Back to blog
Case Study · November 12, 2026

How Battle Creek Games and Kohort took UA predictions agentic

Battle Creek Games runs UA through an agentic workflow, with agents cross-checking each other and a person on every final call. They wanted Ktrl's predictions inside that workflow and built the connection themselves. Kohort took it from there: the Ktrl MCP server is now a standard feature any studio can switch on.

Ethan Silovsky
Ethan Silovsky
Marketing Director, Battle Creek Games · No Limit Drag Racing 2
How Battle Creek Games and Kohort took UA predictions agentic

Battle Creek Games builds racing and hunting titles, including No Limit Drag Racing 2, which has passed 10 million downloads on Google Play. The studio's UA operation runs on an agentic workflow built on MCP: pointed questions rather than open-ended prompts, and cross-checks between agents before any campaign change goes out.

Battle Creek became a Kohort customer in March 2026. Since then their requests have shaped Kohort's roadmap, and Ktrl has become part of how their agents work. Four takeaways from the partnership so far.

Takeaway 01: Treat prediction as a component

Battle Creek wanted live predicted LTV and ROAS for every campaign, network, and geo. The studio runs its own warehouse and data team, and chose to keep them focused on first-party analytics rather than building predictive LTV from scratch. So they treated it as a component decision: connect the MMP and let Ktrl handle the modelling.

Takeaway 02: Put predictions where the work happens

The Ktrl dashboard anchors Battle Creek's weekly UA sync per game: network-level performance first, then campaign-level decisions. Between syncs, the studio's agents query the predictions directly. A question gets a current answer in the tool where it was asked, with no report to pull and no tab to switch to.

The setup: How the loop runs

MMP DATAattribution, spend, revenueSTUDIO DATA FEEDfirst-party pipelineKTRLdaily predictions · recommendations · alertsKTRL MCP SERVERqueryable from Claude, ChatGPT, Cursor, or custom agentsBATTLE CREEK'S AGENT LAYERAI agents, querying Ktrl on demandanswers cross-checked before any campaign changeHUMAN REVIEWevery change approved by a personAD NETWORKSbid and budget changes appliedRESULTS FLOW BACK

In practice, the questions sound like: how do this week's campaigns compare with last week's, which campaigns are running under their target ROAS, and what does Ktrl recommend for this network's bids, at what confidence. The answers come back with the same numbers the dashboard shows, pulled live.

Takeaway 03: If the integration doesn't exist, ask

Battle Creek didn't wait for an integration to exist. In March, the studio asked whether Ktrl's predictions could feed the MCP setup they had built in-house, and wired it in themselves: the first Ktrl deployment inside an agentic stack.

That request shaped Kohort's roadmap. In August, Ktrl shipped a first-party MCP server, connecting predictions, cohort data, and spend recommendations to Claude, ChatGPT, Cursor, or any agent framework. It now ships as a standard feature.

Takeaway 04: Keep a person on every final call

Battle Creek's agents retrieve, summarise, and cross-validate. They don't decide. Every campaign change is approved by a person, and the prediction is treated as one signal among several, weighed against profitability and the goal at hand.

About Battle Creek Games

Battle Creek Games is a US-based mobile game studio behind racing and hunting titles including Offroad Outlaws, No Limit Drag Racing 2, American Marksman, and Offroad Outlaws Drag Racing. No Limit Drag Racing 2, the studio's flagship racing title, has passed 10 million downloads on Google Play.

Ready when you are

Stop hoping your campaigns pay back.

Daily ROAS signals trained on $6bn of UA spend. Free to start, set up in minutes.