Mobile gaming's missing brain

I've been speaking with a lot of people inside mobile gaming over 2026 -- devs, publishers, vendors, practitioners. And what's quietly surprised me across dozens of these conversations is how often the standard approach to optimising the funnel is to manually investigate what's happening at each stage in fairly simple ways -- pulling cohort numbers, eyeballing the curves, looking for which week's update broke retention -- rather than actually building a real understanding of how players flow through the game.

Players play a game because it's fun. And fun isn't a funnel stage -- it emerges from a series of interactions (pacing, friction, surprise, social loops, mastery moments) combined in virtually infinite ways. Reducing that to a D7 retention curve or a level 1 → 2 conversion gate is oversimplifying what you're actually trying to understand.

Every studio is essentially trying to answer the same three questions:

• Acquisition. How do I maximise ROAS by getting high value players at the lowest cost?
• Retention. How do I keep them playing?
• Monetisation. How do I get them paying more?

These are individual player questions -- which specific player is high value, which one is about to churn, which would respond to which offer -- but the tools every studio actually runs on, like ARPDAU dashboards, ROAS reports, and D1/D7/D30 retention curves, only ever tell you what the cohort did on average. Knowing your D7 retention is 18%, or that median D28 retention sits below 1% across mobile games, tells you nothing about which of your players are actually about to leave next week.

Most teams know they're chasing the wrong thing. DAU climbs 20% on the back of a new UA campaign and revenue stays flat, because the incremental DAU is low LTV traffic that was never going to convert in the first place. The number on the dashboard went up; the business didn't. That's the textbook vanity metric -- a measure that improves without the underlying outcome improving. When a measure becomes a target, it ceases to be a good measure.

Every successful game has at least one person obsessed with the individual player, not the cohort. Scaling that obsession across millions of players is the hard part -- what most studios don't have a system for.

The cleanest public example is Royal Match. A player gets stuck, closes the app, comes back the next morning, and finds a 60% off booster bundle on the home screen at exactly the price they would pay. Not someone else's. Theirs.

Which is obviously a model running silently underneath the game -- watching what each player does, predicting churn, picking the right offer at the right moment. So who built that model? Who owns it? And who is it actually optimising for?

Most mobile gaming publishers don't have a confident answer.

How much does this actually matter?

Where your installs actually go

The funnel is brutal. Roughly 1.83% ever make an in app purchase, and the top 1% of those payers generate ~29% of all IAP revenue. 72% of mobile game developers are advancing hybrid monetisation -- IAP plus in-app advertising -- which means every session ends up with a live decision underneath.

The industry benchmark ranges are unforgiving across both IAA (ad-funded) and IAP (purchase-funded) games:

Metric	Median	Top quartile
D1 retention	27%	40%+
D7 retention	8%	18-20%
ARPDAU (casual / IAA)	US$0.05-0.15	US$0.30+
ARPDAU (mid-core / IAP)	US$0.50-1.00	US$2.00+

Sourced from industry benchmarks.

The leverage point is retention. Take a mid-core IAP game running at median benchmarks (ARPDAU US$0.50) and lift D7 retention by 5pp -- from 8% to 13%. Hold the rest of the curve shape constant:

	Baseline	+5pp D7 lift	Delta
D7 retention	8%	13%	+5pp
D7 survivors per 1,000 installs	80	130	+50 (+62%)
90 day LTV per install	~US$1.50	~US$2.25	+US$0.75 (+50%)
90 day revenue, 100k install cohort	~US$150k	~US$225k	+US$75k

Almost all of the lift comes from a healthier post-D7 cohort. Early retention feeds the entire downstream curve, so the effect compounds.

If you can't predict who the 1.83% are early, who the top 1% are even earlier, and what to do with everyone else session by session... you're running a casino with a really nice retention dashboard.

Apple's ATT rollout in April 2021 broke this. only 4% of US iPhone users opted into app tracking; iOS cost per install on casual games rose 88% by Q4 2022. The industry got more expensive overnight, and precision dropped.

ATT broke rented targeting. Studios that owned their player data were fine. Studios that depended on ad network targeting got crushed.

So what would the brain actually do?

A real player intelligence capability has four layers -- events, identity, prediction, activation.

Events. The raw stream, and every dashboard sitting on top. "User opened the app at 7:42pm. Completed Level 47. Watched a rewarded ad." GameAnalytics, Firebase, MMPs. Every ARPDAU and retention curve is an aggregation over this stream. Solved.

Identity. Stitching events into a picture of who. "Same user who installed two weeks ago from a TikTok ad, spent US$4.99 on Day 3." IDFA used to handle this; post ATT it's probabilistic methods, MMP fingerprinting, SKAdNetwork aggregation, and first party logged in IDs where you can get them. Contested, functional.

Prediction. What's likely to happen next? "This player has a 78% chance of churning before Day 7. pLTV US$0.40. Don't waste a personalised offer; show an ad." This is the brain.

Activation. Acting on the prediction -- offer, difficulty, lookalike bidding, ad placement. CleverTap, Braze, MoEngage. Crowded vendor space.

Royal Match is the brain in action. Knows you got stuck. Knows you closed the app. Knows you went quiet. Predicts you'd come back at the right price. Picks it. Triggers the offer. Dashboards are downstream -- how the team measures whether the brain works.

Building this needs three things:

1. Your own player data, owned end to end. Not rented from ad networks. Not stitched from third party MMP postbacks. Your event stream, your identities (as much as ATT allows).

2. Models trained on YOUR game. A churn model calibrated on retail data doesn't predict what your players will do. Match 3 churn looks different from CCG. Hardcore RPG whale precursors look different from casual.

3. Real time plumbing to push predictions into action. Predictions are useless sitting in a warehouse. They need to fire to the offer engine, LiveOps tool, UA bidder, difficulty engine.

The brain is a system that turns events into player level predictions and routes them to the things that act on them.

Why isn't every studio doing this?

If the brain is so obviously valuable -- and the economics so leveraged -- why hasn't every studio built one?

Three reasons. They stack.

1. Cost. A real prediction layer needs a serious data science team. Playtika's Boost Platform has ~40% of headcount in R&D. Supercell, King, Zynga, Scopely all run their own ML. Works at the scale where a fifty person data science org pays for itself -- top 50 publisher, maybe top 100 in 2026. Below that line, the fixed cost doesn't justify.

2. Time. Even with budget, you don't have the months. Event pipelines, model training, validation, integration with activation tools -- six to twelve months before you ship one production prediction. Most studios don't have that runway if economics are wobbling.

3. The vendor market doesn't bridge the gap. Three off the shelf options exist; none close it cleanly.

Rent from ad networks. AppLovin's AXON has run up to three trillion predictions per day since its 2021 IPO filing. Unity Vector launched GA in May 2025 and is on track to do US$1B+ annual run rate by end of 2026. AppsFlyer, Adjust, Singular all ship predict-ROAS products. Good models, more compute than you have. The catch -- every model is owned by the entity that gets paid when ad impressions or installs run. AppLovin's pLTV is calibrated for one thing: who'll look profitable inside the SKAN postback window -- where the network gets paid. That's a different target from Day 90 retention, and a different cohort.

Horizontal predictive SaaS. Pecan AI is the cleanest example -- build a churn or pLTV model without a data scientist. Serves PlaySimple, SciPlay, KSG Mobile, Pixio; 95% accuracy at Day 30 from Day 2 signals on PlaySimple. But Pecan is horizontal -- retail, finance, SaaS, gaming. Doesn't speak session rhythms, monetisation funnels, whale precursors. You end up rebuilding the gaming native layer on top.

Build it in house. See the cost section. Only viable at the top of the market.

The shape for the missing product is clear -- gaming native, predictive, drops in like an analytics SDK. Whoever ships it inside three years wins.

Until that exists, the dashboards stay descriptive. The decisions stay heuristic. The brain stays missing.

What this means if you're running a mobile gaming business

The question to actually ask: where does the brain live in our stack today?

Most studios will say "our analytics dashboards." The dashboards are descriptive. The honest answer: at our ad network, our MMP, our CRM tool, or nowhere. None of those is "in our product, owned by us, optimised for our economics."

Follow up: which decisions depend on prediction we don't actually have? Difficulty curves. Offer timing. Reengagement triggers. UA bidding. LiveOps segmentation. Ad floor pricing. If those decisions run on heuristics, vibes, or rented predictions optimised for someone else's economics -- that's the gap.

Most of the margin shift over the next five years comes down to which side of this gap each studio lands on. Top 50 publishers built the brain themselves. Ad networks absorbed it for everyone else. Studios in the middle without an owned prediction layer get acquired, rolled up, or run on margins that don't justify existing.

What does your stack actually predict?

---

Disclaimer: Thoughts are my own and do not represent any other parties.