Featured research · UCL MSc dissertation
AI, Data Skills and Advertised Pay in UK HR Roles
The Advertised Salary Premium for AI and Data Skills across UK HR Roles: A Quantitative Content Analysis of Online Job Advertisements
Quantitative content analysis and cross-sectional regression research
This study examines AI-related requirements separately from strictly defined advanced data skill requirements. It asks whether each is associated with advertised pay in UK HR roles, and whether those associations vary by job level or HR function.
Research questions
- How are AI and data skill requirements in UK HR job advertisements associated with advertised pay?
- Do these associations vary by job level or HR function?
Where the data came from
The study used a snapshot of job advertisements collected on 14 July 2026 through the official Reed.co.uk Jobseeker API. Searches using 12 groups of terms produced 23,253 deduplicated records. HR role screening and eligibility checks yielded a sampling frame of 2,539 eligible HR advertisements. After a pay audit, identification of fixed pay and restriction to full-time roles, the main analysis sample contained 2,229 advertisements.
The unit of analysis is the job advertisement, not the job seeker. Pay refers to advertised fixed or basic pay, not actual earnings after hiring.
My research work
Collecting recruitment data and auditing pay
I collected and organised job advertisements through the official API, distinguished fixed or basic pay from variable earnings such as commission and on-target earnings (OTE), standardised pay periods and handled extreme pay values according to the study’s rules.
Turning recruitment text into research variables
I first used dictionary searches to identify advertisements for review, then checked their full text in context. I distinguished requirements for candidates’ AI capabilities or willingness to learn from employer-provided AI tools and unrelated mentions, rather than treating every keyword match as a skill requirement.
Statistical modelling and checks
Using Python and statsmodels, I built three OLS models with progressively added controls and HC1 robust standard errors. I also examined interactions with job level and HR function and conducted several robustness checks.
Connecting the findings to HR practice
I discussed the findings in relation to job responsibilities, capability development and pay benchmarking, while distinguishing conclusions supported by the data from explanations requiring further research.
Defining the two types of requirement
AI-related requirements
Advertisements explicitly require candidates to have AI-related capabilities, or express understanding, openness, willingness to learn or readiness in relation to AI. References only to AI tools used or provided by the employer do not count as candidate AI requirements.
Strict Advanced Data requirements
Relevant terms must refer to candidates’ skills or experience, or substantive analytical responsibilities in the role. Tool names or keyword lists without job-specific context are not automatically coded as positive.
Strict Advanced Data is the dissertation’s operational definition, not a definition of all data capabilities.
Main findings
Skill requirements and advertised pay
Fully adjusted model · Percentage differences and 95% confidence intervals
AI-related requirements
+16.34%Statistically significant
95% CI 6.36%–27.25% · p < .001
Strict Advanced Data requirements
+1.97%Not statistically significant
95% CI -2.99%–7.19% · p = .443
Advertised pay difference (%) · Dashed line: 0%
Points show estimated differences in advertised pay; lines show 95% confidence intervals. Both requirements enter the model together, accounting for job level, HR function, London location, recruitment-agency-related variables, pay basis and posting period.
Redrawn from the main model results in section 4.2 of the author’s MSc dissertation.
These estimates describe associations in advertised pay after accounting for observed job characteristics; they are not causal effects. The main sample included 34 advertisements with AI-related requirements. The 16.34% estimate should not be interpreted as the pay rise an individual would receive from using AI.
Finding 1: AI-related requirements retained a significant positive association with advertised pay.
In the fully adjusted model, AI-related requirements were associated with an advertised pay difference of +16.34%, with a 95% confidence interval of 6.36%–27.25%, p < .001.
Finding 2: The adjusted estimate for Strict Advanced Data requirements was not statistically significant.
The estimated difference was +1.97%, with a 95% confidence interval of -2.99%–7.19%, p = .443. This does not mean that data analysis has no value, or that the association between data skills and pay is exactly zero.
Finding 3: No clear differences by job level or HR function were detected.
The interaction analyses did not provide sufficient evidence that these associations vary systematically by job level or HR function. This does not establish that all groups are identical.
As controls were added, the AI estimate fell from 36.79% to 20.22% and finally 16.34%. The Strict Advanced Data estimate fell from 15.90% to 2.54% and finally 1.97%. The study therefore distinguishes specific requirements rather than treating all technical capabilities as one category.
Implications for HR practice
Make job requirements more specific.
Distinguish between using AI tools, implementing systems and taking responsibility for AI-supported decisions, rather than simply asking for “familiarity with AI”.
Link capability development and pay benchmarking to tasks.
Technical depth, implementation scope, decision consequences and governance responsibilities all matter. The 16.34% estimate cannot be applied directly to every role or individual.
Connect analysis with action.
Data analysis should connect with a clear business question, communication of results and subsequent decisions, rather than simply listing tools.
These are practical implications proposed in the dissertation’s discussion, not verified outcomes of implementation in a business.
Research boundaries
This study uses job advertisements from a single platform at a single point in time, with only 34 AI-positive advertisements in the main sample. It estimates conditional associations, not the causal effect of AI requirements on pay, and does not control for employers’ own wage levels. It observes advertised fixed or basic pay rather than actual wages. The findings cannot be directly generalised to the whole UK labour market or all HR roles.