Posts on Workforce Data Lab

The Documented Vendor-Exit Pattern in Enterprise Alumni Software: An n=4 Migration Analysis

Mon, 28 Sep 2026 09:00:00 +0000

In recent years, four top-tier professional-services firms have migrated their corporate alumni programs off incumbent vendors onto a single destination platform. That is the whole dataset. This analysis treats those four documented migrations as the unit of observation, states plainly what a sample of n=4 can and cannot support, and asks a narrower question than “who is winning the market”: what changed about the buying decision at firms where we can observe a completed switch?

Cost-per-Hire Is an Allocation Problem: What Gets Excluded, Why Firms Cannot Compare, and a Decomposition FP&A Will Accept

Tue, 08 Sep 2026 09:00:00 +0000

Cost-per-hire looks like division: recruiting spend over hires. In practice the numerator is an allocation exercise, and most published figures are produced by allocating only the costs that are easy to see. The result is a metric that understates the true cost of hiring, varies with accounting choices more than with recruiting efficiency, and is almost never comparable across organisations. This piece covers the gap between advertised and fully-loaded cost, the costs that are silently excluded, why cross-firm comparison fails, and a decomposition that a finance team will accept.

Absence and PTO Analytics: Skewed Distributions, Calendar Artefacts, and the Burnout-Proxy Problem

Tue, 04 Aug 2026 09:00:00 +0000

Absence data looks like the simplest feed in the HR stack — days off, per person, per month — and is one of the easiest to misread. The distributions are not the shape analysts assume, the calendar contaminates every monthly comparison, and the field’s favourite use of the data, as a burnout early-warning signal, is weaker than its reputation. This piece covers all three, with the statistical posture we recommend for each.

Workforce Planning Scenarios: Driver-Based Models vs Budget Arithmetic, and the Assumptions Register

Tue, 14 Jul 2026 09:00:00 +0000

Most workforce plans are budget arithmetic wearing a strategy costume: last year’s headcount, plus or minus a negotiated percentage, spread across cost centres. That process produces a number. It does not produce a plan, because it contains no model of how headcount relates to anything the business actually does. Driver-based planning is the alternative, and its track record is mixed for an instructive reason — teams build models with dozens of drivers when the variance is carried by five. This piece covers the structural difference, the small driver set that matters, and the publishing discipline that separates a scenario from a guess.

Workforce Data Lineage: An Audit Framework for HRIS-to-Warehouse-to-Dashboard Pipelines

Tue, 09 Jun 2026 09:00:00 +0000

Most people analytics teams did not build the pipelines they report from. They inherited them: an extract written by a former HRIS administrator, a transformation layer maintained by a data engineering team that has since reorganised, a dashboard whose logic lives in a calculated field nobody has opened in two years. When a number is challenged, the team cannot say where it came from with any confidence, and trust in the whole function erodes faster than any single error would justify. This piece sets out a lineage audit framework built around six questions, and explains why running it changes the conversation about whether people data can be trusted.

GenAI in People Analytics: Where LLMs Help, Where They Fabricate, and a Validation Protocol

Tue, 19 May 2026 09:00:00 +0000

Large language models have entered the people-analytics workflow faster than the validation practices around them. Eighteen months of watching teams deploy them has produced a clear pattern: the failures concentrate in specific task types, not in “AI usage” generally. This piece separates where LLMs genuinely add capability from where they manufacture confident error, and closes with the validation protocol we now consider the floor for AI-assisted people analytics.

Where LLMs genuinely help

The reliable use cases share a structure: the model transforms text into structured, checkable categories, and a human can verify any given output.

Pay Equity Analysis: The Specification Choices That Change the Answer

Tue, 07 Apr 2026 09:00:00 +0000

Two competent analysts, given the same payroll extract and asked the same question — “is there a pay gap?” — can return materially different answers, and both can be defensible. The divergence is rarely in the statistics. It is in three specification choices made before the model is fit: the model form, the control set, and how the adjusted and unadjusted gaps are presented alongside each other. This piece walks through each choice and the reporting risks that follow.

Measuring Manager Effects on Attrition: Fixed Effects, Sorting, and What an Identification Claim Requires

Tue, 10 Mar 2026 09:00:00 +0000

“People leave managers, not companies” is the most quoted sentence in retention practice, and most of the analyses offered in support of it are correlations. A dashboard showing that some managers’ teams lose people at three times the rate of others establishes that attrition varies by manager. It does not establish that the manager caused it. The claim “bad managers cause attrition” is an identification claim, and identification requires a design. This piece covers manager fixed effects, the sorting problem, leave-in and leave-out estimation, and what a defensible manager-impact programme can actually estimate.

Internal Mobility Metrics That Hold Up: Definitions, Denominators, and the Gaming Problem

Tue, 17 Feb 2026 09:00:00 +0000

Internal mobility is the metric every organisation wants to report and few organisations define tightly enough to survive an audit. The numbers travel upward — to CHRO decks, board packs, ESG narratives — and each hop strips away a denominator. This piece lays out definitions that hold up for the three core metrics (internal fill rate, time-to-fill for internal moves, and the post-move outcome window), then catalogues the four places teams quietly game them.

How Vendor Benchmark Reports Get Constructed — and a Six-Question Filter for Using One as a Planning Input

Tue, 13 Jan 2026 09:00:00 +0000

People analytics teams are asked to benchmark constantly: attrition against industry, time-to-fill against peers, engagement against a norm. The benchmark usually comes from a vendor report, and the number is treated as an external fact. It is better understood as the output of a data-collection process with specific, usually undisclosed, design choices. Two benchmarks of the same metric from two reputable sources can disagree by a factor of two, and both can be honestly produced. This piece describes how benchmark reports get constructed, why they diverge, and the filter we apply before letting one into a plan.

Skills Data Quality: Inferred vs Attested Skills, and How to Decide with Bad Mirrors

Tue, 02 Dec 2025 09:00:00 +0000

Every skills-based initiative — talent marketplaces, workforce planning, succession analytics — inherits the error structure of whatever produced the skills data. That provenance is almost always one of two processes: inference (resume and ATS parsing, profile extraction) or attestation (self-report, manager endorsement). These two sources fail differently, predictably, and in ways that matter for which decisions each can support. This piece maps the error profiles and proposes a decision-bounding framework for the common situation: no ground truth available.

Time-to-Hire Is a Censored Measurement: Open Requisitions, Bimodal Medians, and a Reporting Spec That Survives Audit

Tue, 18 Nov 2025 09:00:00 +0000

Time-to-hire appears on nearly every talent acquisition dashboard, usually as a single average for the quarter. That average is almost always computed only over requisitions that closed, which means the slowest searches — the ones still open — are excluded by construction. The metric is systematically optimistic, and the error grows exactly when hiring gets harder. This piece covers the right-censoring problem, why the median hides structure across job families, why requisition-level and candidate-level definitions disagree by weeks, and a reporting specification we consider defensible.

Span of Control: What the Distribution Actually Looks Like, and Why the Ideal-Ratio Research Does Not Hold Up

Tue, 21 Oct 2025 09:00:00 +0000

Span of control is the most frequently benchmarked structural metric in organisation design and one of the most frequently misused. Delayering programmes routinely cite an “ideal” ratio — one manager to eight, to ten, and in a widely circulated claim, to fifteen — and set targets against it. The evidence behind those numbers rarely survives inspection. This piece describes what span distributions in large firms actually look like, why the ideal-ratio literature is weaker than its citations suggest, why hierarchy depth is often the better structural variable, and what span data can legitimately predict.

Engagement Survey Methodology: Why the Instrument Moves More Than the Intervention

Tue, 16 Sep 2025 09:00:00 +0000

The engagement survey is the most analysed and least scrutinised instrument in people analytics. Organisations invest in interventions to move scores by a few points, then change the questionnaire in the same cycle and attribute the resulting shift to the programme. This piece covers three methodological problems we consider structural rather than incidental — common-method bias, the anonymity-perception effect, and item-wording effects — and closes with survey-design guidance we now treat as baseline.

Survey Nonresponse Bias: The Missing Third, and Why Engagement Deltas Are Partly Response-Rate Deltas

Tue, 19 Aug 2025 09:00:00 +0000

A typical enterprise engagement survey closes with a response rate somewhere between 60% and 70%, and the readout proceeds as if the respondents were the workforce. They are not. The missing 30–40% are not a random sample of employees, and when the response rate moves between cycles, a portion of the reported score movement is a change in who answered, not a change in how anyone feels. This piece covers why nonresponse is directional, the minimum defensible correction, and the limits of what any correction can recover.

Headcount Analytics as a Slowly-Changing Dimension Problem: Why the Org Chart Needs Effective-Dated History

Tue, 22 Jul 2025 09:00:00 +0000

Ask three teams in the same organisation for last March’s headcount by department and you will usually get three numbers. None of the teams made an arithmetic error. Each pulled the current state of the HRIS and filtered it backwards, which means each answered a slightly different question: how many of today’s employees, in today’s departments, were employed last March. That is not last March’s headcount. This piece argues that the org chart is a slowly-changing dimension problem, sets out why point-in-time snapshots beat current-state pulls, and catalogues what breaks when reports read the current row.

Attrition Risk Models: What a Defensible Model Actually Needs

Tue, 10 Jun 2025 09:00:00 +0000

Most attrition models in production today would not survive a methods review. Not because the algorithms are wrong, but because the surrounding apparatus is missing: the features leak, the probabilities are uncalibrated, and nobody is watching the model after deployment. This piece lays out the four requirements we now treat as the bar for a defensible attrition model — feature families that generalise, calibration alongside discrimination, a disciplined leakage audit, and drift monitoring treated as the actual product.