Pay Ranges for Emerging Roles: How to Price Jobs With No Market Data

Setting pay ranges for emerging roles means pricing a job before the market has a reliable benchmark for it: you build the range from established proxy roles, weight them by the skills the new job really needs, and add a premium only where evidence shows that one is needed. That is how to price a new role such as AI engineer, prompt engineer or AI governance lead in a way you can explain to leadership, candidates and, increasingly, regulators.

The problem is real. TalentUp’s job-ad data shows that AI-titled postings in seven European countries rose from about 10 to 32 for every 100 software engineer ads between 2024 and 2026. Yet for many of these titles, fewer than 20 ads per country state a salary, too few for a reliable benchmark. This guide sets out a five-step proxy method, the 2026 pay premiums we see in job ads, a worked example and the documentation you will need once the EU Pay Transparency Directive is transposed.

Why Standard Benchmarking Breaks Down for New Roles

Traditional benchmarking matches a job title to a dataset, reads the 25th, 50th and 75th percentiles and builds a band around them. That works when the title is used consistently and the sample is large enough. With emerging roles, neither holds.

• Titles mean different things. One company’s “prompt engineer” writes and tests chatbot prompts; another’s builds multi-step agent workflows connected to core systems. “AI engineer” covers anything from an apprentice to a founding engineer at £180,000 plus equity, both visible in our 2026 sample.
• Samples are thin. In January–September 2026, TalentUp found 272 prompt engineer, 441 AI governance/ethics and 874 AI product manager postings across seven large European markets, but no country had more than 20 salaried ads for any of these three titles.
• Candidates fill the gap. Without a reference point, hiring managers anchor on the candidate’s figure, which leads to overpaying (and pay compression with existing staff), underpaying and losing the hire, or delaying the offer.

The answer is not to wait for surveys to catch up, but to use a method that does not depend on a title match. For background on how these roles are emerging, see our guides to AI in HR and AI fluency.

Tyria Taylor-Matthews, a Total Rewards and HR technology executive at Zeus, argues that the fix is to add a skills layer to job architecture rather than to abandon it:

What TalentUp Data Shows About AI Role Pay in 2026

Source: TalentUp job postings database, January–September 2026. Medians of advertised annual salaries, deduplicated.

We compared advertised salaries for emerging AI titles with their most common proxies in the European countries with enough salaried ads. Only cells with at least 40 salaried ads are shown.

Median advertised annual gross salary in EUR (UK and Poland converted from local currency), all seniority levels.

Four findings matter for anyone pricing a new role:

1. The AI premium is real but not universal. It is about a fifth in the UK and Spain, small in Poland and absent in Italy. A single “AI premium” applied across countries would be wrong in most of them.
2. Similar titles do not behave alike. In Spain, ML engineer ads were advertised 18.2% below software engineer ads, while AI engineer ads were 20.5% above. Titles are a weak signal of pay on their own, which is why Step 1 below matters.
3. The premium shrinks with seniority. In Poland, the market with the largest salaried sample, mid-level AI engineer ads were 15.4% above mid-level software engineers (€60,924, n = 87 vs €52,774, n = 332), but senior ads were only 1.5% above (€69,047, n = 155 vs €67,998, n = 481).
4. Ranges are not wider for new roles. In Poland, the median advertised range was 25.4% wide for AI engineers (n = 251), 23.6% for MLOps engineers (n = 42) and 35.0% for ML engineers (n = 124), compared with 32.8% for software engineers (n = 798) and 29.8% for data scientists (n = 114). Employers are not using wider ranges to cover their uncertainty.

Demand has grown fast. Across Spain, the UK, Germany, France, the Netherlands, Italy and Poland, AI-titled ads (AI, ML, LLM and MLOps engineers, prompt engineers, AI product managers and AI governance roles) went from 9.9 per 100 software engineer ads in 2024 to 18.3 in 2025 and 32.1 in 2026 (January–September each year). Measured against all postings, they rose from 12.9 to 35.0 per 10,000.

How to Price a New Role: Five Steps

Step 1: Define and Level the Role

Write a short role definition before you look at any salary data. It should state what decisions the person makes independently, which technical skills are required and at what depth, the scope of impact (one team, one product, the whole company) and what good output looks like after six months. Then assign a level in your existing job architecture.

The level matters as much as the title. Our Polish data shows the AI premium at mid level is ten times larger than at senior level, so a range for a “senior AI engineer” priced with mid-level evidence will overshoot.

Step 2: Choose Proxy Roles With Solid Data

Pick two or three established roles that share the core skills, complexity and scope of the new job, and that have a reliable sample in your country and at your level.

Use proxy data for the same country and level as the hire. Pay differences between countries are large: in our sample, software engineer ads were advertised at €74,500 in the UK and €50,000 in Italy. Our guide to international salary benchmarking explains how to keep the method consistent across borders.

Step 3: Decompose Skills and Weight the Proxies

Split the role into skill clusters and give each cluster a weight that reflects how much of the job it covers. Multiply each proxy’s median by its weight and add them up to get a blended anchor. The weights should come from the role definition in Step 1, not from the result you want.

Blend proxies only where each skill is a lasting part of the job. If 90% of the work is software engineering with a few AI tools added, use the software engineer benchmark and treat the AI component as a possible premium in Step 4.

Step 4: Test for a Scarcity Premium With Evidence

A premium above the blended anchor is justified only when the market shows it. Useful evidence includes:

• Title-level ads, where they exist: compare the median for the new title with the proxy at the same level and in the same country (as in the table above).
• Your own hiring results: offers declined for pay, time to fill compared with the proxy role, and the share of applicants who meet the technical bar.
• Competing offers that candidates can document.

Set the premium from that evidence and cap it so that the new range does not overtake more senior established roles without a reason you can explain. Our 2026 data suggests premiums between roughly 0% and 23% depending on country, and smaller at senior levels. If none of your evidence points to scarcity, use the anchor as it is.

Brandon Patterson, a US Total Rewards and compensation executive, makes the same case against a blanket premium for AI work:

Step 5: Build the Band and Set a Review Date

Turn the anchor into a band in the usual way: midpoint at the anchor, and a minimum and maximum around it. Our guide on how to create salary bands covers spread, overlap and midpoint progression; use range penetration to decide where each hire should sit.

Then set a review date of six months or less. Emerging-role pay moves faster than annual review cycles, and a short cycle is a cheaper way to manage uncertainty than a very wide range.

Worked Example: Pricing a Mid-Level AI Engineer in Spain

The proxy medians below are TalentUp 2026 job-ad data; the weights, premium and band spread are illustrative choices to show the method, not recommendations.

Role: mid-level AI engineer building LLM-based features into an existing product, in Spain. About 60% software engineering, 25% model evaluation and data work and 15% working with product and business teams.

1. Premium. In Spain, AI engineer ads were advertised 20.5% above software engineer ads (n = 46 vs 270). Two recent offers in this example were also lost on pay, so a 20% premium is applied: €44,250 × 1.20 ≈ €53,000 midpoint.
2. Band. With a ±15% spread around the midpoint, the band is €45,000–€61,000. The advertised hiring range for a new joiner covers the minimum to the midpoint: €45,000–€53,000.
3. Internal check. The band maximum stays below the median for senior software engineer ads in Spain (€65,000, n = 79), so the new role does not overtake senior engineers.
4. Cross-check against title data. The median for all Spanish AI engineer ads is €66,250, well above the €53,000 midpoint. The sample is small (n = 46) and mixes levels, so in this example the decision is to keep the proxy-based midpoint, record the gap and review in six months, or sooner if the next two offers are declined on pay.

The output is not just a number but a one-page record of the proxies, weights, premium, evidence and review date. That record is what makes the range defensible.

Pay Transparency: Documenting How a New Range Was Set

The EU Pay Transparency Directive (2023/970) had to be transposed by 7 June 2026, and its obligations apply through each country’s national law once transposed. Several of them affect how you price new roles:

Whether a market-driven premium counts as an objective justification will depend on national law and case law, so keep the evidence behind every premium and review it on schedule. For the full list of obligations, see our complete EU Pay Transparency Directive guide, and use the free pay transparency salary benchmark to check your ranges.

How to Explain the Range to Hiring Managers and Candidates

Hiring managers need the method, not just the number: the proxies, the weights, the evidence for any premium and the review date. When they understand that the range is evidence-based and will be revisited, they are less likely to approve offers outside it under candidate pressure. Show them where the new range sits next to existing roles, so they see the balance between market pay and internal equity.

Candidates for technical roles often know that data for their title is thin. A short explanation, such as “anchored to software engineer and data scientist pay at your level in Spain, with a review in six months”, shows that the offer is reasoned. If a candidate quotes a much higher figure, ask about the country, company size and level behind it; treat a confirmed gap as evidence for your next review, not as an automatic reason to break the band.

When to Move From Proxies to Direct Market Data

Proxies are a starting point. As a title becomes established, title-level data becomes more useful than a blend. Define the switch in advance, for example when you can find at least 40–50 data points for the exact title, level and country in a source you trust (the threshold used for the tables in this article is 40 ads).

At each review:

• Compare your proxy-based midpoint with title-level data where it exists. A gap of more than about 10% is a signal to revisit the weights, the premium or the level.
• Check hiring outcomes: offer acceptance, time to fill and pay-related declines.
• Reclassify the role into an existing job family when its responsibilities and skills have stabilised and match an established level.
• Correct premiums that are no longer justified for new hires, and plan how to handle existing employees without creating unexplained gaps.

Our guide on how to implement salary benchmarking explains how to build these reviews into your annual compensation cycle, and our methodology page explains how TalentUp collects and cleans salary data.

Sources

• TalentUp job postings database, January–September 2024, 2025 and 2026: advertised salaries for AI and proxy titles in the UK, Spain, Poland and Italy (about 3,600 deduplicated salaried ads) and title volumes in seven European countries (about 19,100 AI-titled ads in 2026). See the methodology box above.
• EUR-Lex, Directive (EU) 2023/970 on pay transparency