafted with AI from your statement. If anything doesn't match, your statement is correct; please tell us.”
Before you switch it on
Will it ever produce or change a figure, or only explain existing ones?
How will we check for unequal outcomes between groups?
Can every pay decision be explained without the AI?
Who checks figures before anything reaches an employee?
Has a data protection impact assessment been done?
Where AI can help: Generative AI: Group anonymous survey comments into themes and summarise what people are saying.; Generative AI: Draft action plans and "you said, we did" updates after a survey.; Generative AI: Write clear wellbeing and benefits information, and signpost where to get support.; Generative AI: Draft recognition messages and award nominations.; AI agent: Run pulse surveys and send reminders.; AI agent: Analyse sentiment or monitor messages or activity: the highest-risk use in HR, and never for individuals.
Watch for: monitoring people's messages, activity or mood; "anonymous" comments that identify people; wellbeing data used for other decisions.
Start here: grouping anonymous survey comments into themes.
Keep a person in charge of: any judgement about a named individual's wellbeing or behaviour.
Owner: HR director, with data protection and employee representatives.
View as: Manager
Here are the survey themes for my team (no individual comments): [paste]. Help me plan a 30-minute team conversation: how to open it, two questions for each theme, and how to agree two actions together.
View as: Employee
I want to nominate a colleague for recognition. Here's what they did: [paste, no names]. Turn it into a short, specific nomination: what they did, the difference it made, and one example.
In practice
Illustrative example · fictional Leyfield Group
THE USE
Generative AI: Group anonymous survey comments into themes and summarise what people are saying.
WHAT HAPPENED
A theme summary for a small team included a direct quote that made it obvious who had written it.
WHAT THEY CHANGED
Identifying details are removed before analysis, quotes are never included, and only themes with five or more comments are reported, so small teams are combined with others.
RED FLAGS
sentiment scores for individuals; monitoring nobody told employees about.
WHAT TO TELL YOUR PEOPLE
“We use AI to group anonymous survey comments into themes. We never use it to identify individuals or monitor what you write.”
Before you switch it on
Does this involve monitoring people, and have they been told clearly?
Could anyone be identified from the results?
Will this data ever be used for decisions about individuals?
What's the minimum group size before results are shown?
Has a data protection impact assessment been done, and have employee representatives been consulted?
Where AI can help: Generative AI: Group exit interview and leaver survey feedback into themes.; Generative AI: Draft leaver communications, handover templates and reference wording for HR to check.; Generative AI: Summarise why people say they leave, to inform planning and recruitment.; AI agent: Pull together exit feedback and leaver data from different systems into one report for HR.; AI agent: Run offboarding checklists: equipment return, final pay steps and removing system access, with a person confirming.; AI agent: Keep alumni in touch with news and roles.
GenAI: Lower risk; GenAI: Lower risk; AI agent: Lower risk; GenAI: Take care; AI agent: Take care; GenAI: Take care; AI: Higher risk; AI: Higher risk
Value: Medium
Watch for: comments that identify people; "likely to leave" predictions treated as fact; access left open.
Start here: themes from exit feedback.
Keep a person in charge of: any judgement about a named individual, including "likely to leave" flags.
Owner: HR operations and people analytics.
View as: Manager
A team member who does [role summary, no names] is leaving in four weeks. Draft a handover plan: what to document, who takes over each responsibility, and a week-by-week timeline.
View as: Employee
I'm leaving my role as [role] in four weeks. Help me write a clear handover document: my regular tasks, where things are kept, key contacts by role, and anything in progress.
In practice
Illustrative example · fictional Leyfield Group
THE USE
AI agent: Run offboarding checklists: equipment return, final pay steps and removing system access, with a person confirming.
WHAT HAPPENED
The offboarding agent removed a leaver's system access on their last day, but missed a shared mailbox set up outside the usual process, which stayed open for three weeks.
WHAT THEY CHANGED
The checklist now includes "anything set up outside the standard process", and a person confirms access is fully removed before the case is closed.
RED FLAGS
attrition scores for named people; access removal nobody double-checks.
WHAT TO TELL YOUR PEOPLE
“We use AI to group exit feedback into themes, so we can learn from it. Your comments aren't used to judge you or your team.”
Before you switch it on
What does the agent check, and who confirms access is fully removed?
Will any prediction ever be linked to a named person?
What happens to the leaver's personal data, and when is it deleted?
How long is exit feedback kept, and who can see it?
How will leavers be told how their feedback is used?
One accountable owner for AI in HR overall, plus a named owner for each use.
An agreed list of what AI is used for, kept up to date (Map your organisation is a good start).
Rules on data: what can go into which tool, approved by data protection.
Telling people: employees and candidates know where AI is used and how to ask for a person.
Challenge and correction: a clear route for anyone to question an AI-supported outcome.
Employee relations stays with people: in grievances, disciplinary matters and investigations, AI should only ever assist a named person, never draft findings or decide.
Review: each use reviewed at least yearly, and whenever the tool changes.
Know who owns AI in HR, which tools are approved for your team, and where to raise a concern.
You should be able to find out where AI is used in decisions about you, how to ask for a person, and how to challenge an outcome.
✦ GENAI drafts and summarises; a person decides. ⚙ AI AGENT works through tasks within limits you set. Any AI: applies whichever kind you use.
WHERE AI CAN HELP
LOWER RISK
GENAIDraft and tidy role profiles and job descriptions
GENAIPull skills themes out of job descriptions
TAKE CARE
GENAISummarise planning reports into a one-page brief (the figures must be checked) ★ Start here
GENAI"What if" scenario narratives from your own figures
AI AGENTGather headcount, vacancy and budget figures into a planning pack
AI AGENTAnswer leaders' plain-English questions about live HR data (who can see what must be controlled)
HIGHER RISK
ANY AIIdentifying roles or people at risk in a restructure, or building selection pools
ANY AI"Likely to leave" scores for named people
WATCH FOR
scenarios that quietly become decisions about real jobs; figures nobody has checked.
START HERE
summarising the data a planning meeting already uses.
KEEP A PERSON IN CHARGE OF
any decision about roles, structures or headcount.
OWNER
HR director or workforce planning lead, with finance.
TRY THIS FIRST
summarising the data a planning meeting already uses.
Here is our headcount and turnover report for the last quarter, with names removed: [paste]. Summarise it in one page for a planning meeting: the three biggest changes, which teams they affect, and two questions we should discuss. Don't make up any figures; if something isn't in the report, say so.
Before you paste: Remove names and personal details, and use an AI tool your organisation has approved.
Choose a high-value, lower-risk place to start, then review your first moves.
View as
VALUE AGAINST RISK
Where to start
Five lower-risk uses that add the most value, one from each stage. Every stage also has riskier uses, so check The risk mix below before you go further.
Draft clearer, more inclusive job adverts and check wording for bias
How many uses in each stage sit at each risk level. Select a stage to see them all.
MAKE IT PRACTICAL
Your best first moves
General first moves for HR teams. Answer the map in Check what we use to make them yours.
Start with
Start with HR service and support and Onboarding, then job adverts in Recruitment.
Drafts an adviser checks before sending.
Answering common new-starter questions from approved policies.
Clearer, more inclusive job adverts.
Fix first
✓ Agree who owns AI in HR.
✓ List where it's already used.
Keep an eye on
General examples, not a finding about your organisation: these stages contain the most higher-risk uses in this guide.
The Explore “Start here” badge describes a suggested use, which may still need care; Plan prioritises lower-risk options. Value is a directional practitioner judgement, not a business case or effort estimate. Other uses in the same stage can carry more risk. How I rated the risk →
Map where your organisation uses AI and see your own first moves.
YOUR ORGANISATION · PRIVATE TO THIS BROWSER
Map your organisation
HOW TO MAP YOUR ORGANISATION
Record AI usage and ownership answers, not a risk assessment of your actual uses.
1
Select GenAI, AI agents or both for each stage, or select No or Not sure. Then answer ownership separately where asked. Your answers stay in this browser.
2
Usage answered and stage fully answered are counted separately. Missing owners and uncertain or unanswered ownership get follow-up prompts.
3
When you have answered, download your map as a PDF to share.
Record AI usage for each stage, then actively answer the ownership question unless you selected No AI. Select both GenAI and AI agents where both are used; select a type again to remove it. No and Not sure replace the type selections. Your answers stay in your browser and appear in the PDF. This is an ownership and usage overview, not a risk assessment.
0 of 9 stages have AI usage answered · 0 of 9 stages fully answered
A stage is fully answered when usage and any required ownership answer have been selected. Not sure is an explicit answer, not confirmation that an owner exists.
1 Workforce planning
Usage unanswered
2 Recruitment
Usage unanswered
3 Onboarding
Usage unanswered
4 HR service and support
Usage unanswered
5 Learning and careers
Usage unanswered
6 Performance
Usage unanswered
7 Pay and reward
Usage unanswered
8 Engagement and wellbeing
Usage unanswered
9 Leaving and alumni
Usage unanswered
METHOD
How I rated the risk
Show the rating method
These ratings are my own view, as someone who works with HR technology and AI. They're not legal advice, and they're not an assessment of any organisation or product. Here's how I worked them out, so you can judge them for yourself.
I rate each use, not each stage
The same stage can hold some of the safest uses of AI and some of the riskiest, so every use is rated on its own.
Four questions decide the rating
1Does it shape a decision about a specific person, such as hiring, pay, promotion, discipline, ending employment or which work someone gets?
2Does it use personal data, especially health or other sensitive information?
3Does an AI agent act on its own, without a person approving each step?
4Does it evaluate or monitor people at work?
The four levels
Lower risk
None of these apply, or only published or combined information is used, and a person checks the result.
Take care
It uses personal data, deals with employees or candidates directly, or an agent acts within narrow limits, but it doesn't decide anything about a person.
Higher risk
It shapes a decision about a person, or evaluates or monitors people.
Avoid
This guide separates practices prohibited in a stated legal context from other uses it does not recommend. See the two Avoid groups in the map; a practitioner rating is not a legal finding.
Value
Value is a directional practitioner judgement about where AI may improve time or experience. It is not a business case or an estimate of effort, cost or implementation difficulty.
Where it comes from
The questions reflect the uses the EU AI Act treats as high-risk in employment, the UK GDPR rules on decisions made solely by automated means, the ICO's guidance on AI in recruitment and on monitoring workers, and what I've seen work, and go wrong, in practice.
Your context matters
The same use can be riskier in your organisation, depending on your data, your tools and your people. Treat these ratings as a starting point for your own assessment, not the final word.
Last reviewed October 2026. See it differently? Tell me through the feedback form below.
Risk and value ratings are my own practitioner view, worked out using the method on this page. They're not legal advice or an assessment of any organisation or product. Last reviewed October 2026.
LEILA FIELDING · AI FIELD NOTES12 November 2026
AI HR Lifecycle Map
Where AI can help across the employee lifecycle, and what to weigh at each stage.
1. Workforce planning
2. Recruitment
3. Onboarding
4. HR service and support
5. Learning and careers
6. Performance
7. Pay and reward
8. Engagement and wellbeing
9. Leaving and alumni
9 stages of HR
Where to start
Value against risk · All
Value
Lower risk
Take care
Higher risk
High value
2 Recruitment3 Onboarding4 HR service and support
Medium value
5 Learning and careers
1 Workforce planning6 Performance7 Pay and reward8 Engagement and wellbeing9 Leaving and alumni
✓ Lower risk! Take care× Higher risk⊘ AvoidGenerative AI creates or summarises content.AI agents work through tasks for you: some find and bring together information, others take steps on their own, within limits you set.Each stage can include a range of risks. The grid places the starting use; open the page's Method section for the full approach.
● 4 lower · ● 3 take care · ● 3 higher · ● 2 avoid
Start here: Lower riskValue: High
Higher risk: Screening, filtering, ranking or scoring candidates
Higher risk: Targeting job adverts at who gets to see a vacancy
Higher risk: Background checks that produce a "risk" score
Avoid · legally prohibited in the stated context: Using biometric data to infer candidates' emotions in workplace interviews, where the EU AI Act prohibition applies
Avoid · not recommended in this guide: Rejecting anyone without a person deciding
Start here: Draft clearer, more inclusive job adverts and check wording for bias
Keep a person in charge: shortlisting and every decision about a candidate.
Owner: talent acquisition lead, with HR and data protection.
3. Onboarding
Covers: pre-boarding, contracts, right-to-work checks, induction, equipment and system access, probation.
● 4 lower · ● 2 take care · ● 1 avoid
Start here: Lower riskValue: High
Avoid · not recommended in this guide: Using a general AI tool for right-to-work or identity checks (use the approved Home Office process and providers)
Start here: Answer new starters' common questions from approved policies
Keep a person in charge: right-to-work checks, contracts and anything sensitive.
Owner: HR operations, with the hiring manager.
The nine stages
4. HR service and support
Covers: policy questions, HR help desk cases, employee self-service, personal data changes, absence and holiday requests, employee relations cases (grievances, disciplinary, investigations).
● 4 lower · ● 4 take care · ● 1 higher · ● 2 avoid
Start here: Lower riskValue: High
Higher risk: Anything that recommends an outcome for an individual's case
Avoid · not recommended in this guide: AI drafting findings, outcomes or decisions in grievances, disciplinary matters or investigations
Avoid · not recommended in this guide: A bot handling health or other sensitive disclosures without a person
Start here: Reply drafts that an adviser checks before sending
Keep a person in charge: grievances, disciplinary matters and investigations, health, wellbeing and anything sensitive.
Owner: HR service delivery lead.
5. Learning and careers
Covers: training and learning paths, skills profiles, career paths, internal moves, promotions, succession.
● 3 lower · ● 4 take care · ● 2 higher
Start here: Lower riskValue: Medium
Higher risk: Selecting people for promotion, succession or talent pools
Higher risk: Ranking internal candidates for a role
Start here: Summaries, quizzes and practice material from training content
Keep a person in charge: promotion, succession and selection decisions.
Owner: learning and development lead, with talent management.
Higher risk: Recommending or setting an individual's pay or bonus
Higher risk: Dynamic pay setting
Start here: Explain a reward statement in plain English (quote the figures exactly)
Keep a person in charge: setting or recommending anyone's pay.
Owner: reward lead, with finance.
8. Engagement and wellbeing
Covers: surveys and listening, recognition, wellbeing support, health and absence support.
● 4 lower · ● 1 take care · ● 2 higher · ● 1 avoid
Start here: Take careValue: Medium
Higher risk: Sentiment analysis of named people's messages or activity
Higher risk: Monitoring employees' communications or activity
Avoid · legally prohibited in the stated context: Inferring employees' emotions from biometric data at work in the EU (with narrow medical and safety exceptions)
Start here: Themes from anonymous survey comments (groups of five or more only, with no quotes that could identify anyone)
Keep a person in charge: any judgement about a named individual's wellbeing or behaviour.
Owner: HR director, with data protection and employee representatives.
Higher risk: "Likely to leave" predictions about named people
Higher risk: Scoring people for redundancy selection (this sits with Workforce planning, but it's often triggered here)
Start here: Themes from exit feedback
Keep a person in charge: any judgement about a named individual, including "likely to leave" flags.
Owner: HR operations and people analytics.
How I rated the risk
These ratings are my own view, as someone who works with HR technology and AI. They're not legal advice, and they're not an assessment of any organisation or product. Here's how I worked them out, so you can judge them for yourself.
The same stage can hold some of the safest uses of AI and some of the riskiest, so every use is rated on its own.
Four questions I ask
Does it shape a decision about a specific person, such as hiring, pay, promotion, discipline, ending employment or which work someone gets?
Does it use personal data, especially health or other sensitive information?
Does an AI agent act on its own, without a person approving each step?
Does it evaluate or monitor people at work?
Lower risk
None of these apply, or only published or combined information is used, and a person checks the result.
Take care
It uses personal data, deals with employees or candidates directly, or an agent acts within narrow limits, but it doesn't decide anything about a person.
Higher risk
It shapes a decision about a person, or evaluates or monitors people.
Avoid
This guide separates practices prohibited in a stated legal context from other uses it does not recommend. See the two Avoid groups in the map; a practitioner rating is not a legal finding.
Value. Value is a directional practitioner judgement about where AI may improve time or experience. It is not a business case or an estimate of effort, cost or implementation difficulty.
Sources. The questions reflect the uses the EU AI Act treats as high-risk in employment, the UK GDPR rules on decisions made solely by automated means, the ICO's guidance on AI in recruitment and on monitoring workers, and what I've seen work, and go wrong, in practice.
Your context. The same use can be riskier in your organisation, depending on your data, your tools and your people. Treat these ratings as a starting point for your own assessment, not the final word.
October 2026. See it differently? Tell me through the feedback form below.
Where AI can help: Summarise headcount, turnover and absence reports into a one-page brief for a planning meeting.; Draft and tidy role profiles and job descriptions so they're consistent across teams.; Write "what if" scenario narratives (for example growth, a new site or a hiring freeze) from figures you provide.; Pull skills themes out of job descriptions to show where skills gaps might be.; Gather headcount, vacancy and budget figures from HR and finance systems into a planning pack.; Flag roles that have been vacant longer than an agreed time, or teams close to budget limits.
Watch for: scenarios that quietly become decisions about real jobs; figures nobody has checked.
Start here: summarising the data a planning meeting already uses.
Keep a person in charge of: any decision about roles, structures or headcount.
Owner: HR director or workforce planning lead, with finance.
Illustrative example · fictional Leyfield Group
THE USE: Summarise headcount, turnover and absence reports into a one-page brief for a planning meeting.
WHAT HAPPENED: The planning team asked AI to summarise quarterly turnover by department for the leadership meeting. It merged two departments with similar names, Customer Service and Customer Success, so Customer Service's turnover looked about twice as high as it really was.
WHAT THEY CHANGED: Department names now go in as a fixed list alongside the data, and every figure in the brief is checked against the source report before it's shared.
RED FLAGS: AI scenarios shared without saying they're scenarios; nobody can say where a figure came from.
WHAT TO TELL YOUR PEOPLE: We use AI to help summarise workforce data for planning. Decisions about roles are always made by people.
Where does the data come from, and who keeps it accurate?
Can we see how each figure was produced?
Who reviews outputs before they reach decision-makers?
Has a data protection impact assessment been done if personal data is used?
How will scenarios be labelled so nobody mistakes them for decisions?
Where AI can help: Write clearer, more inclusive job adverts, and check existing ones for jargon or biased wording.; Draft structured interview questions and scoring guides linked to the role's requirements.; Summarise an application for the recruiter, who still reads it in full.; Draft candidate emails: invitations, updates and kind, clear rejections.; Find and pull together a candidate's application, interview notes and scores from the applicant system for the hiring manager.; Schedule interviews across diaries and send confirmations and reminders.; Answer candidates' common questions about the process and the role.; Move applications between stages once a recruiter has made the decision.
Watch for: bias in screening; AI ranking or rejecting people; candidates not knowing AI is involved.
Start here: clearer, more inclusive job adverts.
Keep a person in charge of: shortlisting and every decision about a candidate.
Owner: talent acquisition lead, with HR and data protection.
Illustrative example · fictional Leyfield Group
THE USE: Summarise an application for the recruiter, who still reads it in full.
WHAT HAPPENED: Summaries of applications for a project manager role kept leaving out career breaks and returner programmes, so strong returners looked less experienced than they were.
WHAT THEY CHANGED: The prompt now asks the summary to list every gap and what the candidate said about it. Recruiters read the full application for anyone with a gap, and summaries are never used to sort or reject anyone.
RED FLAGS: recruiters only reading AI summaries; "the system rejected them" being said out loud.
WHAT TO TELL YOUR PEOPLE: We use AI to help draft our job adverts and schedule interviews. Every application is reviewed by a person.
Can we see why it summarised or ranked a candidate the way it did?
Has it been tested for bias with our own roles and applicants?
Can it reject or progress anyone without a person approving?
How will candidates be told, and how can they ask for a person?
Has a data protection impact assessment been completed?
Where AI can help: Draft welcome guides and first-week plans from your current policies.; Answer new starters' common questions using only approved documents.; Turn long policies into short, plain-English summaries for induction.; Find the right policy, form or contact for a new starter's question and point them to it.; Send reminders before day one and chase missing forms or documents.; Set up onboarding tasks across HR, IT and facilities, such as equipment, system access and induction sessions.
Watch for: out-of-date policy answers; personal data in the wrong tool; agents acting on incomplete records.
Start here: answering common new-starter questions from approved policies.
Keep a person in charge of: right-to-work checks, contracts and anything sensitive.
Owner: HR operations, with the hiring manager.
Illustrative example · fictional Leyfield Group
THE USE: Draft welcome guides and first-week plans from your current policies.
WHAT HAPPENED: A welcome guide for new warehouse starters quoted last year's holiday allowance, because an old policy was still sitting in the shared folder used as the source.
WHAT THEY CHANGED: Guides are now drafted only from the current policy library, each guide shows a "policies checked on" date, and HR operations signs off every new version.
RED FLAGS: new starters getting different answers from the guide and their manager.
WHAT TO TELL YOUR PEOPLE: Some answers in your welcome guide are drafted with AI from our current policies. If anything looks wrong, ask your HR contact.
Which documents will it answer from, and who keeps them current?
What personal data will it see, and is the tool approved for it?
If an agent chases documents, what happens when someone doesn't reply?
Who signs off guides before new starters see them?
How will new starters know which answers come from AI?
Where AI can help: Answer routine policy questions (holiday, expenses, hybrid working) from your published policies.; Draft replies to HR enquiries for an adviser to check and send.; Summarise a long enquiry history so an adviser can pick it up quickly (never for employee relations cases).; Keep FAQs and help articles up to date when a policy changes.; Find the relevant policy or document for a question and answer from it, with a link to the source.; Look up an employee's own information when they ask, such as their leave balance or payslip dates.; Route enquiries to the right team and log routine requests and personal data changes.
Watch for: confident wrong answers; one person seeing another's data; sensitive cases handled by a bot; AI summarising or drafting anything in an employee relations case; a chatbot answering questions about an open grievance.
Start here: drafts an adviser checks before sending.
Keep a person in charge of: grievances, disciplinary matters and investigations, health, wellbeing and anything sensitive.
Owner: HR service delivery lead.
Illustrative example · fictional Leyfield Group
THE USE: Draft replies to HR enquiries for an adviser to check and send.
WHAT HAPPENED: Advisers got through routine questions noticeably faster. Then one draft about hybrid working described a manager's informal practice as if it were the policy, and it nearly went out.
WHAT THEY CHANGED: The prompt now says "use only the published policy wording and quote the section", and advisers check the quote against the policy before sending.
RED FLAGS: advisers sending drafts unread; employees asking sensitive questions to a bot because it's quicker.
WHAT TO TELL YOUR PEOPLE: We use AI to help draft replies to your HR questions. An adviser checks every reply before it's sent.
How does it know when a question is sensitive and needs a person?
Can it only ever see the asking employee's own information?
What's the fallback when it doesn't know?
How will employees know they're using an AI assistant, and how to reach a person?
Where AI can help: Turn existing training material into summaries, quizzes and practice scenarios.; Draft a development plan from notes of a conversation between a manager and an employee.; Suggest learning that fits someone's goal or a skills gap.; Describe career paths, and what each next role involves, in plain English.; Search the learning catalogue for courses that match a goal or skills gap.; Build learning plans, enrol people on courses and send reminders.; Suggest internal roles or projects that match someone's skills profile.
Watch for: learning that looks complete but isn't; skills data reused for selection; the same people always being surfaced for opportunities.
Start here: practice material and plain-English summaries.
Keep a person in charge of: promotion, succession and selection decisions.
Owner: learning and development lead, with talent management.
Illustrative example · fictional Leyfield Group
THE USE: Suggest internal roles or projects that match someone's skills profile.
WHAT HAPPENED: Matching kept suggesting the same small group of people, because fully completed skills profiles scored higher. Part-time and frontline staff, who had less time to fill in profiles, rarely appeared.
WHAT THEY CHANGED: Every internal role is now also advertised openly, matches are shown to employees as suggestions they can choose to follow up, and the team checks each quarter who is being suggested.
RED FLAGS: the system didn't suggest you" used as a reason; skills profiles becoming a test nobody agreed to.
WHAT TO TELL YOUR PEOPLE: We use AI to suggest learning and roles you might like. It's a suggestion, not a selection, and every opportunity is open to everyone.
What data does it use to suggest people, and could that favour some groups?
Will skills data be used for anything beyond suggestions?
Can employees see and correct their own skills profile?
How will internal roles still reach people the system doesn't suggest?
Where AI can help: Help a manager organise their own notes into themes before they write a review.; Suggest clearer, measurable wording for goals the manager and employee have already agreed.; Draft agendas and questions for check-ins.; Help a manager prepare for a difficult feedback conversation.; Pull together the goals and check-in notes already recorded in the system, ready for a manager's review conversation.; Send check-in and review reminders, and gather the feedback requests a manager asks for.
Watch for: biased or invented summaries; reviews that sound fair but aren't evidenced.
Start here: helping a manager organise their own notes before writing.
Keep a person in charge of: writing, scoring and deciding any review or rating.
Owner: line managers, with HR business partners.
Illustrative example · fictional Leyfield Group
THE USE: Help a manager organise their own notes into themes before they write a review.
WHAT HAPPENED: A manager pasted a year of one-to-one notes and asked for themes. The output included "time-keeping concerns", based on a single note about one late meeting, and the manager nearly carried it into the review.
WHAT THEY CHANGED: The prompt now asks for each theme to list the notes it came from, any theme based on one note is flagged, and the manager writes the review themselves.
RED FLAGS: reviews in identical wording across a team; managers pasting AI summaries into reviews unedited.
WHAT TO TELL YOUR PEOPLE: Managers may use AI to organise their own notes. Your review is written and decided by your manager.
Will any AI output appear in someone's review, and how will it be checked?
Can the person being reviewed see what was used?
Who signs off that it's being used as agreed?
Has a data protection impact assessment been done?
Where AI can help: Explain reward statements and benefits in plain English.; Draft communications about pay reviews and benefit changes.; Answer common questions about how pay and benefits work, from published policies.; Summarise pay market or survey reports for the reward team.; Find the relevant pay or benefits policy and answer an employee's question from it, with a link.; Prepare and check data for a pay review, flagging gaps and anomalies for a person to look at.; Answer payroll questions, such as payment dates, from the employee's own records.
Watch for: errors in figures; unequal outcomes; decisions nobody can explain.
Start here: explaining a reward statement in plain English.
Keep a person in charge of: setting or recommending anyone's pay.
Owner: reward lead, with finance.
Illustrative example · fictional Leyfield Group
THE USE: Explain reward statements and benefits in plain English.
WHAT HAPPENED: Plain-English explanations cut the number of reward queries, until one rounded a bonus figure up by £50 and the employee queried the difference.
WHAT THEY CHANGED: Explanations now quote figures exactly as they appear on the statement and never calculate anything, and finance checks a sample every pay cycle.
RED FLAGS: anyone saying "the model recommended it" about pay; figures that can't be traced.
WHAT TO TELL YOUR PEOPLE: Your reward statement explanation was drafted with AI from your statement. If anything doesn't match, your statement is correct; please tell us.
Will it ever produce or change a figure, or only explain existing ones?
How will we check for unequal outcomes between groups?
Can every pay decision be explained without the AI?
Who checks figures before anything reaches an employee?
Has a data protection impact assessment been done?
Where AI can help: Group anonymous survey comments into themes and summarise what people are saying.; Draft action plans and "you said, we did" updates after a survey.; Write clear wellbeing and benefits information, and signpost where to get support.; Draft recognition messages and award nominations.; Run pulse surveys and send reminders.; Analyse sentiment or monitor messages or activity: the highest-risk use in HR, and never for individuals.
Watch for: monitoring people's messages, activity or mood; "anonymous" comments that identify people; wellbeing data used for other decisions.
Start here: grouping anonymous survey comments into themes.
Keep a person in charge of: any judgement about a named individual's wellbeing or behaviour.
Owner: HR director, with data protection and employee representatives.
Illustrative example · fictional Leyfield Group
THE USE: Group anonymous survey comments into themes and summarise what people are saying.
WHAT HAPPENED: A theme summary for a small team included a direct quote that made it obvious who had written it.
WHAT THEY CHANGED: Identifying details are removed before analysis, quotes are never included, and only themes with five or more comments are reported, so small teams are combined with others.
RED FLAGS: sentiment scores for individuals; monitoring nobody told employees about.
WHAT TO TELL YOUR PEOPLE: We use AI to group anonymous survey comments into themes. We never use it to identify individuals or monitor what you write.
Does this involve monitoring people, and have they been told clearly?
Could anyone be identified from the results?
Will this data ever be used for decisions about individuals?
What's the minimum group size before results are shown?
Has a data protection impact assessment been done, and have employee representatives been consulted?
Where AI can help: Group exit interview and leaver survey feedback into themes.; Draft leaver communications, handover templates and reference wording for HR to check.; Summarise why people say they leave, to inform planning and recruitment.; Pull together exit feedback and leaver data from different systems into one report for HR.; Run offboarding checklists: equipment return, final pay steps and removing system access, with a person confirming.; Keep alumni in touch with news and roles.
Watch for: comments that identify people; "likely to leave" predictions treated as fact; access left open.
Start here: themes from exit feedback.
Keep a person in charge of: any judgement about a named individual, including "likely to leave" flags.
Owner: HR operations and people analytics.
Illustrative example · fictional Leyfield Group
THE USE: Run offboarding checklists: equipment return, final pay steps and removing system access, with a person confirming.
WHAT HAPPENED: The offboarding agent removed a leaver's system access on their last day, but missed a shared mailbox set up outside the usual process, which stayed open for three weeks.
WHAT THEY CHANGED: The checklist now includes "anything set up outside the standard process", and a person confirms access is fully removed before the case is closed.
RED FLAGS: attrition scores for named people; access removal nobody double-checks.
WHAT TO TELL YOUR PEOPLE: We use AI to group exit feedback into themes, so we can learn from it. Your comments aren't used to judge you or your team.
What does the agent check, and who confirms access is fully removed?
Will any prediction ever be linked to a named person?
What happens to the leaver's personal data, and when is it deleted?
How long is exit feedback kept, and who can see it?
How will leavers be told how their feedback is used?
One accountable owner for AI in HR overall, plus a named owner for each use.
An agreed list of what AI is used for, kept up to date (Map your organisation is a good start).
Rules on data: what can go into which tool, approved by data protection.
Telling people: employees and candidates know where AI is used and how to ask for a person.
Challenge and correction: a clear route for anyone to question an AI-supported outcome.
Employee relations stays with people: in grievances, disciplinary matters and investigations, AI should only ever assist a named person, never draft findings or decide.
Review: each use reviewed at least yearly, and whenever the tool changes.
Leila Fielding
