HR has spent twenty years building systems that understand people. It has built almost nothing that understands work. That is becoming the biggest unmanaged risk in the function.
Ask a modern HR function about its people and the answers arrive in seconds. Headcount by function, by geography, by grade. Attrition, regretted and unregretted. Engagement scores cut twelve ways. Salary distributions, compa-ratios, diversity metrics, absence trends, learning hours, performance distributions, nine-box placements. Two decades of investment in HRIS platforms, analytics teams and dashboards have made the workforce one of the most measured assets in the modern organisation.
Now ask a different question.
What work does every role in this organisation actually perform today?
Not what the job description said when it was written. Not what the title implies. What the role actually does, now, this quarter.
Almost no organisation can answer. That silence matters, because everything HR is being asked to deliver over the next decade depends on the answer.
A system of record for people. Nothing for work.
The history of HR technology is a history of knowing people better. The payroll era captured who we employ and what we pay them. The HRIS era added structure: grades, reporting lines, locations, service history. The analytics era layered insight on top: engagement, skills, sentiment, flight risk. Each generation deepened the employee record.
Open that record for a single person and look at what it holds.
Employee 54872. Salary £68,000. Manager Jane Smith. Grade 7. Location London. Performance rating 4. Length of service six years.
Every field describes the person or their position in the structure. Not one field describes the work. The record cannot tell you that this role now carries regulatory accountability it did not carry two years ago. It cannot tell you that the role's decision rights expanded when a management layer was removed. It cannot tell you that a third of its original tasks are now performed by software, that its stakeholder map has doubled or that its real complexity bears little resemblance to the grade assigned in 2019.
The organisation holds rich, governed, audited data about its people. The work those people perform is documented, if at all, in a Word file nobody has opened since the last restructure.
Work has already drifted away from the record
This is not a theoretical concern. The research is stark.
Deloitte's global study of skills-based organisations, drawing on more than 1,200 workers and executives, found that 63% of executives say work in their organisations is already performed in teams or projects outside people's core job descriptions. Fewer than half believe their job descriptions do an excellent job of specifying the work to be done. Only 19% believe the traditional job is still the best way to organise work at all.
Read those numbers back. The people who run organisations report that most work has escaped its official documentation and that the documentation no longer describes reality. They keep using it anyway, because it is the only work data they have.
The pace of drift is accelerating. The World Economic Forum's Future of Jobs Report 2025, based on responses from more than 1,000 employers representing over 14 million workers, expects 39% of workers' core skills to change or become outdated between 2025 and 2030. The same report anticipates disruption to around a fifth of today's jobs over the period. A job description written this morning describes a role that will be materially different before the next pay review cycle completes.
AI changes the work. The title never notices.
AI is usually discussed as a destroyer of jobs. The data tells a subtler and more awkward story. AI rarely deletes the job. It rewrites the work underneath it, task by task, while the title, the grade and the job description remain untouched.
Anthropic's Economic Index, built from millions of real AI conversations mapped to official task taxonomies, found exactly this pattern. In its first analysis, around 36% of occupations were already using AI for at least a quarter of their tasks. Only about 4% used it across three quarters or more. Adoption is broad but partial. It lands inside roles, not across them.
That distribution has profound implications for pay, organisation design and workforce planning. If AI transformed whole jobs, the organisation chart would show you where change is happening. It does not. Change accumulates invisibly within roles. Two people can hold the same title in the same team while one of them has had 40% of their work absorbed by tooling and replaced with judgement, escalation and oversight. Every HR system still records them as the same job. One of those records is now wrong.
Skills are not the answer. Skills data is still people data.
The market's favoured response to all of this is skills. Build a skills taxonomy, infer capabilities from CVs and systems, deploy people against skills rather than jobs. There is real value in that agenda. It does not solve this problem, for a simple reason.
Skills describe the supply side. They tell you what a person can do. They tell you nothing about the demand side: what a role actually requires, how much judgement it involves, what it is accountable for, what happens when it goes wrong. An organisation can hold a perfect inventory of every capability its people possess and still be unable to say what any given role demands of them.
Deploying skills without work data is matching supply against a demand you have never measured. Pricing skills without work data is worse, because pay ultimately attaches to the value and demands of work performed, not to the possession of capability. Skills intelligence is a genuine advance. It is also, structurally, another layer of people data. The work side of the ledger remains empty.
What the gap costs
Poor work data does not sit quietly. It surfaces in some of the most expensive failures in the corporate landscape. Increasingly, it surfaces in courtrooms.
Equal pay litigation turns on precisely the question organisations cannot answer: whether different roles involve work of equal value in terms of demands such as skill, effort and decision-making. Birmingham City Council has paid out £1.1 billion in equal pay compensation since the Supreme Court ruled against it in 2012, a liability that contributed directly to the effective bankruptcy of Europe's largest local authority. In August 2024 an employment tribunal found that around 3,500 Next store staff, mostly women, had performed work of equal value to higher-paid warehouse staff, leaving the retailer facing back pay estimated at over £30 million. In February 2025 Asda lost a ruling on the same equal value question, with total exposure estimated above £1 billion.
In every one of those cases the employer held payroll data, grading data and employee records in abundance. What it lacked was reliable, current, defensible evidence of what its roles actually demanded.
The same gap runs through quieter failures. Workforce plans model an organisation that no longer exists. Succession decisions compare roles as they were documented, not as they are performed. Recruitment inherits stale specifications: research from Harvard Business School and Accenture found that 88% of employers believe qualified candidates are screened out of hiring processes because they fail criteria written into the job specification. AI programmes stall because nobody can say, at role level, which work is actually there to be automated or augmented.
Each of these problems is treated as a separate discipline with a separate fix. They share a single root cause. The organisation does not know its own work.
Regulation has removed the choice
Until recently an organisation could live with this gap indefinitely. In Europe, it no longer can.
The EU Pay Transparency Directive requires employers to base pay structures on objective, gender-neutral criteria, naming skills, effort, responsibility and working conditions. Workers gain the right to information about pay levels for work of equal value. Where an employer cannot demonstrate that its pay decisions rest on such criteria, the burden of proof in a pay discrimination claim shifts to the employer. The transposition deadline passed in June 2026. Gender pay gap reporting, with mandatory joint pay assessments where unexplained gaps exceed 5%, follows from 2027.
Notice what the Directive is really demanding. Skills, effort, responsibility and working conditions are not attributes of people. They are attributes of work.
The Directive is not really regulating pay. It is regulating work data.
The legislation in effect requires every in-scope employer to hold structured, current, defensible work data and to be able to produce it on request. An organisation that cannot describe its work cannot explain its pay. Very soon, it will not be permitted to try.
What good work data looks like
Closing the gap does not mean writing better job descriptions. Documents begin drifting the day they are approved. It means treating work as governed organisational data, held to the same standard as the employee record.
That data is structured, describing every role through consistent dimensions such as complexity, accountability, decision rights, judgement and stakeholder demands rather than free text. It is current, refreshed as roles change rather than when someone remembers. It is comparable, so that any two roles can be assessed on equal terms regardless of title, function or department. It is governed, with version history and an audit trail, because pay, grading and legal defence all depend on being able to show what a role demanded and when.
None of this is exotic. It is exactly the discipline organisations already apply to financial data, customer data and product data. Work is the last major class of organisational data still managed by folklore.
The next system of record
The first generation of HR technology told us who we employ. The second told us what they cost and how they perform. The prize now sits somewhere else entirely.
The next generation of HR technology will not win because it stores more employee records. It will win because it understands work: what every role actually performs, how that work is changing and what it is worth. Organisations that build that understanding will price work fairly, plan credibly, deploy AI deliberately and defend their decisions when challenged. Organisations that do not will keep managing a modern workforce through the job description, a document that stopped describing the job some years ago.
HR solved the people data problem. The next decade belongs to organisations that solve the work data problem. The next system of record will not be for people. It will be for work.
