AI user acquisition agents in 2026: what they change for indie UA budgets

Direct answer — Should a small or mid-size mobile studio spend its next euro on AI user acquisition tooling or on new distribution channels? AI user acquisition agents such as Kohort’s Ktrl make paid UA spend less wasteful: they predict earlier which campaigns will pay back, feed those predictions into ad-network bidding and flag failing campaigns faster. They do not lower the price of the auction, the store’s commission or the size of the audience paid ads can reach. If paid UA already works for your title and you spend enough to train the models, AI tooling is the cheaper next step; if the paid channel is saturated or structurally unprofitable for your game, the next euro belongs in distribution.

In May 2026 Kohort closed a $7 million Series A led by The Raine Group to build what it calls AI user acquisition agents for mobile game studios. The raise put a name on a category growth leads had been assembling piece by piece: predictive LTV models, automated bid signals, AI-generated creatives and an agent layer acting on all three. The pitch is simple: let the machine run the auction better.

This article answers a narrower question than whether paid UA still works. We made that case in why distribution has to go beyond paid UA. Here the question is what an AI UA agent actually changes in a studio’s paid UA budget, what it cannot change, and how to decide between funding that tooling and funding an additional distribution channel.

What does an AI UA agent actually do in 2026?

An AI UA agent today does two jobs: it predicts the long-term value of the users each campaign buys, and it turns those predictions into actions — bid targets, budget moves, creative refreshes — faster than a human team would. The vendors split along those two jobs.

The decision layer: Kohort. According to the funding announcement, Kohort’s agent suite covers campaign optimisation, deep research and automated reporting, and its flagship product, Ktrl, generates network-specific bidding strategies and targets across return-on-ad-spend, cost-per-install and cost-per-engagement campaigns while integrating directly with ad networks. Kohort’s own site describes the loop: Ktrl trains cohort-based predictive models on the studio’s UA data, issues daily scale, hold and cut recommendations per campaign, and sends predictive ROAS values back to AppLovin, Meta, Google, Unity and TikTok so their bidders optimise for forecast value rather than early conversions. CEO Dan Marcus framed it this way: “The best UA teams operate more like high-frequency traders than marketers.”

The headline numbers are Kohort’s own. The company said its predictive models are trained on approximately $6 billion in historical user acquisition spend data across hundreds of mobile games and produce daily campaign-specific predictions with approximately 95% accuracy. Its website claims 94%+ accuracy on day-365 LTV forecasts. Those are two different measurements, neither page publishes the method behind them, and no third party has audited them. Read them as a reason to run a trial, not as a planning assumption.

The production layer: Layer. Layer sells AI creative production for mobile games across UA, LiveOps and game art. On the UA side, its agent reads campaign results from AppLovin, Meta, Google Ads and TikTok and fans a winning creative out into ranked variants. Layer’s pitch rests on creative fatigue, and its own figure is that a mobile ad wears out in 5-14 days. It does not set bids; it keeps the bidder supplied.

Both layers sit inside the same paid UA machine, and that limits what they can change.

What changes in a studio’s paid UA budget math?

AI UA tooling attacks waste, not price. It does that in four places.

• The lag between a campaign going bad and someone cutting it. Mobile game UA runs on long payback windows. Kohort’s own description of the problem is that by the time D90 ROAS comes in, a studio has already spent another quarter on campaigns that were never going to work. A credible early LTV prediction shortens that window, and the spend it saves is the main return on the tool.
• The quality of the signal sent to the networks. Value-based bidding is only as good as the value signal. Feeding a network a predicted long-term ROAS instead of an early event can move spend toward users who keep paying.
• Analyst hours. Automated reporting and always-on monitoring replace part of the daily spreadsheet work, often the scarcest input on a small growth team.
• Creative throughput. If fatigue is what kills your winners, faster variant production keeps a working campaign alive longer at the same cost per creative.

Now the cost side. On its site in October 2026, Kohort lists a Startup tier for studios spending under $100k a month on UA, from $600 a month per title billed annually, and a Growth tier for studios spending over $100k a month, from $1.2k a month per title, also billed annually. The structure matters more than the price: the fee is fixed per title, while the waste it removes scales with spend. Heavy spend on one title absorbs the fee easily; small tests across several titles pay it several times over a much smaller base.

There is a second, less visible cost: data. A cohort model trained on your own UA data needs cohorts large and mature enough to learn from. A title in soft launch, or one buying a trickle of installs per network, gives the model little to work with, and a confident prediction built on thin data is worse than an honest “too early to tell”.

What AI UA tools do not fix

Every gain above happens inside an auction you do not control, on a store whose terms you do not set. Four things stay exactly where they were.

The auction itself. An AI agent makes you a better bidder in the same auctions as everyone else, and your competitors can buy the same class of tool. As more studios bid on predicted value, the edge shifts to whoever has better data and better creative, not to whoever has an agent. The auction price is set by the competition for the audience, and an agent does not reduce that competition.

Store economics. Your margin after acquisition still passes through the store’s fee schedule. Since 30 June 2026 in the EEA, the UK and the US, Google Play charges 10% plus a 5% billing fee, where Google Play Billing is used, on a developer’s first $1M of annual earnings, with higher rates above it (our breakdown of the Play repricing covers the full grid). On the App Store, the Small Business Program’s 15% commission applies to developers who opt in and earned up to $1 million in proceeds in the prior calendar year. No UA agent touches those lines.

Audience reach. Paid networks reach the people who use those networks’ inventory. If your genre’s profitable audience there is already being bought by better-funded competitors, a smarter bidder finds the remaining profitable pockets faster; it does not create new ones. A good prediction model may simply tell you, earlier and with more confidence, that there is no profitable scale left in the channel. That is valuable information, not growth.

Payment rails. A user acquired in a market with low card penetration converts no better because an AI chose them. If the bottleneck is payment friction, the fix is billing and checkout — carrier billing, a web shop run as a real channel — not bidding.

AI UA tooling or another distribution channel: where should the next euro go?

The decision turns on where your problem actually sits. AI UA tooling improves a paid channel that already works; distribution adds reach the paid channel cannot reach.

Your situation	Where the next euro goes	Why
Some campaigns pay back, decisions are slow and manual, spend per title is steady	AI UA decision tooling	The waste is in the lag; predictions and automation recover it
Winning creatives die quickly and the team cannot keep up	AI creative production	Throughput is the bottleneck, not bidding
Campaigns fail to pay back across networks even with good creative	A non-auction distribution channel	The auction price is the problem, and no bidder changes it
Players arrive but do not pay, or store fees eat the margin	Billing, web shop or alternative store terms	The leak is after the install, not before it
One title in soft launch, small test budgets	Neither tool yet; fix retention and monetisation first	Models need mature cohorts; distribution needs a product that holds players

The rows are not mutually exclusive, but one caution applies: an AI tool’s dashboard will report on the channels it sees. It will not tell you what an OEM preload, a store featuring slot or an instant-play surface would have delivered, because those installs never enter its auction. Our multi-channel distribution framework sets out how to score channels side by side so the paid line does not win by default simply because it is the best-instrumented one.

What should a studio do next?

A practical sequence for a small or mid-size studio weighing an AI UA tool against a distribution test:

1. Measure your own lag. For the last quarter, count how long campaigns ran after they had in fact stopped paying back. That spend is the ceiling on what a prediction tool can save you.
2. Trial on one title with a holdout. Run the vendor’s predictions on one title and keep a comparable set of campaigns on your current process. Judge the tool on predicted versus realised ROAS once the cohorts mature, not on the vendor’s headline accuracy.
3. Price the tool against the waste. Compare the per-title fee with the waste measured in step 1. If the fee is a large share of that number, the tool is not your next euro.
4. Price one channel outside the auction in parallel. Pick the alternative channel that fits the title — an OEM store, an alternative app store, carrier billing in an emerging market, a web build — and estimate its integration cost and expected cohort quality.
5. Reallocate each quarter. Move budget to whichever line returned more per euro, and keep paid UA as one line in the plan rather than the plan itself.

FAQ

What is an AI user acquisition agent for mobile games?

It is software that predicts the long-term value of users bought through paid campaigns and acts on those predictions: it sets or suggests bid targets, recommends scaling or cutting campaigns, monitors for anomalies and, on the creative side, generates new ad variants. Kohort’s Ktrl covers the bidding and prediction side; Layer covers creative production.

How much does Kohort’s Ktrl cost?

On its site in October 2026, Kohort lists a Startup tier for studios spending under $100k a month on UA, from $600 a month per title billed annually, and a Growth tier above $100k a month, from $1.2k a month per title, also billed annually. Enterprise pricing for large portfolios is quoted on request.

Is Kohort’s 95% prediction accuracy reliable?

It is a company claim. Kohort said its models produce daily campaign-specific predictions with approximately 95% accuracy, and its website claims 94%+ accuracy on day-365 LTV forecasts; neither publishes a method or an independent audit. Test it on your own matured cohorts before relying on it.

Can AI UA tools replace alternative distribution?

No. AI UA tools optimise bids, budgets and creatives inside the same paid auctions and the same store economics. They do not reduce store fees, add audiences outside the ad networks or fix payment friction. They make a working paid channel more efficient; distribution channels add reach the paid channel cannot buy.

---

If you are a small studio deciding whether your next budget line is a UA tool or a new channel, the Founding Developer Program is built for exactly that conversation. Larger publishers can go straight to our distribution team.