Why Job Levelling & Evaluation is Broken

Last year we researched how organisations level and evaluate jobs, not how the textbooks say they should.

We spoke with customers and other global organisations about job levelling and evaluation - including how these practices both look in skills-first, Agentic operating models - and corroborated our findings with independent research covering hundreds of companies. Under the EU Pay Transparency Directive, the burden of proof now sits with the employer, so when an employee asks why a comparable colleague earns more, you have to evidence the decision, not just defend it. Against that standard, what we found was consistent: the systems most organisations rely on to decide what a job is worth are indefensible, slow, unscalable, disconnected and already out of date. 

Indefensible: You Can't Explain Why Pay Differs for Comparable Work 

Ask why two comparable roles are paid differently, and the honest answer is often that the evidence was never captured. Most job levelling and evaluation systems record the outcome and nothing else, so there is no audit trail showing how the decision was reached; it's no wonder 78% of organisations find it challenging to govern pay decisions. Roles drift out of line and grade inflation sets in quietly, unnoticed until an audit or a challenge forces a look back. AI has made this harder before it has made it easier: it can level a job in seconds, but speed alone strips out the factor-by-factor reasoning that makes a decision defensible when someone asks you to prove it. 

Slow and Costly: Months Lost to Back-and-Forth Over Grade 

Where a process does exist, it is often too slow to trust, with grading taking months and vacancies stuck behind endless back-and-forth over how the job should be defined and what it's worth. One global company told us hiring for a graded role took three to five months, largely due to the review and challenge cycle. Across the market, 83% of evaluations get redone at least once, usually because the underlying job information wasn't good enough to evaluate against. Organisations end up choosing between rigour and speed, when the process should never have forced that choice. 

Unscalable: The Logic Lives in One Person's Head 

Job levelling and evaluation was built for trained experts to manage, and most organisations have only a handful. Everyone else finds the system either too complicated to use with confidence or too light-touch to trust, so business areas build their own version, one that drifts from the core framework the moment nobody's watching. In-house methodologies often live in one person's head and a spreadsheet, so when that person leaves, the levelling logic leaves with them. With 86% of organisations still running levelling manually, that risk is the norm, not the exception. 

Disconnected: The HCM Captures the Outcome, Not the Decision 

Even where the process works, the evidence behind it doesn't live anywhere useful. Levelling sits in one tool, job architecture in another and job descriptions and skills data end up scattered across local drives and spreadsheets.

An HCM records the level but not the decision, while the rationale and factor scores behind it sit elsewhere, so there's no way to drill from a level back to the evidence that produced it. Meanwhile, managers are drafting job descriptions in Copilot, ChatGPT and Claude, with no levelling framework behind the output and no audit trail attached. 62% of organisations are dissatisfied with their job description governance, with just 16% connecting skill proficiency to pay, meaning skills and reward stay as two separate conversations. 

Out of Date: Built for a World of Work That No Longer Exists 

Underneath it all is a deeper problem: most frameworks were designed for a different era. Job content is changing faster than traditional methodologies were built to accommodate, and those methodologies still weight people leadership and budget most heavily, which undervalues exactly the roles carrying the most impact in an AI and agentic organisation.

Organisations want to move to a skills-first or Agentic operating models, but the underlying architecture isn't there to carry it. And because job architecture is treated as a one-time project rather than a living system, it goes stale within months of sign-off. 

These aren't five separate problems because they compound: a job levelling and evaluation framework that few people can use gets applied inconsistently, on out-of-date evidence, in a system nobody else can see into, and the result is a grade nobody can defend when it's challenged. That's the brief we set out to answer with RoleEvaluate: a hybrid job levelling and evaluation methodology built against the pay transparency legislation it will be tested by, with the governance and AI decision support needed to scale beyond the few experts holding it together. 

If you want to see what a connected, governed approach to job levelling and evaluation looks like in practice, join one of our webinars or watch a session on demand.