Strategic workforce planning with AI: a guide for HR

By Johannes Sundlo, AI and future of work advisor, formerly HR at Spotify. Updated 2 November 2026.

The short answer

Strategic workforce planning makes sure an organization has the people and skills it needs to reach its goals, typically two to five years ahead, instead of solving one vacancy at a time. AI mostly changes tasks inside jobs and rarely whole jobs, so the analysis has to go down to task level. I suggest five steps. Start from the business plan, break the priority roles into tasks, calculate the gap in people and skills, choose a route for every gap, and write a one-page plan you review every six months. The seven routes are build, redeploy, buy, borrow, retain, bot and reduce.

Start with The business plan, then the tasks in three to five priority roles.
Seven routes Build, redeploy, buy, borrow, retain, bot or reduce.
Review Every six months. The tools change faster than a three-year plan.

What is strategic workforce planning?

CIPD, the UK professional body for HR, defines workforce planning as “the process of balancing labour supply (skills) against the demand (numbers needed)”. The aim is “the right number of people with the right skills employed in the right place at the right time” (CIPD).

The word strategic changes the time frame and the starting point. CIPD separates operational planning, such as staffing a call center or a project, from longer-term planning for the talent you’ll need. A strategic plan starts from where the business is going, often two to five years out, and works backward. In my view, a plan that only lists training courses has skipped that step. Start with what work needs doing, and by whom.

What is skills-based workforce planning?

Classic workforce planning counts headcount per role. Skills-based workforce planning counts the skills you have against the skills you’ll need, across roles. My guide to skills-based organizations covers how to build a short skills vocabulary without a giant taxonomy. Skills are one input. The plan also decides who does the work and which tasks AI takes on.

How AI changes workforce planning

A classic plan counts jobs, such as how many customer service agents, analysts and engineers you’ll need in three years. AI fits badly into that count, because it usually changes parts of a job and rarely the whole job.

25%of global employment is in occupations potentially exposed to generative AI (ILO and NASK, 2025)

49%of occupations have seen Claude used for at least a quarter of their tasks, up from 36% in January 2025 data (Anthropic, 2026)

39%of workers’ existing skills are expected to change or become outdated by 2030 (WEF)

59%of workers will need training by 2030, according to the employers in the WEF survey

The ILO and Poland’s NASK research institute find that full automation of jobs remains limited, because many tasks still need a person even when they go faster (ILO). Anthropic’s 49 percent comes from usage of one AI tool, Claude, mapped to the US Department of Labor’s O*NET task lists and pooled across its reports (Anthropic, January 2026). Its March 2026 report found the figure had barely changed (Anthropic, March 2026).

Of the 59 in every 100 workers who need training, employers in the World Economic Forum’s survey say 29 could be upskilled in their current role and 19 could be upskilled and redeployed elsewhere in the organization (WEF, Future of Jobs Report 2025). Those two groups match the build and redeploy routes further down.

A task-level plan also has to name the skills people must keep even when AI does much of the work. Someone who reviews an AI draft needs to be able to make the judgment themselves, or they won’t notice when it’s wrong.

Strategic workforce planning in five steps

The model is my own proposal. It starts from the business, plans at task level and treats AI as one of the options for closing a gap.

Start from the business plan. Ask each business leader what changes in their area over the next two to five years, such as volume, products, markets or budget, and which three to five roles matter most for it. For each of those roles, write down today’s headcount, expected departures such as known retirements and a normal year’s attrition, and open vacancies.
Break the priority roles into tasks. List the ten most common tasks in each role, estimate the share of working time each one takes and tag it as “AI can do much of it”, “AI can help” or “a person does it”. Let the people in the role make the list with their manager, because a job description rarely shows where the time goes.
Calculate the gap in people and skills. Add up the time in tasks AI can do much of, and subtract the time it takes to check the output. Compare the result with the demand from step one to see whether AI reduces the hiring need or mainly frees time for other work. Then write five to ten skills per role, with AI literacy as a baseline for everyone, and count how many people have each skill today against how many need it in two years. My AI fluency framework helps set the level each role should reach.
Choose a route for every gap. Pick build, redeploy, buy, borrow, retain, bot or reduce for each gap. Most roles need more than one.
Write one page per role and review it every six months. The page says what happens, who owns it, by when and how you’ll know it works. AI tools change faster than a three-year plan, so redo the task analysis for the priority roles every six months and adjust the plan.

How to run the task session in step two

My suggestion is an hour per role, with three to five people who do the work and their manager.

10 minutes. Everyone writes down what they did last week, task by task.
20 minutes. Group the notes into eight to twelve tasks and estimate the share of time for each.
20 minutes. Tag each task, and agree what someone has to check when AI does it.
10 minutes. Pick two tasks to test with AI before the next review.

AI can draft a first task list before the session. Paste a job description without names or personal data into a tool your organization has approved, then let the people in the role correct the list.

Here is a job description for [role]. List the ten tasks that probably take the most working time and estimate the share of working time for each. Tag each task "AI can do much of it", "AI can help" or "a person does it", and give a short reason. Finally, list the knowledge the person needs to keep to be able to review what AI produces. Answer in a table.

A made-up example of the math in step three

Four finance assistants spend about a third of their time on reconciliations and the monthly report. Say AI can do two thirds of that work, and checking the output takes a quarter of the time saved. A third times two thirds times three quarters is a sixth of the team’s time, or about two thirds of a full-time role. If volume stays flat, the team needs about 3.3 people instead of four. Test the assumptions on real work for a month before they go into the plan.

Seven routes to close a gap

SHRM calls build, buy and borrow a popular framework in HR and describes bot, using AI and automation to deliver work, as a fourth B added in recent years (SHRM). Some versions also add bind, meaning keeping key people (BDO). When AI changes the work, I’d use seven routes. I call bind retain, and I add redeploy and reduce.

Route	When it fits	Watch out for
Build	The task stays but is done in a new way, and the person has the foundation	Train on real tasks and give people time to practice afterward. A course without practice time changes little. That’s the gap between Knowledge and Ability in ADKAR.
Redeploy	AI takes over much of one role while another role is short of people	Often needs longer training and a receiving manager who wants the person. Plan the move before the old role shrinks.
Buy	The skill is missing and you’ll need it for a long time	Write the hiring profile around the tasks the role will have in two years.
Borrow	The need is temporary or too narrow for a role of its own. Use contractors, consultants or partners.	Decide how the knowledge stays with you when they leave.
Retain	The skill is critical and the person who has it may retire or leave soon	Let experienced people document routines and judgment calls before they go. The same documents can feed an AI assistant for new hires.
Bot	The task is rule-based and frequent, and errors are easy to spot	Someone still owns the result and checks it. Count the review time in the business case.
Reduce	AI or lower volume shrinks the need in a role over time	Start by not refilling departures and by redeploying. If redundancies are possible, involve employee representatives early and check the notice rules further down.

Half of the employers in the WEF survey plan to move staff from declining to growing roles (WEF). I think redeployment is underrated. It takes longer than hiring, but you keep people who already know your customers, systems and culture.

One example from Sweden combines retain and bot. By 2024 the City of Lidingö had an AI mentor on every work phone in its home care service, where turnover is high and many staff are hourly or temps. It was built to help new hires and temps learn the routines, was kept strictly to the city’s own routine documents and cited its source in every answer (SKR, in Swedish). You can copy the pattern. Have experienced staff document routines before they leave, then let an approved assistant answer only from those documents.

Task analysis examples for five roles

The examples show what a task analysis can look like. They are not measurements. How much AI can do in your organization depends on the tools you’ve approved, the data they may use and how the work is organized. Skills to keep are what people must still do well to catch AI’s mistakes.

Role	AI can do much of	AI can help with	Skills to build	Skills to keep
HR generalist	First drafts of job ads, policies and answers to common questions	Summarizing free-text survey answers, preparing guidance for managers	Reviewing AI drafts, working with data and processes	Employment law and judgment in individual cases
Customer service agent	Draft replies to common questions, summaries of case history	Finding answers in the knowledge base	Handling escalations, checking AI answers against policy	Product knowledge and judgment in complaints
Financial analyst	Compiling reports and first explanations of variances	Forecasts and models in Excel	Asking the right questions of the data and explaining results to managers	Understanding the business behind the numbers
Marketing specialist	First drafts, proofreading and versions for different channels	Market monitoring and channel analysis	Editorial judgment and checking sources	Knowledge of the customers and the brand
Home care worker	Little. The work happens in the client’s home.	Answers about routines, especially for new hires and temps	Questioning answers that look wrong	Care and the relationship with the client

A strategic workforce plan example and template

Copy the table into a document and make one per role or area. The example column follows the made-up finance team from the math above.

Field	What to write	Example
Role or area	The role or group the plan covers	Finance assistants, four people
What changes in the business	Volume, products, markets or budget over two to five years	Volume stays flat, and managers want more analysis support
Departures and hiring need	Expected retirements and other departures within three to five years, and open vacancies	One retires in 2028, no open vacancies
Tasks that change	Tasks AI can do much of or help with, with the rough share of working time	Reconciliations and the monthly report, about a third of the time
Tasks that grow	Work that gets more time or matters more	Analysis and budget support for managers
Skills to build	Five to ten skills	Reviewing AI analyses, Copilot in Excel, explaining numbers to managers
Skills to keep	What has to work without AI	Accounting rules and spotting errors in source data
Gap today	How many have the skill now, and how many need it in two years	One of four can review an AI analysis and find the errors. The three who stay all need to by 2028.
Route	Build, redeploy, buy, borrow, retain, bot or reduce	Build for the three who stay. Retain by having the retiring colleague document the year-end routines. Borrow a controller for the year-end close.
Headcount decision	Replace, don’t replace or replace with a different profile, and when you’ll decide	Don’t replace the 2028 retirement one for one. Three people plus a borrowed controller at year-end cover the work. Make the final call in autumn 2027, after measuring the time saved.
Entry-level places	Trainee, graduate or intern places in the area, now and planned	Keep the one trainee place, and let the trainee do the monthly report by hand before using AI
Action, owner and date	What happens, who owns it and by when	Three two-hour sessions on their own tasks, owned by the finance manager, done by March
How you’ll know it works	One or two measures	Days to a finished monthly report, hours spent on analysis
Employee voice	How employees and their representatives take part, and when	Team meeting before any decision, consultation with the union or works council where there is one

I’d let the business leader own the plan, while HR runs the process and combines the plans into one picture. My AI transformation playbook has a four-part message for telling teams what has and hasn’t been decided.

Check the rules before you act on the plan

If a plan could lead to redundancies, involve employee representatives early and check the rules where you operate.

EU. Article 4 of the AI Act requires providers and deployers of AI systems to take measures that support their staff’s AI literacy, without requiring a specific level. AI used to make decisions affecting terms of work, promotion or termination, to allocate tasks based on behavior or personal traits, or to monitor and evaluate performance and behavior is high-risk under Annex III, with rules that apply from 2 December 2027 (EUR-Lex).
UK. Collective consultation rules apply when you propose 20 or more redundancies at one establishment within 90 days. The minimum consultation period before any dismissal is 30 days for 20 to 99 redundancies and 45 days for 100 or more, and you notify the Redundancy Payments Service on form HR1 before consultation starts (GOV.UK).
US. Under the WARN Act, employers with 100 or more employees, not counting part-time staff, must give 60 days’ written notice before a covered plant closing or mass layoff (29 U.S.C. 2101 and 2102).

Where will junior people learn the job?

Many entry-level roles are built from tasks AI is good at, such as compiling, drafting and preparing material. When those tasks go, so does the place where the next generation of specialists learns the work. Stanford’s Digital Economy Lab used ADP payroll data covering millions of US workers through June 2026. Employment of 22- to 25-year-olds in AI-exposed occupations is 19 percent below where it would be had it kept pace with less exposed peers, mainly because fewer are hired. Experienced workers show no comparable gap. The authors call the results early indicators, not causal estimates (Stanford Digital Economy Lab).

My own analysis of Swedish job ads for graduate-level occupations points the same way. Ads with a junior profile fell 66.2 percent from 2022 to 2025 in the more AI-exposed occupations, against 36.0 percent in the comparison group. All ads in the sample fell 40.5 percent. Ads are not hires, and the report can’t rule out the economy, interest rates or the tech cycle as causes (Vad händer med de juniora jobben, in Swedish). My guide to the jobs AI will replace covers which roles are most exposed.

I think every workforce plan should answer where your future senior people will learn the job, and you don’t need a new program to begin.

Let junior staff do some tasks themselves first, even when AI could, and then compare their work with AI’s version.
Have them review AI drafts with a senior colleague, so they learn the judgment behind the result.
Count entry-level, graduate and internship positions in the plan, so you see if they’re shrinking.

Where AI can help with the planning itself

In the AI & HR Barometern 2026 from humAIn Academy, 22 percent of respondents say AI is used in people analytics and workforce planning, against 67 percent in HR administration and documents. Only one of the nine named HR areas scores lower. The respondents are 329 HR professionals in Sweden who chose to answer between 8 June and 24 September 2026, probably more interested in AI than HR overall (State of AI in HR). These are the three uses I’d start with.

A first task list. Use the prompt above, or start from O*NET, the US Department of Labor’s occupation database, which lists 26 tasks for human resources specialists (O*NET). ESCO, the European Commission’s free classification, describes 3,039 occupations and 13,939 skills in 28 languages (ESCO).
Scenarios. Compare a few futures in a spreadsheet, such as attrition staying flat or rising, or AI taking a quarter of a role’s time instead of half. My guide to AI for people analytics and workforce planning covers scenarios and the data cautions.
A second opinion on exposure. The ILO rates exposure by occupation, and Anthropic releases the data behind its Economic Index openly. Use them to challenge your own task tags.

What should you measure?

In the same barometer, 70 percent of the 329 respondents say their organization doesn’t measure the effect of AI in HR, and 5 percent have defined KPIs. Without measures, a workforce plan turns into a list of activities. Six measures are enough to begin with.

The share of priority roles where the task analysis has been done or updated in the last six months.
The gap per role, in people against the demand in the business plan and in skills against what the role needs in two years.
The share of open roles filled internally, especially in growing roles.
Time to full productivity for new hires and temps.
The share of employees who use AI in their regular tasks every week, and what the freed time goes to.
The number of entry-level, graduate or internship places per year in roles where AI takes over junior tasks.

Compare the measures with the hiring need in your own forecast. That tells you more than the number of training hours.

Common mistakes in workforce planning with AI

Planning in headcount only. A headcount forecast can’t show a role that keeps its title while half its tasks change.
Planning only for the roles that shrink. Growing roles need people trained and moved in time, and that takes longer than a hire.
Buying a skills platform before you know the questions. A rough list of skills for five roles is more useful than a complete taxonomy nobody keeps up to date.
Letting HR own the plan alone. If the business leader hasn’t signed it, the budget won’t follow.
Cutting the entry-level work that trains the next seniors. You’ll notice it in five years, when nobody is ready to step up.
Writing the plan without the people who do the work. They know where the time goes, and they’re the ones who have to change.

Frequently asked questions

What is strategic workforce planning?

Strategic workforce planning is the process of making sure an organization has the people and skills it needs to reach its goals, usually two to five years ahead. It compares the workforce you have with the one you’ll need, finds the gaps and decides how to close them.

What is the difference between workforce planning and strategic workforce planning?

Operational workforce planning covers the near term, such as staffing a call center or a project. Strategic workforce planning starts from the business strategy and looks two to five years ahead.

How is AI changing workforce planning?

AI mostly changes tasks inside jobs rather than whole jobs, so the analysis has to go down to task level. The plan also has to decide which tasks AI takes on, which skills people must keep to check AI’s work, and where junior employees will learn the job.

What should a strategic workforce plan include?

The business change behind it, the role or area, expected departures and hiring need, tasks that change and grow, skills to build and keep, today’s gap, the chosen route, the headcount decision, entry-level places, actions with owners and dates, measures, and how employees take part. The template on this page has a field for each.

What do build, buy, borrow and bot mean?

They are routes for closing a gap. Build means developing your own people, buy means hiring, borrow means using contractors or partners, and bot means using AI and automation to do the work. Some versions add bind, meaning keeping key people. I work with seven routes. I call bind retain, and I add redeploy, moving people from shrinking roles to growing ones, and reduce, letting a role shrink, starting with departures you don’t refill.

Who should own strategic workforce planning?

I’d let the business leader own the plan for each area, while HR runs the process, brings the data and combines the plans into one picture. If the business leader hasn’t signed the plan, the budget won’t follow.

How often should you update a strategic workforce plan?

I’d review it every six months, because AI tools change faster than a three-year plan. Anthropic’s share of occupations where Claude is used for at least a quarter of the tasks went from 36 to 49 percent between its January 2025 data and its January 2026 report. Redo the task analysis for the priority roles at each review.

Can you use AI to assess employees’ skills?

With care. In the EU, AI used to make decisions affecting terms of work, promotion or termination, to allocate tasks based on behavior or personal traits, or to monitor and evaluate performance and behavior is high-risk under Annex III of the AI Act, with rules that apply from 2 December 2027. Let AI draft a skills inventory, and let people check and confirm their own skills.

Sources

Want to start with your leadership team’s own tasks?

In a three-hour AI workshop, your team brings its real tasks, practices with AI and leaves with a 90-day plan with owners and milestones. For a longer change across the organization, I run AI change programs for leaders, managers and HR over 3 to 12 months, in English or Swedish. The 13 leadership teams I’ve measured rated their workshop 4.97 out of 5 on average.

Discuss a workshop