The convenient answer was not always the one I could defend.

The 2021 staffing discussion began with a register showing 164 open positions. Operating leaders saw a recruiting problem; finance saw a potential cost commitment. Both reactions assumed the entries described current requirements. My team found that the register combined active approvals, superseded requests and duplicate records created when a role moved between regional and central systems. The headline number was precise, but the decision it invited was unclear.

I asked managers to confirm the work behind each request before approving a campaign to fill the entire list. Some colleagues objected that reconciliation would distract recruiters during a difficult hiring period. I accepted that delay had a cost, but argued that recruiting against an obsolete requirement could create a larger problem. We spent USD 210,000 on temporary analytical support and manager review, while continuing recruitment for requirements already confirmed as current.

The final account identified 92 current approved requirements, 48 obsolete requests and 24 duplicates. Those categories reconcile to the original 164 entries. Each current requirement had an operating owner and a date for reconsideration. The 48 obsolete requests were removed from the active queue, while the duplicate entries were connected to the surviving record. The review did not decide that every current vacancy should be filled immediately; managers still had to consider priorities and affordability.

I opposed describing the 72 removed entries as workforce savings. No corresponding 72 employees had left, and the salary assumptions did not represent expenditure we had demonstrably avoided. The useful result was a more credible starting point for hiring decisions. It also exposed a responsibility within my function: we had allowed an administrative total to circulate without an adequate explanation of its contents. I changed the reporting note and the review ownership, rather than treating the correction as a one-time data clean-up that absolved us of the original weakness.

In 2023 a people-analytics pilot examined workforce patterns across 686 employees in ten service teams. The proposed next step was to give managers individual rankings indicating which employees appeared more likely to leave. Data colleagues had invested substantial work, and operating leaders wanted a practical way to focus retention conversations. I understood that intention. I was concerned about the meaning a named ranking would acquire once it left the analytical team.

A model estimate was not an employee’s stated intention. A manager might interpret a high score as disloyalty, withhold an opportunity or begin a conversation that the employee could neither understand nor correct. I worked with data director Daniel Corvess and the appropriate privacy and employment advisers to examine the proposed use. I asked the team to separate a potentially useful pattern about working conditions from a judgement attached to an identifiable person.

I declined the named-list rollout and supported a narrower aggregate review of conditions across teams. The pilot had already cost USD 390,000; redesign required a further USD 160,000. Some operating colleagues felt that the revised output was less actionable because it no longer identified whom to approach. I accepted that limitation. Managers retained responsibility for ordinary conversations and for responding to concerns employees actually raised, rather than relying on an undisclosed probability as the reason for attention.

The result was a changed use of the analysis, not evidence that turnover fell or that every risk was removed. We documented the intended audience, limits and review responsibility for the aggregate work. My conclusion was specific to this proposed use and the evidence available. It did not establish that all individual analytics are inappropriate, nor did specialist participation amount to a public certification. The decision showed that money spent developing an output does not settle whether management should put it into use.

An acquisition in 2024 brought 126 roles into a mapping exercise intended to support common personnel reporting. The initial plan emphasised matching titles and salary bands to the receiving organisation’s categories. That would have produced a quick completion date. It would also have treated several unlike responsibilities as equivalent before managers had examined the work. I wanted the record to distinguish a supported match from a convenient administrative choice.

We reviewed decision scope, reporting relationships and the responsibilities expected to continue after the transaction. Ninety-one roles had a supported match. Twenty-three needed a revised description before placement, while twelve required separate review because a material question remained. These were roles in the exercise, not a statement that 126 employees received identical changes to their terms. Individual employment matters followed the appropriate processes and advice outside the public account.

The integration sponsor wanted a single date at which all entries could be marked complete. I opposed that description while the twelve remained unresolved. I agreed a ten-week extension and USD 280,000 of additional review and communication expense. The people team explained the categories and the next step to affected managers, including which decisions still belonged to transaction and operating leaders. We did not use a new job description to imply that those wider decisions had already been made.

The cost was a later reporting milestone and more work for managers during an already demanding integration. The benefit was a more honest account of what had actually been compared. I would not infer from the completed mappings that employees experienced the transition as fair or that the acquisition achieved its commercial objectives. The twelve open roles remained visible in the final review rather than being hidden in a rounded completion percentage. That was a better basis for the next decision than a finished-looking report whose categories could not withstand questions about the work.

In 2016 an employee survey invited 880 colleagues across nine operating units and received 586 responses. A combined result was prepared for the leadership team. I initially accepted the presentation because the response count looked substantial. When my analysts examined participation and comments by context, it became clear that the single figure implied a common experience the evidence did not establish. I had approved a summary too early.

Some operating leaders wanted to preserve the headline for consistency with the previous year. Others wanted detailed unit comparisons that could expose small groups or make a difference in participation look like a judgement about local management. I proposed a more limited account: explain who had been invited, how many responded and which questions required further discussion, while respecting the boundaries necessary to protect individuals. We sought the appropriate advice on handling the material and did not circulate identifiable comments as illustrative anecdotes.

We allocated USD 145,000 to facilitated follow-up and reporting work. Managers received a clearer distinction between a recurring theme, a local concern and something that the survey could not tell us. The 294 nonresponses were not assigned an invented motive. Silence might reflect many circumstances; the response count alone could not establish them. Follow-up also needed to say what management could act on and which questions would take more time.

I accepted the loss of a simple year-on-year story. The revised report was harder to summarise and did not offer a single score that could stand for the entire organisation. It was more faithful to the evidence we actually held. I do not claim the expenditure increased trust or solved the issues raised. My own lesson was to examine the conditions of participation before allowing an aggregate result to become an assertion about what employees collectively believed.

In 2008 I supported a screening approach for regional service-planning recruitment that placed excessive weight on one educational route. The intention was to make a large application pool easier to review consistently. I had helped define the screen and initially regarded consistency as evidence that it was working. Hiring managers later showed us applications where relevant operating experience had been excluded before it could be considered.

We reviewed 216 applications and identified 54 for reassessment against a fuller account of the role’s requirements. That did not mean all 54 applicants were qualified or that the other 162 had been handled incorrectly. It meant the original shortcut had prevented a material part of the evidence from being examined in those 54 cases. I worked with recruiting colleagues and the appropriate advisers to revise the process and explain the additional review to the hiring team.

The reassessment led to 18 interviews, six offers and four acceptances. Those are stages within the reviewed group, not separate populations to be added together. The correction cost USD 95,000 and extended the timetable by five weeks. Some managers objected that reopening the process would undermine confidence in the first decision. I considered the opposite risk more serious: continuing to use a criterion we could not adequately connect to the work simply to protect the appearance of consistency.

The outcome did not prove that the four hires later performed better, or that a single revision settled every question about recruitment. We retained qualification requirements that could be explained and made room for relevant experience to be assessed. I acknowledged my part in the original design. That mattered because a people leader should be able to revise a method without pretending that the initial mistake belonged only to the system. I carried the distinction between consistent processing and defensible judgement into the analytical roles that followed.