The UA skills that will still matter

A UA skill will still matter in five years if it sits on the human side of a decision you can't trust the platform to make. Not can't make. Can't trust to make, because its incentive points the other way. Everything else, including a great deal of what practitioners are currently proud of, is on a path to being a platform feature.

That test is stricter than the usual framing, which sorts skills into technical versus strategic and declares strategy safe. Automation is eating plenty of strategic work and leaving some technical work alone. The dividing line is not how clever the work is. It's who benefits when someone does it badly.

The three tests

The first test is incentive conflict. Ask whether the platform, if it did this job for you, would have a reason to do it in its own favour. Setting the optimisation target is the clearest case. A network asked to choose the event it optimises to will pick the one it can hit cheaply. Deciding what a valuable player is, and translating that into a target the model cannot game, will stay human because the platform is structurally the wrong party to do it.

The second test is missing context. Ask whether the decision depends on information the platform cannot see. The margin on a web-shop customer versus an App Store customer under the tiered EU commission changes Apple has scheduled for October is invisible to any network. So is the roadmap, so is the liveops calendar and the cash position, and so is the reason the last update disappointed. Skills that combine platform data with studio-only context are safe. Skills that only manipulate platform data are not.

The third test is accountability. Ask whether a human has to sign for the outcome. The platform does not present to the board, does not explain a missed quarter and does not choose between two titles when the budget is cut. Whoever does needs to understand the machine well enough to defend or overrule it.

A skill that passes two of the three is durable. One is a warning. Zero means the skill is already a feature in someone's roadmap.

The skills that pass

Target design. Choosing the event the platforms optimise to, along with its window and value definition, is the single most consequential act in modern UA, and it fails all three tests for the platform and passes all three for the human. It requires understanding both the game economy and how a delivery model behaves when handed a proxy. Very few people are good at it, and fewer still are getting any training.

Independent measurement. Designing and running holdouts or geo tests, plus any incrementality read that doesn't route through a party with a stake in the answer. UA Ledger's recent piece on the consolidation endgame argued that the measurement layer is acquiring shareholders among the networks. The skill of producing a number nobody else owns becomes more valuable as that happens, not less.

Creative judgement grounded in player evidence. Not taste alone, which is unfalsifiable, but the ability to look at a player observation and a creative concept and say why one should move behaviour, then check. The generation is now machine work. The hypothesis is not, and the platform cannot form one because it does not know your players outside the ad.

Economic translation. Turning platform outputs into margin statements and cash statements and payback statements a finance team will accept, and turning finance constraints back into targets a platform can act on. This passes the context and accountability tests comfortably, and it is the skill most often missing in teams that grew up inside dashboards.

Vendor and platform negotiation with a technical floor. Knowing what a stack is actually doing with your data, what a rate card conceals and what a self-attributing network's numbers do and do not prove. Automation raises the stakes here because the buyer touches fewer levers and so must be surer of the ones that remain.

The skills that look durable and are not

This is where honest disagreement lives, so it is worth being specific.

Channel expertise. Knowing the quirks of a particular network's ad manager was a real advantage for a decade. It fails all three tests: the platform has every incentive to make its own interface unnecessary, it sees more of its own context than you do, and nobody is accountable for a bid. The knowledge decays with each product release.

Creative production craft as a standalone skill. Editing and motion and asset production remain valuable inside a studio, but as UA skills they fail the context test: the tools now have most of it. The exception is craft attached to judgement, which passes.

Reporting and dashboard building. AppsFlyer's January report with Unity and Newzoo already described AI assistants used for UA reporting among top advertisers. Assembling the report is going. Deciding what belongs in it, and what it means, is not.

The trade-off most career advice misses is that the durable skills are slower to acquire and harder to demonstrate in an interview. A buyer can show a screenshot of a campaign that scaled. A target designer has to explain a counterfactual. Studios that hire on visible artefacts will keep hiring for the decaying skills.

A worked illustration

Consider a UA manager with five years in, most of it running channels. A skills audit against the three tests might find channel operation and reporting at zero and creative iteration at one; target design barely registers, and neither incrementality nor economic translation does, because the org never let the role near them.

The practical plan is not a course. It is a reallocation of the job. Take one title's optimisation target and own its definition end to end, including the argument with finance about what value means. Run one incrementality test a quarter, with a written pre-registration of what result would change a decision, and write the monthly economics note for one game in language the CFO uses. Take the feedback. Eighteen months of that produces someone with three durable skills and a portfolio of decisions rather than campaigns.

The org has to allow it, which is the real constraint. Most studios still measure UA people on the skills that are disappearing, because those are the ones the dashboards make visible. The person who wants to still matter in 2030 will often have to ask for the harder job before anyone thinks to offer it.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• The UA lead as translator, not buyer
• Shooter UA on mobile: the attention problem