Setting pay ranges for emerging roles is one of the most frustrating problems in compensation today. A hiring manager asks what a Prompt Engineer should earn. The job did not exist two years ago, the major salary surveys do not have a reliable sample size for it yet, and the candidates who apply quote figures all over the map. HR needs to give a number, and the number needs to be defensible to leadership, fair to the candidate, and consistent with every other offer the company makes. This guide walks through a five-step methodology for building salary ranges for roles with no market data history, using skills decomposition, proxy benchmarking, and a structured review cadence to produce ranges that hold up under scrutiny.

Why Standard Benchmarking Breaks Down for New Roles

Traditional salary benchmarking works by matching a job title to a survey dataset, reading the 25th, 50th, and 75th percentile figures, and building a band around them. That process assumes the role has existed long enough to accumulate a meaningful sample in the survey and that job titles are being used consistently across the market. For established roles, both assumptions hold. For Prompt Engineers, AI Trainers, Automation Architects, and Responsible AI Specialists, neither does.

The result is that HR teams either delay making an offer until they can find data that does not exist yet, overpay relative to internal benchmarks because the candidate asked for a high figure and no one had a reference point to push back, or underpay and lose the candidate to a company that has already built a methodology for pricing these roles. All three outcomes are avoidable. The solution is not to wait for surveys to catch up. It is to use a structured approach that does not depend on a title match.

Step 1: Define the Role Before You Benchmark Anything

The first error most HR teams make when pricing a new role is searching for a salary before they have defined what the role actually does. A Prompt Engineer at one company writes and tests prompts for a customer service chatbot. At another, the same title covers someone building multi-step agentic workflows that integrate with core business systems. These are not the same role, and they should not sit in the same pay range.

Before opening any salary platform, write a one-paragraph role definition that answers: what decisions does this person make independently, what technical skills are required and at what depth, what is the scope of impact (a single team, a product, the whole company), and what does a good output look like in the first six months? The answers to these questions determine which proxy roles are relevant and where in a range this specific version of the job should sit. Skipping this step produces a range that is technically defensible but practically wrong for the actual hire you are making.

Step 2: Identify Proxy Roles and Build an Anchor Range

Once the role is defined, the next step is to identify two or three established roles with meaningful market data that overlap significantly in required skills, complexity, and scope. These become your proxy roles, and their salary data becomes your anchor range.

For most AI-adjacent roles that emerged between 2023 and 2025, the most useful proxies are roles the market has priced for longer: Machine Learning Engineer, Data Scientist, NLP Engineer, and Senior Software Engineer. The table below shows current median gross annual base salary data for Machine Learning Engineers in Spain and Germany by seniority, drawn from the TalentUp Salary Platform (data retrieved 17 July 2026). These figures serve as anchors for emerging AI roles that share the same technical depth and scope of responsibility.

The gap between countries is as important as the absolute figures. A senior ML Engineer in Germany earns 60% more at median than the same level in Spain. Any emerging AI role priced using German proxy data should not be applied to a Spanish hire without recalibration. Country context is non-negotiable even when you are working with proxy benchmarks.

For a role like AI Trainer, where the work is closer to Data Scientist territory (labelling, evaluation frameworks, RLHF pipelines), Data Scientist benchmarks are more relevant. In France, TalentUp platform data (retrieved 17 July 2026) shows Data Scientist medians at €30,300 for junior, €36,500 for mid-level, and €43,300 for senior. Using a weighted average of two or three proxies, adjusted for the skill overlap percentage, gives you a defensible anchor that is grounded in real market data rather than a single candidate’s expectation.

Step 3: Decompose Skills and Apply a Weighting

Proxy roles give you an anchor range, but the emerging role rarely maps perfectly onto any single proxy. The methodology used by compensation teams with the most defensible ranges for new roles is skills decomposition: breaking the role down into its component skill clusters and weighting each cluster against a proxy that has market data.

A Prompt Engineer role, for example, might decompose as follows: 40% ML fundamentals and LLM architecture knowledge (proxy: ML Engineer), 30% software engineering and API integration (proxy: Software Engineer), and 30% evaluation design and iterative testing methodology (proxy: QA Engineer or Data Scientist). Applying those weights to the relevant seniority-level medians for each proxy in the target country produces a blended anchor salary. From there, you build the band in the same way you would for any established role: minus 15 to 20% at the minimum, plus 15 to 20% at the maximum, with the midpoint at or near the blended anchor.

This approach integrates naturally with the broader trend toward skills-based pay structures, where the market value of a skill cluster matters more than the title on a job description. For emerging roles, skills-based decomposition is not just philosophically aligned with modern compensation thinking. It is the only methodology that produces a number when title-based data does not exist.

Step 4: Apply a Scarcity Adjustment

Proxy benchmarks reflect what the market pays for roles with an established supply of candidates. Emerging roles frequently have a supply problem: the number of people who can credibly perform the role is smaller than the number of companies trying to hire for it. That imbalance justifies a scarcity adjustment above the proxy anchor, but only when supply data supports it.

The scarcity adjustment should be evidence-based, not guesswork. Signals that support a premium above the proxy anchor include: the average time to fill for the role exceeds 90 days in your market, fewer than 30% of applicants meet the minimum technical bar, and competing offers from the same candidate are consistently 20% above your proxy anchor. If none of these signals are present, the proxy anchor is likely sufficient and adding a scarcity premium will simply create internal equity problems when the market normalises.

A reasonable scarcity premium for genuinely scarce emerging roles sits between 10 and 25% above the proxy anchor at midpoint. Anything above 25% creates significant internal equity risk: you are now paying a Prompt Engineer more than a Senior ML Engineer with five more years of experience, which will become a problem the moment the team starts talking about pay. Before applying a scarcity premium, confirm it against the existing salary band structure to understand where the new role sits relative to established roles and whether any overlap creates anomalies.

Step 5: Build a Mandatory Review Date Into the Range

Salary ranges for established roles are typically reviewed annually or biennially. For emerging roles, six months is the maximum defensible review cadence. The market for AI-adjacent roles has been moving faster than annual survey cycles can track, which means a range built in January using the best available proxy data may be 15 to 20% off market by September of the same year.

When you publish a pay range for a new role internally, document it with three things: the proxy roles used, the weighting applied, and the review date. This creates an audit trail that protects HR when the range is challenged, and it signals to the business that the range is a living document rather than a permanent fixture. It also creates a natural moment to reclassify the role if the market has evolved enough that the original proxy roles are no longer the best match.

The EU Pay Transparency Directive adds urgency to this documentation discipline. Employers with 150 or more employees will need to report gender pay gaps by worker category from June 2027. An emerging role with an undocumented pay range will create a category that cannot be defended if a gap appears. Building the methodology documentation into your range-setting process from day one means you have an audit-ready record when reporting season arrives rather than a gap you are reconstructing under pressure.

What to Tell Hiring Managers and Candidates

The biggest operational challenge of pricing emerging roles is not building the range. It is communicating it credibly to hiring managers who will question why you are using proxy data and to candidates who may have been offered more elsewhere.

For hiring managers, the message is: the range is built on the closest available market references, weighted by skill overlap, and will be reviewed on a fixed date. It is not arbitrary, and it is not a ceiling: it is a starting point that will be adjusted as the market matures. Giving managers the proxy roles and the methodology, rather than just the final range, builds credibility and prevents them from approving offers outside the band based on candidate pressure alone.

For candidates, transparency about the methodology is increasingly effective, particularly for technical roles where candidates understand that market data for their title is limited. Framing the offer as “anchored to ML Engineer benchmarks for your level in this market, with a built-in review at six months” positions the company as sophisticated rather than evasive, and it preempts the objection that you pulled the number from thin air.

This transparency approach also aligns with the direction that pay practices are moving under the EU Pay Transparency Directive, which requires employers to be able to explain the objective criteria behind any pay decision. Building and communicating a documented methodology now is not just good compensation practice. It is preparation for a legal environment where the ability to explain a pay decision on demand will be mandatory.

How to Use Real Market Data as the Range Matures

Proxy benchmarks are a starting position, not a permanent solution. As a role accumulates market data, the transition from proxy-based to title-based benchmarking should happen at a defined threshold: typically when three or more credible salary surveys carry a sample size of at least 50 data points for the exact title in your geography. Below that threshold, proxy data is more reliable than a thin survey sample that can be distorted by a handful of outlier companies.

Track the salary data available for your emerging roles at each review cycle. The TalentUp platform can be used to query adjacent role data that forms the basis of your proxy calculation, and to monitor when the role itself begins to appear with sufficient sample depth. When it does, run a comparison between your current proxy-based range and the emerging title-specific data. If they diverge by more than 10%, that is a signal that your proxy weighting needs to be adjusted, or that the role has evolved enough to be reclassified. The process of implementing salary benchmarking systematically, including how to structure that transition from proxy to direct data, is covered in detail in the guide on how to implement salary benchmarking in your organisation.

Frequently Asked Questions

What proxy roles work best for Prompt Engineers?

Machine Learning Engineer and Senior Software Engineer are the most reliable proxies, weighted by the proportion of the Prompt Engineer role that involves LLM architecture knowledge versus general software development. If the role is more evaluation and testing focused, Data Scientist is a stronger proxy. Use two or three proxies weighted by skill overlap rather than relying on a single match.

How do I handle candidates who quote salaries far above our proxy range?

Ask for the context behind the figure: what country, what company size, and what seniority level is it based on? Outlier figures for scarce roles often reflect either a very large company paying above market to move fast, or a candidate in a different geography. If the figure is genuinely above your proxy range after accounting for those variables, treat it as a signal to review your scarcity adjustment, not as an automatic reason to match it without recalibrating your internal structure.

Can I use the same proxy methodology for non-technical emerging roles?

Yes. The methodology applies to any role where market data is thin or inconsistent. An AI Ethics Officer, for example, can be anchored to a weighted blend of Legal Counsel, Compliance Manager, and Senior Policy Analyst benchmarks, weighted by how much of the role involves legal interpretation versus stakeholder communication versus technical policy design. The skill decomposition step is the same regardless of whether the role is technical or functional.

How do I prevent pay equity problems when emerging role salaries are set above the proxy anchor?

Document every scarcity premium decision with the evidence that justified it (time to fill, offer competition data, candidate feedback) and set a review date. When the market normalises and the scarcity premium is no longer justified, the range should be adjusted downward for new hires while existing employees are managed within a corrective window. Leaving scarcity premiums in place indefinitely creates structural inequity between employees hired during the shortage and those hired after it resolved.

When should I reclassify an emerging role into an existing job family?

When the role has stabilised enough that its core responsibilities, required skills, and scope of impact consistently match an established job family at a specific level, reclassification is appropriate. The trigger is usually when the role has been filled more than five times in the organisation and the variation in what people in it actually do has narrowed significantly. At that point, maintaining a separate emerging-role range adds administrative complexity without adding accuracy.

Sources