Bokili
Responsible AI Practice4 min read

AI Training for Employees: Add the Affected-Person Check

AI training for employees should teach one check before an output is used: who is affected, what could change and how can they challenge it?

Bokili Editorial· Verified October 29, 2026
ShareX
A reviewer checks an AI-assisted document while colleagues, a customer, a candidate and a supplier can challenge its use.

AI training for employees should teach one question before an AI-assisted output is used: who could be affected if this is wrong, incomplete or applied without context? Employees often learn to protect inputs and check factual claims. They may still miss the colleague, customer, candidate or supplier who carries the consequence.

The affected-person check turns responsible AI from a broad principle into a visible work behaviour. Before sharing, approving or acting on an output, the employee names the people affected, the possible consequence, the evidence needed and the route to question or correct the result.

AI training for employees needs an affected-person check

The European Commission’s current AI-literacy guidance says organisations should consider the context in which an AI system is used and the people or groups on whom it is used. Its Q&A also explains that affected people may need enough understanding to see how an AI-assisted decision could influence them.

UK guidance for automated decision-making adds a practical standard: impacted people should have a simple way to request human intervention or challenge a decision, and explanations should suit the audience. Even when an employee is using a general-purpose AI tool for an ordinary work product, those questions improve judgement about when the output is ready.

Output-only checkAffected-person check
Primary questionIs the draft accurate?Who carries the consequence if it is used?
EvidenceSources support the textEvidence supports the intended use and affected group
ReviewSomeone proofreads the outputA responsible person can correct, pause or reject its use
Challenge routeNot specifiedAffected people know where to question or correct it

Use the FACE method before the output moves

FACE

1

Find the affected people

Name the people or groups whose access, workload, opportunity, money, safety or reputation could change.

2

Assess the consequence

Describe the plausible effect and whether it is easy to reverse. Do not replace this with a generic low, medium or high label.

3

Check evidence and context

Verify the important claim, missing perspective and intended use. Ask whether the source material represents the affected group well enough.

4

Enable challenge and correction

Name the human owner, the route to question the result and what happens while the issue is reviewed.

FACE does not turn every employee into a legal or ethics specialist. It gives people a short pause that exposes when qualified review is needed. The depth should match the task: a private brainstorming note is different from a hiring recommendation, customer eligibility decision or supplier assessment.

Worked example: an AI-assisted supplier summary

A procurement analyst asks AI to summarise supplier-renewal evidence. The draft says one small supplier has “repeated delivery problems” and recommends reducing its allocation. The sentence is based on two late deliveries, but the source pack omits a temporary warehouse closure recorded elsewhere.

Reading is a start. Practice makes it stick.

Start learning

Apply FACE before using the recommendation

  1. 1

    Find

    Name the supplier, procurement owner and internal teams whose plan would change.

  2. 2

    Assess

    Reducing allocation could affect revenue and future access to work; reversing the decision later may not repair the relationship.

  3. 3

    Check

    Open the delivery records, confirm the comparison period and add the missing closure context. Separate observed delays from a general reliability claim.

  4. 4

    Enable

    Route the corrected summary to the authorised buyer and preserve a contact path for the supplier to challenge factual errors.

  5. 5

    Decide

    Use the evidence to continue, revise or reject the recommendation. Do not let the model make the commercial decision.

The improved output may still identify a delivery risk. The difference is that the claim is bounded, the missing context is visible and the affected party has a route to correct the record. Responsible use becomes a work step, not a slogan.

A fluent output can still be incomplete in the way that matters most

Accuracy is not the only question. The intended use, consequence and challenge route decide whether the output is ready.

What employee practice should make observable

The UK Government AI Playbook recommends meaningful human control, clear responsibility and mechanisms for contestability and redress. Training should therefore test whether employees can identify the human decision owner, notice who may be affected, explain the evidence in plain language and stop the workflow when a challenge cannot be handled safely.

Score the affected-person check

  • The affected people or groups are named.
  • The consequence and reversibility are described.
  • Important evidence and missing context are checked.
  • A human owner can change or stop the outcome.
  • The explanation fits the affected audience.
  • A clear challenge or correction route exists.

Connect the check to a wider learning path

Pair FACE with the whole use-case chain, the sharing decision and the four review passes. Bokili’s training for HR and L&D turns these behaviours into short, role-based practice.

Run a ten-minute affected-person check
  1. Choose one fictional AI-assisted workplace recommendation.
  2. Name every person or group whose situation could change.
  3. Write the most important consequence and whether it is reversible.
  4. Check one claim and one missing perspective.
  5. Name the human owner and the route to challenge or correct the result.

Employees do not need to predict every possible impact. They need to notice when an output moves from private assistance into a decision that affects someone else. Teach that transition explicitly.

Make responsible use practicable

A short affected-person check helps employees connect AI literacy with real work. Bokili supports focused practice adapted to role, tool and context.

Sources

  1. AI talent, skills and literacy — European Commission
  2. AI Literacy — Questions & Answers — European Commission
  3. Ethics, Transparency and Accountability Framework for Automated Decision-Making — UK Government
  4. Artificial Intelligence Playbook for the UK Government — UK Government
ShareX

Reading is a start. Practice makes it stick.

Bokili turns skills like this into ten-minute missions for your whole team, with instant feedback and progress you can see.

Start learning

Keep reading

All articles