How to price a new AI role
The market data for new AI roles is thin and it moves in months, not years. Make a careful estimate now, say it is an estimate and look again in six months.
Short answer: To price a new AI role, define the work before the title. Blend the two or three established jobs it most resembles and add a premium only where your own hiring market shows one. Treat the result as provisional. Pay for these roles can swing by a fifth within a year, so set a review date six months out and check the price against real offers.
Three numbers for one title
Picture a 190-person software company in Nashville that sells claims tools to hospital billing teams. Its leaders decide to build AI into the product. They plan to hire two AI engineers this quarter. They also want an AI enablement lead to bring new tools into the company's own support and finance teams.
Three numbers arrive for the AI engineer role in the same week. The hiring manager looks up the title on a free pay site and finds a median well above anything the company pays its senior software engineers. The recruiter brings a figure from news stories about offers at large AI companies, which is higher again. The company's salary survey has no job called AI engineer at all, so the HR lead suggests waiting for next year's edition.
Meanwhile one of the company's own senior engineers has spent the past year building the first AI prototype on nights and weekends. That person is paid on the senior engineer band and has not been asked about the new roles.
The first offer goes out at the hiring manager's number, because a candidate has another offer. The second hire, five months later, comes in lower, because the market has cooled for that kind of work. Now two people in the same job are far apart, the first is above any band the company could defend and the engineer who built the prototype has seen a job ad that pays more for the work they already do. The enablement lead role is still unpriced, because nobody can agree what it is.
This is a composite of a common situation at companies this size, not a client. The details are illustrative. The pattern is common because a new AI role usually gets its first price under hiring pressure, from whatever number is nearest to hand.
Why pricing a new AI role is hard
Our guide on pricing a role with no market match covers the general method: Blend the closest matches openly, weight them by how much of the work each covers and record the guess as a guess. New AI roles add three problems on top.
The first is that the title does not describe a job. "AI engineer" can mean someone who builds product features on top of a model another company trained. It can also mean someone who trains and tunes models and owns the data behind them. Those are different jobs with different markets. The same is true of titles like AI lead, AI product manager and head of AI.
The second is that the data is thin and lopsided. Few salary surveys carry these jobs yet and the cells that exist often rest on a small number of companies. Self-reported pay sites fill the gap, but their high end is shaped by a handful of very large employers. One such site said its rebound in AI engineer pay in 2025 was driven by more high packages reported at two of them [3].
The third is speed. On the same site, the median reported pay for AI engineers in the United States peaked at $295,000 in March 2024 and fell to about $228,500 by January 2025, a drop of more than a fifth in ten months [3]. The premium over other software engineers fell at the entry level over the same period and rose at the staff level [3]. A price set from last spring's data can be wrong in either direction by autumn.
The headline numbers do not settle it either. A 2026 study of more than a billion job ads put the average wage premium for AI skills at 62%, up from 57% a year earlier [1]. But a February 2026 survey of more than 4,000 companies found employers paying an average of about 10% above market for AI roles, with most between 5% and 15%. In the same survey, 64% of compensation professionals said they were unsure what premium to offer [2]. The two figures measure different things. Our note on pricing AI skills in existing jobs explains why a job ad average is not your premium. The point here is simpler: Most companies are making this call without a reliable number and they know it.
Three roles that share one title
Before anyone looks up a number, the company has to decide which job it is hiring for. Most new AI roles at companies of 50 to 500 people are one of three kinds, or a mix. This table shows how they differ and how thin the data is for each. It is illustrative, not a matching guide.
| Kind of role | What the work mostly is | Established jobs it most resembles | How thin the data is |
|---|---|---|---|
| Applied AI engineer | Building product features on models from outside providers: Connecting them to the product, testing output quality and controlling cost | Software engineer or data engineer at the same level | Moderate. Often reported inside software engineering |
| Machine learning engineer | Training, tuning and running models, with the data pipelines and quality checks behind them | Machine learning engineer or data scientist, where surveys carry them | Thinner at small companies. The high end is shaped by large tech employers |
| AI enablement lead | Bringing AI tools into business teams: Choosing vendors, changing workflows, training staff and managing risk | Business systems analyst, process improvement lead or internal tools product manager | Very thin. Few surveys carry it under any title |
Many companies need the first kind and advertise for the second. That one mistake can put the price tens of thousands of dollars too high before any other judgment is made.
Five traps in pricing new AI roles
Each of these starts from a real pressure: A candidate with another offer, a board that wants AI on the roadmap or a survey with nothing to say. The damage comes from treating a fast, thin market as if it were a settled one.
"Search the title on a pay site"
What it looks like: Someone types "AI engineer" into a free pay site, takes the median and uses it as the target for the role.
Why smart teams fall into it: It is fast and it looks like market data. When the survey is silent, any number from outside the company feels better than none.
What it costs: The number blends every kind of work that shares the title, at every kind of employer that reports it. If the role is an applied AI engineer at a mid-size software company, a median pulled up by model trainers at the largest tech firms will overprice it. If the role is a model trainer, a median diluted by people who mostly connect tools can underprice it. Either way the price follows the label, not the work. Free pay data is useful for direction, as our note on free salary sites explains. It is not a price for a job the company has not yet defined.
"Match the offers in the news"
What it looks like: The recruiter or a board member points to reported pay at the best-known AI companies and argues that the company must get close to compete for anyone good.
Why smart teams fall into it: The stories are real and they are everywhere. Losing a strong candidate is painful and visible. Nobody wants to be the person who priced the company out of AI.
What it costs: Those packages come from a few employers competing for a small group of researchers, often with large equity grants that a private company of 190 people cannot match in kind. Most of the people a mid-size company hires will never get those offers. Pricing against them overpays for the work the company actually needs. It also pulls up every engineering band near it, because the people next to the new role will notice. The market that matters is the set of companies the business really hires from and loses people to. That market should be written down before the first offer, not argued out during it.
"Wait for the survey to catch up"
What it looks like: The salary survey has no match, so the role is left unpriced until a future edition adds it. In the meantime offers are approved one at a time.
Why smart teams fall into it: It feels disciplined. Nobody wants to put a number into the structure that the data cannot support.
What it costs: Waiting does not stop the company from pricing the role. It just means the role gets priced by each negotiation, with no record of why. Surveys add a job once enough participating companies report it and official job classification moves slower still. The US occupation list in use today dates from 2018 and its next version is planned for use from 2028 [5]. A provisional price now, built from the closest established jobs and labeled as provisional, is more defensible than a run of one-off offers. Our guide on pricing a role with no market match covers why the first candidate's ask is not a market.
"Pay the new hire the premium and leave the builder where they are"
What it looks like: The company hires its first AI engineer from outside with a premium. The employee who built the first prototype, or who has quietly taken on most of the AI work in their team, stays on their old band with their old title.
Why smart teams fall into it: The outside hire is the one with a competing offer. The internal person has not asked for anything. Changing their job feels like a separate project for later.
What it costs: The internal person is often the most valuable one, because they know the product and the data. They also see the job ads. In a 2026 survey of Indian employers and employees, 66% of employers said they offered salary premiums for AI-skilled hires, while 54% of employees in AI-exposed roles said their own pay had stayed flat or fallen over the past year [4]. A gap between two people doing the same work also needs a reason that holds up. In the United States, market value is accepted as a reason for a pay difference only where the employer can show it assessed the value of the job and the person's qualifications [6]. In the EU, pay structures must let an employer compare the value of work on skills, effort, responsibility and working conditions and the criteria used to set pay must be objective, gender-neutral and accessible to staff [7]. Our guide on writing down your pay criteria covers that. When a new AI role is created, the people already doing that work should be assessed against it in the same pass.
"Lock this year's premium into base pay for good"
What it looks like: The company pays a premium to land its first AI hires and draws a new band around those offers. A year later the band is moved forward with every other band, premium included.
Why smart teams fall into it: A band looks permanent and tidy. A separate premium feels like an admission that the company is not sure.
What it costs: The company is not sure and neither is anyone else. A premium set at a peak can sit in the band long after the market has cooled, as the 2024 to 2025 swing in AI engineer pay showed [3]. A premium set too low can leave the band behind a market that is still rising. Once the premium is inside the band, nobody can see it, so nobody checks it. A visible premium, shown on top of the band for the closest established job with the evidence behind it, can be confirmed, reduced or folded into the band at review. Our note on when to update salary bands covers the yearly check. For new AI roles, a shorter first review makes sense.
What a sound price for a new AI role looks like
A good price is not the highest or the lowest one on the table. It is one the company can explain to the new hire and to the people next to them. It can also change without embarrassment when the evidence changes.
The work is defined first. The role has a short description of what the person will mostly do and at what level, written before anyone looks up a number.
The closest established jobs carry the price. Two or three jobs that each cover a real part of the work set the base, at the level the role sits in the company's own structure.
The comparison market is the company's own. The price is checked against the companies the business really hires from and loses people to, not against the best-known AI employers.
Any premium is separate and shown. If the company's own hiring evidence shows a premium, it sits on top of the base with a note of what it rests on: Declined offers, competing offers or a thin survey cell.
Internal people are assessed in the same pass. Anyone already doing the work is compared with the new role before the first offer goes out.
A review date is set. The price is labeled provisional, with a date about six months out and a short list of what will be checked: Offers accepted and declined, new survey data and any departures.
The reasoning is written down. One page records the matches, the weights, the comparison market and the premium. The person who reviews it in six months may not be the person who set it.
The calls only you can make
The method can be learned. The choices inside it belong to the company.
Build or buy? Some companies hire AI specialists from outside. Others train people who already know the product and add one experienced hire to guide them. The second route usually costs less in pay and more in time. Either can be right. The answer changes what the company is pricing.
Which market do you compete in? A software company in Nashville may lose engineers to remote employers on the coasts or to local health systems. Which of those it chooses to compete with is a strategy call, not a data question. Our guide on which market percentile to target covers the same choice for the whole structure.
How much risk will you take? Pricing high reduces the risk of losing a hire and raises the risk of a band the company regrets. Pricing low does the reverse. A visible premium with a review date narrows both risks. It does not remove them.
Cash, equity or a bonus? Some companies pay part of the premium as a hiring bonus or an equity grant, so base pay stays near the established jobs. That keeps the band honest. It also makes the offer harder to compare and may not hold someone past the first year.
Where does the role sit in your levels? A new title tempts people to place the role a level higher than the work requires. The level should follow the scope of the work, the same as any other job.
The data, the law and the numbers
- Job ad premiums are high and rising. PwC's 2026 Global AI Jobs Barometer, based on more than one billion job ads in 27 countries and territories, put the average wage premium for AI skills at 62%, up from 57% the year before. The premium ranged from 118% in some sectors to 16% in government and public sector work. Jobs requiring specific AI skills grew 69% against 9% for the overall jobs market [1]. These are averages across job ads, not what any one employer pays.
- What employers say they pay is much lower. A survey of more than 4,000 companies in 133 countries, fielded in February 2026, found employers paying an average of about 10% above market for AI roles, from 5% at the low end to 15% at the high end. 64% of compensation professionals said they were unsure what premium to offer [2].
- Pay for AI engineers swings. On a self-reported pay site, the median pay for AI engineers in the United States peaked at $295,000 in March 2024 and fell to about $228,500 in January 2025. The premium over other software engineers was 6.2% at entry level in 2025 (10.7% in 2024) and 18.7% at staff level (15.8% in 2024) [3]. Self-reported data skews toward large tech employers.
- New hires and existing staff are treated differently. In India, a 2026 survey of 1,267 employers and 2,541 employees reported that 66% of employers offered salary premiums for AI-skilled professionals, while 54% of employees in AI-exposed roles said their pay had stayed flat or fallen over the past year [4]. This figure comes from a news report of the survey.
- Official classification moves slowly. The US Standard Occupational Classification in use today dates from 2018. The revision now under way is planned to be final in 2027 and used from reference year 2028 [5].
- United States, equal pay. Equal pay guidance accepts market value as a factor other than sex only where the employer shows it assessed the marketplace value of the job and the person's qualifications and that the gap is not based on sex [6].
- European Union. Pay structures must let employers assess equal value on skills, effort, responsibility and working conditions (Article 4(4)). The criteria used to set pay, pay levels and pay progression must be easily accessible to staff and must be objective and gender-neutral (Article 6) [7].
This is general information, not legal advice. Equal pay and pay transparency rules differ by state and country and change over time. Have an employment lawyer review how you document pay decisions.
Our view
We think a new AI role should be priced like any other job with thin data, with two extra safeguards. Define the work before the title, build the price from the established jobs it most resembles and check it against the companies you really compete with. Then show any premium on its own, so it can be seen and questioned.
We would set the first review six months out, not twelve. That is long enough to see a few real offers and a new round of survey data. It is short enough to catch a market that has turned before the price hardens into the structure.
And we would never price the new hire without looking at the people already doing the work. The fastest way to lose the person who knows your product is to hire someone beside them at a premium for the job they built.
Where we help
Our market pricing work prices new and hard-to-match roles, including new AI roles: Defining the work, choosing the closest established matches, setting the comparison market and recording any premium with a date to review it. See market pricing for how a project is scoped and priced. If a conversation would help first, Book a free call.
Sources
- PwC, 2026 Global AI Jobs Barometer, press release (15 June 2026), checked 2026-10-28
- Korn Ferry, Only modest pay bumps for AI skills (16 March 2026; survey of more than 4,000 companies in 133 countries, February 2026), checked 2026-10-28
- Levels.fyi, AI Engineer Compensation Trends Q3 2025 (17 July 2025; self-reported data), checked 2026-10-28
- People Matters, New AI hires earn more while existing employees see little pay growth (16 July 2026; reports Indeed's AI at Work Tracker 2026, India), checked 2026-10-28
- US Bureau of Labor Statistics, 2028 SOC revision (page last modified 30 September 2026), checked 2026-10-26
- US EEOC, Compliance Manual Section 10: Compensation Discrimination, 10-IV.F.2.g (market factors), checked 2026-10-26
- Directive (EU) 2023/970 on pay transparency, Articles 4(4) and 6 (EUR-Lex), checked 2026-10-28