Source: [Turn AI exposure scores into a task-level workforce plan — Skills for Business](<https://skillsforbusiness.ai/blog/ai-exposure-scores-workforce-planning/>)

[← All articles](https://skillsforbusiness.ai/blog/)

HR & people · Guide

# Turn AI exposure scores into a task-level workforce plan

How HR and business leaders can use AI-exposure research: inspect actual tasks, check deployment readiness and plan skills around changing workload.

[Skills for Business](https://skillsforbusiness.ai/blog/editorial/)31 October 20267 min read

On this page

1.  [First establish what the exposure measure means](https://skillsforbusiness.ai/blog/ai-exposure-scores-workforce-planning/#first-establish-what-the-exposure-measure-means)
2.  [Observed use adds evidence without completing the forecast](https://skillsforbusiness.ai/blog/ai-exposure-scores-workforce-planning/#observed-use-adds-evidence-without-completing-the-forecast)
3.  [Build a task map with the people doing the work](https://skillsforbusiness.ai/blog/ai-exposure-scores-workforce-planning/#build-a-task-map-with-the-people-doing-the-work)
4.  [A hypothetical facilities-coordination role](https://skillsforbusiness.ai/blog/ai-exposure-scores-workforce-planning/#a-hypothetical-facilities-coordination-role)
5.  [Keep capability, readiness and demand separate](https://skillsforbusiness.ai/blog/ai-exposure-scores-workforce-planning/#keep-capability-readiness-and-demand-separate)
6.  [Derive learning needs from the redesigned work](https://skillsforbusiness.ai/blog/ai-exposure-scores-workforce-planning/#derive-learning-needs-from-the-redesigned-work)
7.  [Give the workforce scenario an evidence trail](https://skillsforbusiness.ai/blog/ai-exposure-scores-workforce-planning/#give-the-workforce-scenario-an-evidence-trail)
8.  [Sources](https://skillsforbusiness.ai/blog/ai-exposure-scores-workforce-planning/#sources)

What to take away

-   An occupational exposure score identifies work to investigate; it does not specify a staffing change.
-   Separate technical capability, local deployment readiness and demand for the service.
-   Plan learning around the work people will perform, including checking, coordination and exception handling.

Use an AI-exposure score to choose work worth investigating, then build the workforce plan from evidence about the actual service. Before drawing a staffing conclusion, establish which tasks can change, whether the organization can deploy the relevant system and what demand remains for the work.

A job title near the top of an exposure ranking is a reason to ask more precise questions. It does not establish that a particular team can remove the same proportion of its workload, or that fewer people will be needed. Our proposed method is to translate the ranking into a task map, a readiness review and a set of workload scenarios.

This guide concerns planning work and learning at team level. It is not advice about individual employment decisions, and it does not present an AI system’s assessment of a person as workforce evidence.

## First establish what the exposure measure means

The ILO’s April 2026 research brief, [Workers’ exposure to AI](https://www.ilo.org/publications/workers%E2%80%99-exposure-ai-what-indicators-tell-us-%E2%80%93-and-what-they-don%E2%80%99t), explains that exposure indicators vary in their definitions and methods. It highlights limits including static task descriptions, omitted adoption constraints and imperfect transfer of occupational information between countries. Such indicators describe technological susceptibility rather than predicting employment outcomes.

The [OECD’s May 2026 working paper](https://www.oecd.org/en/publications/the-oecd-ai-exposure-measure_f3da0f0a-en.html), revised in July, offers a useful example of why definitions matter. Its capability-gap approach compares occupational requirements with AI capabilities across cognitive, social and physical domains. A smaller gap implies greater potential exposure. This is broader than asking what a text-generating assistant can do.

Before using any ranking, record the measure’s date, technology scope, occupational classification and meaning of a high score. Check whether it concerns potential capability, observed use or another construct. Do not combine scores from different methods as though they were interchangeable percentages of labor that can be removed.

For a business operating across countries, also check whether a common job title conceals different responsibilities. A role classification can be useful for screening opportunities while remaining too broad for a local operating plan.

## Observed use adds evidence without completing the forecast

Anthropic’s March 2026 [labor-market research](https://www.anthropic.com/research/labor-market-impacts) combines theoretical task feasibility with its own usage data. Its observed-exposure measure gives greater weight to work-related and automated use. That is vendor-produced research using Claude activity, not an independent census of workplace adoption across products or companies.

The distinction is useful: something judged technically feasible may not yet appear in observed use. But evidence that people use a product for a task still does not establish the net effect on your team’s workload, service quality or staffing requirements.

Use public usage research to refine the questions for a local investigation. If a task appears suitable for AI assistance, ask what the implementation needs and what remains outside the observed interaction. A generated answer may be one step in a process that also includes gathering information, resolving an exception and obtaining acceptance from another team.

Avoid treating a lack of observed usage as proof that a task is impossible. Equally, avoid treating observed use as proof that the whole workflow is ready to change.

## Build a task map with the people doing the work

Choose a service rather than an entire department. “Facilities requests” is easier to inspect than “administration.” Ask employees and service owners to describe the work from intake to resolution, including the cases that do not follow the usual path.

For each activity, our proposed task map records the input, expected output, volume pattern, dependencies and responsibility for the result. Include coordination and checking. If the official role description is incomplete, record the actual work before assessing what might change.

Keep task frequency separate from task importance. A rare escalation may require little annual time while still determining who must remain available. Likewise, a task that looks short in isolation may involve waiting for information or coordinating several people.

Use representative work examples and aggregate process information. The purpose is to understand the service and its constraints, not to rank employees by how replaceable an AI system says they are.

## A hypothetical facilities-coordination role

Imagine a coordinator who receives maintenance requests, clarifies missing details, arranges access and follows up with suppliers. This is a fictional example, not an assessed occupation or a customer case.

An occupational ranking might suggest that administrative information processing deserves attention. The local task map could look like this:

| Activity | Candidate change to investigate | Evidence still needed |
| --- | --- | --- |
| Summarize incoming requests | Draft a structured intake record | Whether required details and uncertainty survive summarization |
| Classify routine requests | Suggest a queue or service category | Accuracy on local categories and unusual cases |
| Arrange access to a building | Assist with correspondence | Current permissions, schedules and coordination responsibilities |
| Resolve conflicting supplier updates | Assemble the evidence | Who can establish the authoritative status and make the decision |
| Close a completed request | Prepare a completion note | Confirmation that the service was delivered and accepted |

The table does not assign automation percentages. It identifies testable changes and the evidence that would justify them. Drafting an intake record and authorizing building access are different responsibilities, even if both appear under one job title.

Suppose summarization works well but employees spend longer clarifying incomplete requests. The planning opportunity may be better intake design, not more summarization. If routine cases become easier while exceptions remain, the role’s skill mix could shift toward diagnosis and coordination.

Those are hypotheses for the pilot. They should not be presented as observed outcomes before the team tests them.

## Keep capability, readiness and demand separate

Use three planning questions for each proposed change. Can the supported system perform the task to the required standard? Can this organization deploy it with the necessary information, access and ownership? What happens to the volume and complexity of work if the service becomes easier to use?

These questions lead to different actions:

| Local finding | Useful planning response |
| --- | --- |
| Capability looks promising, but source information is unreliable | Improve the information process before estimating capacity gains |
| The task works, but a downstream team cannot absorb more cases | Plan across the service boundary and examine the bottleneck |
| Routine handling improves while difficult cases remain | Examine coverage, escalation and learning needs for the changed case mix |
| Lower effort enables a previously unaffordable service | Model the additional demand and its support requirements |
| The system needs extensive correction | Revise or stop the proposed change before using it in a workforce scenario |

These are editorial recommendations, not thresholds derived from an exposure index. Keep the findings in separate fields rather than reducing them to one reassuring readiness score.

For staffing scenarios, make the demand assumption visible. Holding demand constant is one possible scenario, not an established consequence of introducing AI. Consider existing backlogs, service standards and unmet requests alongside the time spent on routine cases.

## Derive learning needs from the redesigned work

In their September 2026 [ILO practitioner article](https://www.ilo.org/resource/article/exposure-opportunity-why-skills-shape-employment-effects-new-technologies), economists Isaure Delaporte and Veronica Escudero emphasize skills that complement new technology, including communication, coordination and management. Their discussion draws on research about emerging digital technologies across regions; it is not a controlled evaluation of a specific generative-AI training program.

The implication we draw for employers is to identify the responsibilities people will retain or acquire before selecting courses. “High AI exposure” alone does not specify whether someone needs source checking, exception diagnosis, process design or technical configuration skills.

For the fictional coordinator, a useful learning exercise could involve spotting a missing access constraint, reconciling contradictory updates and explaining why a request cannot yet be closed. Knowing how to request a polished summary would cover only part of that work.

Our [guide to learning after AI assistance](https://skillsforbusiness.ai/blog/ai-workplace-training-measure-learning-after-assistance/) explains how to check what people can still do when assistance is unavailable. Apply that principle to the responsibilities the revised service will depend on.

## Give the workforce scenario an evidence trail

Keep the original exposure source alongside the local task map, pilot findings, demand assumptions and unresolved constraints. Label each conclusion by its basis: external research, an observed local result or a planning assumption.

Use the [completed-work evaluation method](https://skillsforbusiness.ai/blog/workplace-ai-adoption-measure-completed-work/) to capture review, rework and acceptance during the pilot. Then examine what changed across the service, rather than extrapolating from a faster drafting step.

Agree in advance when the plan will be revisited: after the pilot, a significant change in demand or a material system update. A workforce scenario should remain revisable as evidence changes.

The next useful step is one team-level review with employees, the service owner and the people responsible for learning and technology. Bring a bounded service and its actual work examples. Leave with a short list of changes to test, the responsibilities that remain and the evidence needed for a later capacity decision.

## Sources & further reading

Sources checked for this article. Research findings and vendor claims are distinguished in the text.

1.  [Workers’ exposure to AI: What indicators tell us – and what they don’t](https://www.ilo.org/publications/workers%E2%80%99-exposure-ai-what-indicators-tell-us-%E2%80%93-and-what-they-don%E2%80%99t)International Labour Organization · Accessed 31 October 2026
2.  [The OECD AI exposure measure: Mapping the OECD AI Capability Indicators to occupations](https://www.oecd.org/en/publications/the-oecd-ai-exposure-measure_f3da0f0a-en.html)OECD · Accessed 31 October 2026
3.  [Labor market impacts of AI: A new measure and early evidence](https://www.anthropic.com/research/labor-market-impacts)Anthropic Economics · Accessed 31 October 2026
4.  [From exposure to opportunity: Why skills shape the employment effects of new technologies](https://www.ilo.org/resource/article/exposure-opportunity-why-skills-shape-employment-effects-new-technologies)Isaure Delaporte and Veronica Escudero; International Labour Organization · Accessed 31 October 2026

Produced with AI-assisted research and writing. Examples and recommendations are our editorial synthesis unless attributed. This article does not claim independent product testing or human expert review. [Read our editorial policy.](https://skillsforbusiness.ai/blog/editorial/)

## Keep reading

[When an HR policy changes, update the AI answer path](https://skillsforbusiness.ai/blog/hr-ai-assistant-policy-change-workflow/)[AI workplace training: check what people can do after assistance](https://skillsforbusiness.ai/blog/ai-workplace-training-measure-learning-after-assistance/)