Ad tech companies are budgeting for ML engineers, but their AI projects are more likely to stall on the identity, consent and clean-room data underneath. The engineers who build that layer are harder to find.
About half of senior buy-side decision-makers in IAB’s State of Data 2026 report cite significant or critical concerns about data quality, accuracy, and legal and security risk as they bring AI into measurement. Fewer than 40% have a fix in place or planned. Those are data problems, not model problems. Yet when ad tech and media companies staff up for AI, most of the headcount goes to ML engineers. The harder hire sits one layer down: the data engineer who can move auction-scale data through consent rules, identity graphs and clean rooms without breaking any of them.
Key findings
About half of US brand and agency measurement leaders cite serious concerns about data quality, accuracy and legal and security risk in AI measurement, and fewer than 40% have or plan a fix, per IAB’s report.
Gartner predicts organizations will abandon 60% of AI projects through 2026 that aren’t supported by AI-ready data.
Twenty US states have comprehensive privacy laws in effect in 2026, with no federal law to unify them.
Data engineers start at a midpoint of $156,250, compared with $170,750 for AI/ML engineers.
IAB’s September 2026 clean room framework finds only 5 of 14 common commerce use cases typically need a clean room. The ones that do require substantial technical resources.
The model is the easy part
Gartner predicts that through 2026, organizations will abandon 60% of AI projects that lack AI-ready data, based on a survey of 1,200 data management leaders. In that survey, 63% of organizations said they either lacked the right data management practices for AI or weren’t sure whether they had them.
Ad tech has a sharper version of the problem. Bidding models, marketing mix models, incrementality tests and AI buying agents all run on joined data. Exposure logs come from one system and outcomes from another. Identity resolution links them, and every record carries a consent state. If any of those joins is wrong, the model is trained on a distorted picture and nobody finds out until the results don’t hold up. In the same IAB survey, 60% to 75% of buy-side users of advanced measurement said it falls short on rigor, timeliness, trust or efficiency. AI built on top of that measurement inherits those gaps. As we covered in our AdCP/AAMP piece, an agent that buys media is only as good as the audience and identity data it runs on.
The data layer got harder to build
Two things have raised the bar for that data work since most ad tech data teams were staffed.
The first is regulation. Twenty states now have comprehensive privacy laws on the books, with new laws in Indiana, Kentucky and Rhode Island taking effect at the start of this year. Each has its own thresholds and opt-out rules, and a growing number require businesses to honor browser-level opt-out signals such as Global Privacy Control. For a data engineer, that means consent has to travel with each record through every downstream system, including the ones built before anyone asked.
The second is clean rooms. IAB’s Data Clean Rooms in Commerce framework warns that clean room collaborations can require substantial spending on platform fees, data science and technical resources, and ongoing operations. It also states that a clean room does not by itself establish a legal basis for processing or sharing data. FTC technology staff made a similar point in 2024. The framework names interoperability across clean room providers and identity frameworks as the next big challenge. Cross-platform reach measurement, for example, only works as well as the parties’ ability to resolve audiences through common identifiers. Every new specification from IAB Tech Lab, from minimum dataset sizes to conversion-data formats, is more engineering work for someone.
A role most job descriptions don’t describe
A generic data engineer posting asks for SQL, Python and a cloud warehouse. The ad tech version needs much more:
Pipelines that handle log-level auction data at volume.
Identity work: hashed emails, ID graphs and match rates.
Consent propagation, so an opt-out recorded in one system reaches every system that uses the data.
Clean room operations: preparing datasets, enforcing aggregation thresholds and turning outputs into something a measurement team can use.
Clean, timely inputs for MMM and incrementality testing.
Few engineers have all of those skills, and fewer have built them under 20 state privacy regimes. A requisition posted as "data engineer" attracts the generic profile. The candidates who can do this work tend to recognize the systems by name, so the job description should name them.
Retraining data engineers into ML can empty the bench
Some companies close their ML gap by retraining adjacent talent. Talent Observer reported that data and backend engineers can move into ML work within six to twelve months. That approach adds ML headcount. In ad tech, though, it often moves the one person who understands the identity pipeline onto model work, and the data layer loses its owner just as the new models start depending on it.
Pay pushes engineers in the same direction. In Robert Half’s 2026 Salary Guide, data engineers’ starting salaries run from $127,000 to $180,750 with a midpoint of $156,250. AI/ML engineers have a $170,750 midpoint, and data scientists $153,750. For a data engineer, a move into ML usually comes with a raise. Employers are already paying above plan to hold onto specialists: looking ahead to 2027, 65% of technology leaders said they planned to offer salaries above their planned range, and data architecture and infrastructure was among the skills commanding higher pay.
How to staff the layer underneath
Teams that get AI into production in ad tech tend to staff in this order.
Data before models. The first AI hire on an ad tech team is often a data engineer who owns identity and consent, not another modeler.
Specific job descriptions. Name the consent framework, the clean room platform, the identity vendors and the log volumes the hire will handle.
Privacy counsel from the start. A clean room is a technical control, not a compliance answer, and engineers need someone who can say what the law requires.
Contracts for builds that end. Bring in contract specialists for work with an end date, such as a clean room integration or an onboarding pipeline, and give each one a full-time owner who keeps the knowledge. We describe this model further here.
A wider search. AI and data skills are now the hardest to hire worldwide, and shortage rates vary sharply by country. Teams that search beyond New York fill these roles faster.
An early start. Technical roles take about 10 weeks to fill. The data layer has to be in place before the models can use it.
The models will keep getting better and cheaper to access. The data they run on won’t fix itself, and in ad tech it now has to satisfy 20 state laws, several identity frameworks and every partner’s clean room. Companies that staff that layer first will have AI projects that hold up in production.
Frequently asked questions
What does a data engineer do in ad tech? They build and run the pipelines that feed bidding, measurement and AI systems. That includes processing log-level auction data, resolving identity across partners, carrying consent and opt-out signals through every system, and preparing data for clean rooms and measurement models.
Why do AI projects in advertising stall? Usually because of data, not models. Gartner predicts 60% of AI projects without AI-ready data will be abandoned through 2026. In IAB’s State of Data 2026, about half of buy-side leaders cite serious concerns about data quality, accuracy, and legal and security risk.
Does every ad tech company need a data clean room? No. IAB’s 2026 framework finds that only 5 of 14 common commerce use cases typically require one. Those are mostly cases where two companies need to match customer records without sharing them. Many measurement tasks can run on existing tools when the data sits with one company.
How much do data engineers cost compared with ML engineers? Robert Half’s 2026 Salary Guide puts the data engineer starting midpoint at $156,250, compared with $170,750 for AI/ML engineers. Engineers with ad tech identity and privacy experience typically command more than the generic midpoint.
Should ad tech companies hire data engineers or ML engineers first? In most cases, data engineers. A model trained on poorly joined or non-compliant data fails in production, and the data layer takes longer to build than the model.
What should an ad tech data engineer job description include? The specific systems: consent frameworks, clean room platforms, identity providers, data volumes and the state privacy laws the pipelines must respect. Generic postings attract generic applicants.
How do US state privacy laws affect ad tech data engineering? Twenty states have comprehensive privacy laws in effect in 2026, each with its own thresholds and opt-out rules. Pipelines have to record and honor consent record by record across every system that touches the data.
Can we retrain data engineers into ML roles? Yes, and it can take six to twelve months. But moving the engineer who owns identity or consent pipelines leaves that layer without an owner, so plan a backfill first.
Should clean room and identity work be contract or full-time? Use full-time engineers for the people who own identity, consent and the data platform. Use contract specialists for builds with an end date, such as a clean room integration, paired with a full-time owner who keeps the knowledge.
