Strategic HR decisions — powered by data

The report that asks rather than lists

Most HR reports are descriptive: they show what is recorded. Useful for reconciliation, but blind to what is missing — an unregistered employee is not in them at all, and a certificate with no expiry date appears in no alert.

The different report compares two states. The insurance follow-up compares who is at the company against who is registered, producing two actionable lists: those past their waiting period and not yet registered, and those past the age at which the subscription ends who are still registered.

The qualification check does the same for training: whose certificate is nearing expiry, and who was assigned a competency and is missing one of its courses — with one datum that prevents duplicated work: is training already scheduled for them?

The difference between the two kinds is the difference between knowing the present and knowing what it lacks. The first reassures you about what you have; the second tells you what to do this week.

Who sees whom: visibility at the row

The most common data leak in HR systems comes not from a breach but from a permission: a manager opened a report they were entitled to open and found the salaries of people who do not report to them. They broke nothing — the system gave them what they asked for.

So permissions work at two levels: which screens a user opens and what they may do there, then which rows they see inside them. The second is what actually protects privacy, and it applies to every list and every report in the system — not to one screen.

The common wrong fix is blocking the manager from the screen entirely, so they request reports by email — and the data travels in attachments governed by no permission and sits in mailboxes for years. A total block produces a wider leak than the one it prevents.

Why a past month's figures move

Run a past month's payroll report twice, six months apart: do the same figures come out? In many systems the answer is no — because the report does not read a stored figure but recomputes it from current data.

If an attendance rule changed, an insured wage was amended or an old punch corrected, a past month comes out different from what was actually paid. The danger is that the change is silent: nobody reruns an old report to compare, so the gap persists until an external party asks about a figure.

Closing the month is what prevents that: afterwards, transactions in the locked period cannot be edited. The practical test for any system: export a closed month's report and keep it, rerun it three months later and compare — if a single character differs, the system is not protecting your figures.

Obligations that never appear on an expense statement

Some of a company's largest obligations to its people appear in no expense line because they have not been paid yet — and are nonetheless real and owed.

The leave liability

— unused leave is owed either as days or as money at end of service. The leave-cost report turns it into a figure instead of leaving it unknown until someone leaves.

Compensatory leave

— overtime compensated as leave is a deferral, not a cancellation — it later appears as paid days away or as an amount at settlement.

The employer's insurance share

— a real cost that never shows on a payslip because it is not between company and employee — and failing to record it makes labour cost look lower than it is.

Medical insurance

— the company share is recorded separately and never posted to pay. Its value is measured at renewal by usage against premium, not by headcount.

Aggregation needs an affiliation

An individual figure's value is limited; the value is in aggregation: is this team better than that one, is this line improving or declining? Questions that go unanswered unless each employee is tied to their place in the tree and to their supervisor.

The common error is aggregating by department while actual operations are organised into lines of business that cut across departments — producing reports that are structurally correct and operationally meaningless.

The same applies to money: a cost centre is a decision independent of the org structure, not a copy of it. A project staffed from three departments is one cost centre, not three — and assuming they coincide is convenient at installation and costly at the first report needing detail the structure cannot provide.

Who did what, and when

Closing the month prevents changing the past, but says nothing about what happened before it. Who changed a wage, who deleted a transaction, who edited a permission — questions asked later that only an operations log answers.

A log's value is not catching bad faith — that is rare — but explaining an error: a figure came out odd, and the log says when it changed and at whose hand, so the cause is known in minutes rather than a days-long investigation ending nowhere.

It is completed by the permissions report showing what each user can do. Because permissions accumulate and never shrink on their own: granted when needed and forgotten afterwards, leaving people with access earned for a task that ended a year ago.