Restructuring an HR Function Without Losing Those Who Carried It

Context

A regional food retailer operating in Western Europe employs between one thousand and five thousand people depending on the season. Its store network rests on an HR function historically sized for personnel administration and payroll, on a largely manual model.

The retailer had launched a shift toward an integrated HR platform and the automation of part of its transactional processes. The consequence was mechanical. The HR function — around sixty strong — had to be cut by nearly half over eighteen months.

Senior management wanted to avoid two pitfalls: a brutal plan that would have degraded the social climate at an already tense moment of transformation, and a blind reduction that would have kept the wrong people.

Problem

The classic reflex would have been to keep the best-rated profiles from the annual reviews and the most senior people in the function. That reflex was the problem. The best-rated were often the strongest performers on the very tasks heading for automation. Their excellence rested precisely on what was about to disappear.

The payroll mass at stake was a significant budget issue, but it was not the heart of the risk. The heart of the risk was losing the rare profiles able to steer the HR function in its new configuration — profiles who did not necessarily sit at the top of the grids, because the grid was measuring mastery of the old world.

Management had less than a year to identify who to keep, with no real ability to trust the existing evaluation instruments.

Intervention

Management set aside the annual performance grid as the sole criterion. It built a reading overlay — a Trajectory Radar — from data that was already in the system but had never been cross-referenced.

Four signals were drawn on. The history of internal moves, to spot the people who had already pulled off role transitions. Usage traces from digital tools, to measure who ramped up fast. Documented peer-support activity, to identify the colleagues people turned to spontaneously. And end-of-project feedback, more revealing than the smoothed annual rating.

Cross-referencing these signals produced a map that looked nothing like the one the evaluations were drawing. Several payroll administrators, rated average on the classic grid, showed a tool-adoption speed well above average and intense support activity. Conversely, some highly rated profiles turned out to be rigidly attached to the manual processes they had mastered.

The Potential Stack was used to go deeper on the ambiguous profiles. Beyond the inventory of their current skills, management looked at their cognitive flexibility, their capacity to unlearn, and their ability to weave their knowledge into the knowledge of others.

Keeping the best-rated would have meant holding on to the best specialists of a job that was vanishing. The question was not who excels today, but who can learn tomorrow’s job fast enough to hold it.

Retention decisions never rested on the radar alone. Each profile was validated by triangulation across three viewpoints: the direct manager, peer appraisal, and usage data. The radar flagged; humans decided. That discipline also protected the system legally, avoiding any consequential decision based on a single automated indicator.

Outcome

The HR function was cut by 46% over the planned period. Retention of the profiles flagged as critical by the radar reached 91% at eighteen months — against the high voluntary attrition usually seen in this kind of restructuring.

Two unforeseen effects emerged. First, three payroll administrators repositioned into HR data-analysis roles absorbed their new scope faster than externally recruited profiles in equivalent roles. Second, the partial transparency of the system — explained to employee representatives without disclosing individual data — defused part of the distrust. The restructuring was perceived as less arbitrary than previous ones, because it rested neither on seniority alone nor on the annual rating alone.

Lessons

1. In an automation-driven restructuring, keeping the best-rated often means holding on to the best specialists of the very tasks that are disappearing. Past performance is measuring the wrong thing.
2. Trajectory signals — successful moves, adoption speed, peer support — predict the ability to hold the future job better than the annual rating does.
3. A decision to keep or release someone should never rest on a single automated indicator. Triangulation across manager, peers, and data protects both accuracy and legal defensibility.
4. Explaining the system’s logic to employee representatives, without disclosing individual data, reduces the perception of arbitrariness and eases social acceptance.