Keeping a hold on judgement when AI enters HR decisions

The HR team was not starting from scratch. Before the first day, five participants out of six said they had already used AI for a management task: preparing an interview, drafting a memo, analysing feedback. The words they chose to describe their stance spoke for themselves: curious, open, architect, opportunist. Nobody described themselves as resistant.

So the question was not whether to adopt AI. It was how far to let it decide, in the name of which values, and how the internal leadership model should take this into account.

What the participants put on the table:

• Unresolved trade-offs: is using a predictive score for absenteeism or disengagement still management, or already a breach of trust?
• Candidates taking online reasoning tests with an AI open beside them, and scores that no longer mean anything.
• An expert employee wearing themselves out rechecking every line of an automated chain, for lack of trust in the tool.
• Internal communication where the tone suggested by AI goes off track, and everything has to be rewritten without the original context.
• A majority of profiles with no direct reports, who nevertheless have to support managers across the whole organisation.

We designed a two-day programme, facilitated by a senior Lead coach-consultant. The principle: 30% short input, 70% exercises, and every exercise produces something concrete. The programme was linked, pillar by pillar, to the organisation’s internal leadership model: each competency expected of a manager was reread through the lens of the tension AI introduces into it.

Before: listen and recalibrate

• A scoping call with the HR manager to set the intention: to question the internal leadership model, not to learn tools.
• An individual twelve-question questionnaire: stance on AI, actual usage, a recent decision AI could have changed, an unresolved trade-off, a difficult conversation ahead.
• An anonymised summary shared back with the group, and a programme adjusted accordingly: a dilemma card on predictive HR scoring added to the game, a case study on candidates passing tests with AI, wording adapted for profiles with no direct reports.

Day 1: diagnose and demystify

Intention: start from what is really happening, not from fantasy.

• Shadow AI made visible: the gap between what the organisation believes it oversees and what teams actually do with consumer tools.
• A four-level governance grid, run in parallel: policy, tooling, training, field monitoring.
• A bingo of received ideas, checked live on screen, to bring down the beliefs that skew decisions.
• A Stop, Start, Continue sort on fifteen real tasks drawn from each person’s diary, then a collective wall: what this group refuses to delegate to AI.
• A hackathon on four cases, including an HR case from the questionnaire, with mandatory iterations and a vote on “most useful on Monday morning”.
• La Boussole (the Compass) × AI, to put values under pressure.

Day 2: decide and commit

Intention: turn clarity into a stance, then into a team rule.

• The shift from expert leader to orchestrator leader: setting direction, assembling, protecting human judgement.
• A seven-level delegation grid, from “AI executes” to “AI decides alone”, crossed with the type of task and the type of context to spot where judgement is atrophying.
• A one-sentence leadership signature, anchored in three values and one acknowledged tension, with a variant for profiles without direct management responsibility.
• Four employee profiles in the face of AI (the naive enthusiast, the rational sceptic, the silent worrier, the opportunist) and a simulated difficult conversation.
• A prototype AI delegation policy: twelve real tasks, current level, target level, a first guardrail signed and dated.
• A commitment card shared with two sparring partners, and milestones at 30, 60 and 90 days.

After: embed

• Slides, the participant workbook (manual and exercise pages, references included) and reminder sheets sent out.
• An “AI integrator” ideal profile read through the 7 Islands, to spot the territories that are over-inhabited and those that are avoided when integrating AI.
• A debrief with the HR manager.

The signature tool: La Boussole × AI

Lead’s La Boussole game, extended with a set of Dilemma cards specific to AI. Each card presents a concrete management situation (a report that is too brilliant, client data pasted into a free tool, a job made redundant by automation) and pits two of the seven values of responsible management against each other: efficiency, boldness, integrity, agility, solidarity, dedication, humility.

For this team, one card was written from a question raised in the questionnaire: a predictive tool flags two employees as “at risk of disengagement”. A preventive conversation based on the score, or wait for them to speak up?

The rule of the game fits in one sentence: “A value is not a goal or a slogan. It is a decision criterion, even when it costs.”

• A personal Stop, Start, Continue map of real tasks.
• A leadership signature in the face of AI, anchored in three La Boussole values and one acknowledged tension.
• A prototype AI delegation policy, to present to the team, with a first dated guardrail.
• A draft five-rule AI use charter, drawn from the game.
• A reading of the internal leadership model in the age of AI: one tension named per competency.
• A commitment card, two sparring partners and milestones at 30, 60 and 90 days.
• The participant workbook, the Dilemma cards, the reminder sheets and the ideal profile in 7 Islands.

The programme aims to move the HR team out of the role of tool relay and into that of guardian of human judgement in the organisation. In practice, it should make it possible:

• to speak the same language about what can and cannot be delegated to AI, task by task;
• to treat Shadow AI as a matter of psychological safety rather than as a fault to be sanctioned;
• to settle sensitive HR-specific cases (predictive scoring, recruitment, appraisal) on the basis of explicit criteria;
• to equip profiles with no direct reports to support managers across the whole organisation;
• to develop the internal leadership model by naming, competency by competency, what AI puts under tension.

A first effect is documented even before the training: the questionnaire responses changed the content of the two days (a new dilemma card, a new case study, an adapted signature formula). Participants worked on their own situations, not on generic cases.

• A programme calibrated on participants’ real situations, gathered before the training and fed back into the exercises.
• Lead tools that work on decision-making rather than theory: La Boussole and its Dilemma cards, the 7 Islands reading.
• Input grounded in research (psychological safety, the jagged frontier of AI, adaptive challenges), not in passing trends.
• A combination of coaching, facilitation and consulting, serving the organisation’s own leadership model, without imposing another one.

If you want your managers or your HR team to take a position before usage decides for them, we can build a programme with you, rooted in your own situations. Let’s talk.