Skip to content
HR HubsMedia
Subscribe
Core HR (HCM / HRIS)

The State of HR Tech 2026

The AI Act deferral to December 2027 and what it requires, the four foundations every HR technology initiative depends on, the evidence gaps across engagement, learning and workforce planning, and a buyer's framework.

22 Sep 20268 min read

This report reviews the published record. It is not a survey, and where it would ordinarily quote an industry benchmark it frequently does not, for a reason set out in section four.

Four movements that define the yearTHE 2026 PICTUREFour movements that define the yearOne regulatory date now governs planningAI in employment and worker management is high risk under EU law, withobligations applying from 2 December 2027 after a sixteen-monthdeferral.FixedThe foundations did not changeJob architecture, position data, retained justification and mapped dataflows still determine whether anything else works.UnchangedAI value is administrative, not decisionalThe defensible wins are in high-volume, low-consequence work. Thedecisional applications carry an evidence regime.SortedThe evidence base did not improveThe categories with the most confident marketing remain the ones withthe least published evidence.UnchangedHOW THIS REPORT IS BUILTIt reviews the published record and setsout frameworks. Where a widely usedfigure could not be traced to a sourcewith a sample size and a date, we haveleft it out rather than repeat it.HR Hubs Media, September 2026.
The structural position of HR technology heading into 2027, each stated with the evidence available rather than the evidence one would want.

1. One date now governs HR technology planning

Annex III of Regulation (EU) 2024/1689, the AI Act, classifies AI systems used in employment and worker management as high risk. That classification reaches recruitment and selection, decisions affecting terms of work, promotion and termination, task allocation, and monitoring and evaluation of performance and behaviour.

Those obligations were originally to apply from 2 August 2026. They now apply from 2 December 2027, deferred by Regulation (EU) 2026/1744, the Digital Omnibus on AI, which entered into force on 27 July 2026. Annex I product-embedded systems move to 2 August 2028.

Anyone planning against those dates should confirm them against the operative article of the amending regulation rather than a summary page, and check whether the Annex III numbering changed.

What high-risk classification requires is an evidence regime rather than a prohibition: risk management, data governance, technical documentation, logging for traceability, information to the deployer, human oversight, accuracy and robustness, quality management, post-market monitoring and serious-incident reporting. Obligations fall on deployers as well as providers, which means buying a compliant product does not produce a compliant deployment.

Every obligation is a record, and records cannot be created retrospectivelyWHAT THE SIXTEEN MONTHS ARE FOREvery obligation is a record, and records cannot be created retrospectively01InventoryEvery system thatinfluences a decisionabout a person,including ones boughtas something else.02ClassifyAssistive ordecisional, measured byoverride rate ratherthan asserted from theproduct name.03InstrumentLog inputs and outputsfrom now. This is theobligation that cannotbe met retrospectively.04TestOn your own population,at a fixed interval,with a named reviewer.05DocumentPurpose, data,oversight, monitoring.The deliverable is afile, not a feature.The second and third steps are the ones organisations most often defer, and they are the two that gate everything else.
The deferral to December 2027 is build time. An organisation arriving at that date without these will be in the position it would have been in August 2026, having had sixteen extra months.

The single most useful governance metric in HR technology is the override rate: how often a human rejects the system's recommendation. Almost nobody measures it, and without it every classification of a tool as assistive is an assumption.

2. The foundations did not change, and they still decide everything

Four capabilities determine whether any HR technology initiative produces a result. None is a product. All four appear repeatedly across this publication's launch coverage because the same four keep surfacing under different initiative names.

A job architecture that means the same thing everywhere. Pay transparency reporting, skills-based hiring, workforce planning and internal mobility all aggregate through job structure. An organisation whose levels differ by region or by acquisition history cannot do any of the four properly, and typically funds them as four separate programmes each of which quietly fails for the same reason.

Position data distinct from people. Without positions as objects, a vacancy is an inference from a headcount number held in finance, and that inference breaks on part-time arrangements, long-term absence, contractor cover and deliberately held roles.

Retained justification for decisions. Systems record what someone is paid and when it changed. They rarely record why. That gap becomes acute the moment a pay transparency regime, an employment claim or an algorithmic accountability question arrives, and it cannot be filled retrospectively.

Mapped data flows. Most organisations discover during a mapping exercise that employee data reaches more places than anyone listed, generally through reports and exports rather than documented integrations. Every data protection obligation and every vendor-change decision depends on that map existing.

The pattern worth noticing is that all four are unglamorous, none is a purchase, and each supports several initiatives at once. They are consistently funded by no one because they belong to no single programme.

3. Where AI in HR is defensible, and where it is a decision

Sorting AI applications by capability produces an exciting list. Sorting them by consequence produces a useful one.

The line is where a mistake stops being an inconvenienceSORTED BY CONSEQUENCEThe line is where a mistake stops being an inconvenienceAnswering policy questions from documentedsourcesLowSummarising and routing HR service ticketsLowExtracting document data into structured fieldsLowScheduling and candidate communicationsLowSummarising employee feedback into themesModerateScreening or ranking candidatesHighScoring performance, potential or flight riskHighFeedback summarisation is included in the upper group deliberately: if it determines what leadership hears, it sets the agenda.
An assessment of common HR AI applications by the consequence of an error rather than by technical sophistication. The lower group attracts the AI Act evidence regime and, where decisions are solely automated, data protection obligations as well.

The defensible pattern across organisations getting value is consistent. Deploy into administrative volume where errors are visible and reversible, keep people in the consequential decisions, ground everything in the organisation's own documents, and be able to state which category each tool sits in.

The application most often misclassified is the recommendation that is technically advisory and practically determinative. A ranking followed in the overwhelming majority of cases is a decision with a human signature on it.

4. The evidence base, and what it costs buyers

This is the section we would most like to be able to shorten in future editions.

Across HR technology, and across the adjacent categories our sister titles cover, the same failures recur: figures that cannot be traced to a sample size or a fielding period, correlational findings presented as effects, and at least one widely quoted statistic that turns out to be a remark made once from a conference stage and never replicated.

Preparing this publication's launch coverage, we looked for credible published answers to questions HR leaders ask routinely. These are the ones we could not find.

  • Whether engagement platforms change engagement. No controlled study establishing that deploying a platform changes the outcome. What exists is benchmark databases, correlational outcome studies, selected case studies and action-completion metrics, none of which addresses the question.
  • Learning completion benchmarks that separate mandated from discretionary content. Without that separation the aggregate figure is carried by compliance training and means nothing.
  • Whether onboarding or enablement programme design shortens ramp. No source publishes design against measured outcome.
  • HR technology implementation cost against budget. Discussed constantly in private, published nowhere.
  • Override rates on automated recommendations in hiring. The most important governance number in the category. No organisation publishes it.
  • Any HR technology market size or stack size figure that we could obtain from a primary source in this research pass.

An absence is a finding. A category that has sold for a decade on an outcome it has never evidenced is telling buyers something, and it is not what the case study says.

5. A buyer's framework for the next twelve months

  • Ask for the sample size and the fielding period on every figure in a vendor deck. Note which produce an answer.
  • Ask what the product can remove, not what it can rank. The governance weight follows the ability to exclude a person.
  • Ask what it logs, and for how long. Under an evidence regime the log is the deliverable.
  • Ask to see a real explanation output for a rejected candidate or a low score, not a description of one.
  • Hold out a comparison group in any pilot where the claim is about an outcome. Assign it rather than letting teams volunteer.
  • Fix the foundations before the feature. Job architecture, position data, decision justification and data flow mapping. They are cheaper than any of the initiatives that depend on them.
  • Instrument the override rate now on anything that produces a recommendation about a person.

On this report and the next one. This edition reviews the published record and sets out frameworks. It is not original research and it does not pretend to be. Our People Stack research series specification is published separately and in advance of fielding, which is the standard we intend to hold ourselves to: sample size, fielding period, the questions as asked, distributions rather than only averages, and the findings that are inconvenient reported alongside the ones that are not.

Our coverage of vendors, products and workplace regulation is journalism. Nothing on this site is legal, employment or procurement advice. This report reviews the published record and sets out a framework; it is not original survey data and we do not present it as such.

References

Every figure and legal citation in this article is drawn from the sources below. Where an instrument is proposed rather than in force we say so in the text.

  1. European Union, Regulation (EU) 2024/1689, the AI Act, Annex III on employment and worker management. artificialintelligenceact.eu/annex/3
  2. European Union, Regulation (EU) 2026/1744, the Digital Omnibus on AI, deferring Annex III obligations to 2 December 2027, in force 27 July 2026. eur-lex.europa.eu/eli/reg/2026/1744/oj/eng
  3. European Commission, AI Act policy page, application dates and high-risk obligations, updated 27 July 2026. digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai
  4. European Union, Regulation (EU) 2016/679, the GDPR, including Article 22 on automated individual decision-making. eur-lex.europa.eu/eli/reg/2016/679/oj
  5. Government of India, The Digital Personal Data Protection Act, 2023. meity.gov.in/data-protection-framework
  6. Information Commissioner's Office, Guide to PECR, under review following the Data (Use and Access) Act 2025, page dated 20 August 2025. ico.org.uk/.../guide-to-pecr

How we work. This report was researched and written by the HR Hubs Media editorial team. We do not republish press releases. Every number and legal citation is checked against a primary source, named and linked above. Corrections are made openly on the article itself, never by silent edit. If you believe something here is wrong, write to info@hrhubsmedia.com and tell us what and why.

Related coverage

More reports

All reports →