In the lumi reward benchmark, 57.7% of UK organisations report no AI use in pay decisions, and 42.4% use AI tools in pay decisions in some form (19.9% + 16.5% + 6.0%) (n = 267, 2026 H1).
Key findings
- 57.7% of organisations report no AI use in pay decisions (n = 267), the common answer. 42.4% use AI tools in pay decisions in some form (19.9% + 16.5% + 6.0%).
- 16.5% use AI in pay decisions with no formal oversight. 6.0% combine human oversight with bias auditing (n = 267).
- 75.8% do not disclose to employees when AI has been used in decisions affecting their pay (n = 186, "Not applicable" excluded). 9.1% always disclose.
- 60.5% of reward functions do not use AI tools for benchmarking or market analysis (n = 266). 10.9% have them embedded in process.
- 92.1% pay no premium specifically for AI skills (n = 267). 89.2% say AI proficiency does not influence base pay review outcomes for any population (n = 268).
- 64.9% operate no differentiated reward arrangements for AI or digital talent (n = 265). 15.5% run a formal programme.
- 64.3% have not re-graded or re-evaluated any role in the last 12 months specifically because AI changed job content (n = 263). 6.8% have done so for several roles.
About the data
This paper reports how UK employers answered lumi's questions on AI in reward. The lumi reward benchmark covers 269 UK organisations across 14 sectors and five size bands. Data was collected in the 2026 H1 collection window. Each figure gives its base (n): the number of organisations that answered that question. Percentages are rounded to one decimal place, so a distribution can add up to 100.1%. Figures here are national. Sector and size comparisons are available to lumi members. See how lumi works for the method.
The legal context
The practices covered in this paper, from the use of AI in pay decisions to premiums for AI skills, are not set by statute.
How many UK employers use AI in pay decisions, and with what governance?
The first question asks two things at once: Do you use AI tools in pay decisions, and with what governance? Each answer records whether AI is used and, where it is, what checks sit around it (n = 267).
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No AI use | 57.7% | 154 | common |
| AI as decision-support with human oversight | 19.9% | 53 | rare |
| AI with no formal oversight | 16.5% | 44 | rare |
| AI with human oversight and bias auditing | 6.0% | 16 | rare |
No AI use is the common answer, at 57.7%. Taken together, 42.4% of organisations use AI tools in pay decisions in some form (19.9% + 16.5% + 6.0%). Each AI answer is rare; the most frequently chosen of them is decision-support with human oversight, at 19.9%. Human oversight in either form accounts for 25.9% (19.9% + 6.0%), and 6.0% (16 organisations) add bias auditing to that oversight.
For someone designing reward, the split matters because the same tool can sit inside very different processes. Human oversight places a person between the tool's output and the pay outcome. Bias auditing adds a check at a different level: on the pattern of outcomes across a group of employees, not only on each decision. The "no formal oversight" answer does not describe the tool or how it is used; it records only that no formal check sits around it. It could cover anything from a feature built into a pay system to informal use by individual managers.
Disclosure is asked separately: Do you disclose to employees when AI has been used in decisions affecting their pay? This question excludes "Not applicable" from its base. 186 organisations gave an answer on disclosure (n = 186); 80 answered "Not applicable" and are not counted.
"No" is the common answer, at 75.8%. 15.1% disclose on request and 9.1% always disclose; both answers are rare. Some form of disclosure, on request or always, accounts for 24.2% (15.1% + 9.1%).
The two disclosure answers put the onus in different places. Disclosure on request depends on an employee knowing there is something to ask about. Always disclosing makes AI use part of how a pay outcome is communicated. These figures are shares of the 186 organisations that answered this question, not shares of those that use AI in pay decisions.
Do UK reward teams use AI tools for benchmarking or market analysis?
The next question turns from decisions about individuals to the reward team's own analytical work: Does your reward function use AI tools for benchmarking or market analysis? (n = 266)
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No | 60.5% | 161 | common |
| Exploring | 15.8% | 42 | rare |
| Piloting | 12.8% | 34 | rare |
| Yes — embedded in process | 10.9% | 29 | rare |
"No" is the common answer, at 60.5%. 10.9% have AI tools embedded in their benchmarking or analysis process. Piloting and embedded use together account for 23.7% (12.8% + 10.9%). Adding those exploring, 39.5% report some activity (15.8% + 12.8% + 10.9%).
Exploring, piloting and embedded are statuses at the time of collection. The data does not show whether organisations move from one to the next.
For reward design, AI in benchmarking affects pay at one remove. Benchmarking produces the reference points, such as range midpoints and positioning against comparator groups, that later pay decisions draw on. A tool used here shapes the frame for decisions rather than the decisions themselves. The oversight questions differ in kind: they concern the source data, the method and whether a result can be traced back to its inputs, rather than a review of each outcome.
Do UK employers pay more for AI skills and AI talent?
Three questions ask whether AI skills or AI talent attract different pay treatment. They work through different mechanisms: a premium attached to a skill, the base pay review, and separate arrangements for a talent group.
The first is direct: Do you pay a premium specifically for AI skills? (n = 267). "No" is the common answer, at 92.1% (246 organisations). 7.9% (21 organisations) pay a premium, which is rare.
The second looks at the annual review of base pay: Does AI proficiency influence base pay review outcomes for any population? (n = 268)
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No | 89.2% | 239 | common |
| Yes formally | 4.1% | 11 | rare |
| Planned | 4.1% | 11 | rare |
| Yes informally | 2.6% | 7 | rare |
"No" is the common answer, at 89.2%. AI proficiency influences pay review outcomes, formally or informally, at 6.7% of organisations (4.1% + 2.6%). A further 4.1% report it as planned. Including plans, the figure is 10.8% (4.1% + 4.1% + 2.6%).
The difference between formal and informal influence matters for design. Formal influence implies a defined measure of proficiency and a documented link to the pay outcome. Informal influence works through managers' judgement, which is harder to see and to check afterwards. The question asks about "any population", so an organisation answering yes may apply this to a single group of employees.
The third question is about group-level arrangements: Do you operate differentiated reward arrangements for AI/digital talent (e.g. separate ranges, RSUs, retention vehicles)? (n = 265). "No" is the common answer, at 64.9%. 19.6% decide case by case and 15.5% run a formal programme; both answers are rare. Some differentiation, case by case or formal, accounts for 35.1% (19.6% + 15.5%).
This question covers AI and digital talent together, and names specific tools, so it is wider in scope than the skills premium question. A case-by-case approach gives flexibility on individual hires and retention risks. A formal programme sets out who is eligible and on what terms, which makes the arrangement easier to explain and to apply consistently.
Are UK employers re-grading roles because AI has changed job content?
The last question looks at grading: Have any roles been re-graded or re-evaluated in the last 12 months specifically due to AI changing job content? (n = 263)
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No | 64.3% | 169 | common |
| Review underway | 16.0% | 42 | rare |
| Yes — isolated cases | 12.9% | 34 | rare |
| Yes — several roles | 6.8% | 18 | rare |
"No" is the common answer, at 64.3%. 19.7% have re-graded or re-evaluated at least one role for this reason (12.9% + 6.8%): 12.9% in isolated cases and 6.8% across several roles. With reviews underway included, the figure is 35.7% (16.0% + 12.9% + 6.8%).
Job evaluation links what a job involves to where it sits in a grading structure. When job content changes, re-evaluation is the route by which that change reaches grade and pay range. The question does not record the direction of any change: a re-evaluation can move a role up, move it down or confirm its current grade. It also asks about changes made "specifically due to AI", so re-grading where AI was one factor among several may sit within the "No" answers.
What the data means for reward decisions
The figures describe choices, each with its own costs. A rare answer is not a wrong one.
Oversight of AI in pay decisions. Human oversight adds a review step to each AI-supported decision. Bias auditing adds a further check across outcomes, with its own demands on data and time. Using AI with no formal oversight avoids those steps, and leaves no formal record of how a tool's output became a pay outcome. Questions a reward team might ask:
- At which point does a person review the tool's output, and can they override it?
- Is an override recorded, and by whom?
- Could the organisation explain an individual pay outcome where a tool contributed?
Disclosure to employees. Not disclosing keeps pay communication as it is. Disclosure on request responds to employees who ask. Always disclosing makes AI use visible by default, and needs a working definition of when AI "has been used". Questions a reward team might ask:
- Does drafting, analysis or a recommendation count as use?
- Who answers an employee's follow-up question, and with what information?
AI in benchmarking. A tool used for analysis shapes the reference points behind many pay decisions at once. Questions a reward team might ask:
- What data sits behind the tool's output, and can the team trace a result back to it?
- How are tool outputs checked against other sources before they inform ranges?
Paying for AI skills and talent. A premium or a formal programme signals that an organisation values a skill and gives a clear basis for offers. It also creates a pay element that needs a definition of the skill, a way to assess it and a rule for when it stops applying. A case-by-case approach avoids that structure, at some cost to consistency across similar cases. Questions a reward team might ask:
- How would the organisation define and assess AI proficiency?
- Would the same treatment apply to employees who gain the skill in role and to new hires?
- How would the arrangement be explained to colleagues in similar roles who sit outside it?
Job content and grading. Re-evaluating roles when AI changes their content keeps grades tied to what jobs involve, at a cost in evaluation time. Leaving grades unchanged avoids that cost, and can open a gap between what a job involves and how it is graded. Questions a reward team might ask:
- What triggers a re-evaluation: a manager's request, a scheduled review or a change in role content?
- How would a re-evaluation that moves a role down be handled?
Frequently asked questions
What percentage of UK employers use AI in pay decisions? In the lumi reward benchmark, 42.4% of UK organisations use AI tools in pay decisions in some form (19.9% + 16.5% + 6.0%) and 57.7% report no AI use (n = 267, 2026 H1). Within that, 19.9% use AI as decision-support with human oversight and 16.5% use it with no formal oversight.
Do UK employers tell employees when AI has been used in decisions affecting their pay? In the lumi reward benchmark, 75.8% of UK organisations do not disclose to employees when AI has been used in decisions affecting their pay (n = 186, "Not applicable" excluded, 2026 H1). 15.1% disclose on request and 9.1% always disclose.
Do UK employers pay a premium for AI skills? In the lumi reward benchmark, 92.1% of UK organisations pay no premium specifically for AI skills (n = 267, 2026 H1). 7.9%, or 21 organisations, do pay one.
Does AI proficiency influence base pay review outcomes in the UK? In the lumi reward benchmark, 89.2% of UK organisations say AI proficiency does not influence base pay review outcomes for any population (n = 268, 2026 H1). 4.1% say it does formally and 2.6% informally, with a further 4.1% reporting it as planned.
Do UK reward teams use AI tools for benchmarking or market analysis? In the lumi reward benchmark, 60.5% of UK reward functions do not use AI tools for benchmarking or market analysis (n = 266, 2026 H1). 10.9% have them embedded in process, 12.8% are piloting and 15.8% are exploring.
Do UK employers offer differentiated reward for AI and digital talent? In the lumi reward benchmark, 64.9% of UK organisations operate no differentiated reward arrangements for AI or digital talent (n = 265, 2026 H1). 19.6% decide case by case and 15.5% run a formal programme.
Have UK employers re-graded roles because AI changed job content? In the lumi reward benchmark, 64.3% of UK organisations have not re-graded or re-evaluated any role in the last 12 months specifically because AI changed job content (n = 263, 2026 H1). 12.9% have done so in isolated cases and 6.8% across several roles, and 16.0% have a review underway.
Notes
- Base (n) is the number of organisations that answered each question. Percentages are rounded to one decimal place and are not re-normalised, so a distribution can add up to 100.1%.
- Combined figures add the listed percentages of options within a single question, with the parts shown. They carry the rounding of their parts.
- "Not applicable" answers sit inside the base unless a question states otherwise. One question excludes "Not applicable" from its base: the question on disclosing to employees when AI has been used in decisions affecting their pay. Its base is 186 organisations; 80 answered "Not applicable" and are excluded.
- Prevalence words: common is the most frequently chosen answer; alternative is 20% or more but not the most common; rare is under 20%. No answer in this paper falls in the alternative band.
- All figures are national, from the 2026 H1 collection window.
Data appendix
Every question on this page, with its base and full distribution, as a spreadsheet: download the CSV.
All national figures used in this paper, as recorded in the lumi reward benchmark (collection window 2026 H1).
Do you use AI tools in pay decisions, and with what governance?
Base: 267 organisations
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No AI use | 57.7% | 154 | common |
| AI as decision-support with human oversight | 19.9% | 53 | rare |
| AI with no formal oversight | 16.5% | 44 | rare |
| AI with human oversight and bias auditing | 6.0% | 16 | rare |
Do you pay a premium specifically for AI skills?
Base: 267 organisations
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No | 92.1% | 246 | common |
| Yes | 7.9% | 21 | rare |
Does AI proficiency influence base pay review outcomes for any population?
Base: 268 organisations
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No | 89.2% | 239 | common |
| Yes formally | 4.1% | 11 | rare |
| Planned | 4.1% | 11 | rare |
| Yes informally | 2.6% | 7 | rare |
Does your reward function use AI tools for benchmarking or market analysis?
Base: 266 organisations
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No | 60.5% | 161 | common |
| Exploring | 15.8% | 42 | rare |
| Piloting | 12.8% | 34 | rare |
| Yes — embedded in process | 10.9% | 29 | rare |
Do you operate differentiated reward arrangements for AI/digital talent (e.g. separate ranges, RSUs, retention vehicles)?
Base: 265 organisations
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No | 64.9% | 172 | common |
| Case-by-case | 19.6% | 52 | rare |
| Yes — formal programme | 15.5% | 41 | rare |
Have any roles been re-graded or re-evaluated in the last 12 months specifically due to AI changing job content?
Base: 263 organisations
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No | 64.3% | 169 | common |
| Review underway | 16.0% | 42 | rare |
| Yes — isolated cases | 12.9% | 34 | rare |
| Yes — several roles | 6.8% | 18 | rare |
Do you disclose to employees when AI has been used in decisions affecting their pay?
Base: 186 organisations · "Not applicable" excluded from base
| Answer | % of base | Organisations | Prevalence |
|---|---|---|---|
| No | 75.8% | 141 | common |
| On request | 15.1% | 28 | rare |
| Yes — always disclosed | 9.1% | 17 | rare |
See how your organisation compares with other UK employers. Book a call
David Whitfield, AI in reward and pay decisions: UK benchmark 2026. lumi, collection window 2026 H1. https://lumihr.co.uk/research/ai-in-reward-2026Last reviewed 20 September 2026 · Figures describe organisations in the lumi reward benchmark, not a random sample of UK employers. Every figure on this page states the question asked and the number of organisations that answered it.