Ktrl drove a 20% ROAS lift for Leke Games, with zero data science overhead
Leke Games' data scientists are 100% on product. Growth Manager Melih Yurduseven was building pLTV models by hand - until Ktrl took over prediction, cut decision cycles from weeks to the same morning, and flagged a hidden winner now paying out at ~20% higher D30 ROAS than the rest of his portfolio.
Leke Games' data scientists are 100% dedicated to product and gameplay, none assigned to marketing. That left Growth Manager Melih Yurduseven building pLTV models by hand to run UA on TDZ Pro, with no data science hours to draw on. Ktrl took over prediction, turned two to three week decision cycles into same-morning calls, and flagged a below-benchmark campaign as a hidden winner weeks before the data would have proven it, a call now paying out at ~20% higher D30 ROAS than the rest of his portfolio.
Running UA while data science builds a better game
Our company is focused on building a great game. The team is game development-first: the roadmap, the game itself, comes before anything else. That means our data scientists are fully focused on product and gameplay, which is where they should be if we want to make a great game. But it also meant that there's no data analysis tool, and no specialist, in my own team. If I wanted LTV predictions to run UA properly, I had to build them myself.
The biggest change for me is that I don't have to choose between deciding early and deciding right anymore. Before, I was waiting for the data to mature before touching a campaign. Now I get a forward-looking view on day one, and I can move budget while it still matters.
Before Ktrl, I built our ROAS targets from our own historical cohorts, looking at how past players monetised over time and turning that into the early ROAS levels a campaign needed to hit. For pLTV, I used growth multipliers in our BI tool: take a cohort's early revenue on D1 or D7, and multiply it by ratios calculated from matured cohorts.
It worked, but every time market conditions changed, we shipped a big update, or the traffic mix shifted, the multipliers had to be recalculated by hand. For established markets, accuracy was fine. The data was ours and there was a lot of it. But a fresh geo, a fresh channel, or a campaign only a few days old had no matured data behind it, so accuracy dropped exactly where decisions mattered most.
The approach did not scale with the number of decisions I had to make. Keeping projections fresh for every campaign, network, and country at the same time was hugely time-consuming. It also made me more conservative than I wanted to be on new campaigns, because acting early without a matured signal was a real risk.
I open the dashboard in the morning and the predictions are already there. They are current, campaign-level, and they come with a confidence score. Prediction stopped being a project I did from time to time and became part of how I manage spend every day.
From signup to partner in two weeks
Leke Games found Kohort themselves and signed up self-serve. Within days, they had walked through the platform, ingested their data, and proved the uplift: a backtest on TDZ Pro's own historical cohorts had hit 92% D90 accuracy and they decided to partner with Kohort after a 7-day trial.
When I first compared Ktrl's numbers against my own curves, they were close, and that showed me the foundation was solid. The real trust came from checking predictions against actuals as cohorts matured, and seeing they stayed accurate even on noisy campaigns.
The confidence score is what makes it usable day to day. It tells me which calls are safe to act on immediately, and which ones need another day or two of data before I touch a budget.
The campaign we would have waited on
The clearest example is a new campaign where early signals came in below our benchmarks. Normally I would not touch the budget. I'd wait for D7 and D30 data to mature, especially running many campaigns on a limited budget.
Ktrl showed us early, with a high confidence score, that we actually had a strong signal. We scaled the campaign weeks before I normally would have, and it turned out we had reached a new audience that monetises over the long term.
That campaign is still one of our main campaigns today, generating around 20% higher D30 ROAS than our other BAU campaigns.
It shortens the decision cycle. A decision that used to need two or three weeks of cohort data now happens in the same morning. The confidence score also tells me which decisions are safe to make immediately and which ones need another day or two of data.
One person, the whole platform
Without a data scientist assigned to UA, the breadth matters as much as the predictions. The daily recommendations with confidence scores are what I use most. The incremental blended view, measuring how much organic traffic each paid campaign actually drives, shows the real contribution of every channel and gives us a much stronger base for budget decisions. And since we connected our ad networks, bid targets come paced: already adjusted to the right size and timing for each network, so I can apply them directly.
I didn't have any BI tools or data scientists, so having Ktrl is like having one dedicated just to me. So far it's been very useful, and has made my job much, much easier.
What one UA manager does with the analysis hours back
With per-country forecasts refreshed daily, I notice a market performing well and put money behind it early, instead of finding out in a monthly review. Expansion became continuous small steps rather than a few big planned bets.
It changed the leadership conversation too. The discussion starts from where campaigns are heading, not where they were three weeks ago, and we all look at the same numbers. Asking for more budget is a very different discussion when I can show projected payback with a confidence score next to it.
With Ktrl covering the prediction side, the team is freed up to focus on the parts of UA that still need a human.
With Ktrl covering the prediction side, we will act much more like creative and business strategists on UA. The hours I used to spend rebuilding projections now go into creative strategy, bids, and budget changes. The energy that used to go into analysis goes into how we present a better product and how we reach the right users, while we keep scaling our flagship title TDZ Pro and bring new titles to market. Our big focus for 2026 is a faster creative testing loop to keep CPIs under control. Creatives reach saturation much faster than before, and on some networks we need to refresh them two or three times a week to stay competitive. Since Ktrl took the prediction workload off us, we can shift our focus exactly there.
Key outcomes
A full prediction stack, no marketing data science required. Leke's data scientists stay 100% on product and gameplay. Ktrl replaced the growth-multiplier models Melih rebuilt himself after every update and traffic-mix shift.
Decision cycle cut from 2–3 weeks to the same morning, with confidence scores gating which calls to act on immediately and which need another day of data.
~20% higher D30 ROAS on a campaign they would have waited on. Flagged early with high confidence despite weak early signals, scaled weeks ahead of schedule, and still one of Leke's main campaigns.
92% D90 backtest accuracy, self-serve to partner in two weeks. Validated on TDZ Pro's own cohorts during a 7-day trial, with whole-platform adoption by a single UA manager: confidence-scored recommendations, incremental blended measurement, and paced bid targets.
About Leke Games and TDZ Pro
Leke Games (Leke Yazılım ve Oyun A.Ş.) is an Istanbul-based mobile game developer and publisher founded in 2022. The studio is best known for its TDZ (Traffic Driving Zone) franchise, producing multiplayer racing, simulation, and casual games for iOS and Android, including TDZ Pro and TDZX.
TDZ Pro - Bus Truck Simulator is Leke's flagship driving simulator: 50+ vehicles spanning trucks, buses, coaches and supercars, a tycoon business mode, multiplayer, a realistic economy with freight auctions, and an open-world map with dynamic weather. The title has passed 2 million downloads, with most of that growth in just the last 6 months, and holds a 4.3-star rating.