Appvertiser AI Logo

AI Platform

Growth Services

Studio Sáng Tạo

Tài nguyên

Studio Sáng Tạo

Gaming

Solitaire Spellbook by DeNA: 30% Higher IPM

Solitaire Spellbook by DeNA: 30% Higher IPM

How Appvertiser AI produced AI creatives for DeNA's Solitaire Spellbook that outperformed in-house and agency work, delivering 30% higher IPM across 5 channels.

How Appvertiser AI produced AI creatives for DeNA's Solitaire Spellbook that outperformed in-house and agency work, delivering 30% higher IPM across 5 channels.

Trusted by teams behind leading apps & games

Trusted by teams behind leading apps & games

EXECUTIVE SUMMARY

Appvertiser AI produced multi-format gaming creatives for DeNA, including statics, 2D assets, gameplay storytelling, and interactive playables.

Appvertiser AI produced multi-format gaming creatives for DeNA, including statics, 2D assets, gameplay storytelling, and interactive playables.

REGION

US

CÁC KÊNH

+5

CAMPAIGN GOAL

Creative Performance & IPM Optimization

2

2

OS

+5

+5

Growth Channels

+30%

+30%

IPM

4

4

Creative Formats

01 / CHALLENGE

The constraint

The constraint

DeNA needed its Solitaire Spellbook launch creative to outperform both an established in-house production team and an outside agency, while producing enough format variety to keep IPM climbing across five acquisition channels in the competitive US casual games market.

DeNA needed its Solitaire Spellbook launch creative to outperform both an established in-house production team and an outside agency, while producing enough format variety to keep IPM climbing across five acquisition channels in the competitive US casual games market.

02 / STRATEGY

The operating model

The operating model

AI creative production spanning four formats — static ads, 2D video, gameplay storytelling, and interactive playables — tested head-to-head against DeNA's in-house and agency-produced assets across TikTok Ads, Unity Ads, Meta Ads, AppLovin, and Google Ads to isolate the highest-IPM concept.

AI creative production spanning four formats — static ads, 2D video, gameplay storytelling, and interactive playables — tested head-to-head against DeNA's in-house and agency-produced assets across TikTok Ads, Unity Ads, Meta Ads, AppLovin, and Google Ads to isolate the highest-IPM concept.

03 / EXECUTION

The system in motion

The system in motion

Produced a full mix of AI-generated static ads, 2D video, gameplay storytelling, and interactive playables for Solitaire Spellbook. Ran the creative set alongside DeNA's in-house and agency-produced assets on iOS and Android across five channels, tracking IPM by concept and format to surface the top performer.

Produced a full mix of AI-generated static ads, 2D video, gameplay storytelling, and interactive playables for Solitaire Spellbook. Ran the creative set alongside DeNA's in-house and agency-produced assets on iOS and Android across five channels, tracking IPM by concept and format to surface the top performer.

04 / RESULTS

Measured outcomes

Measured outcomes

30% higher IPM than DeNA's second-best creative. Outperformed every in-house and agency-produced asset on the account. 4 AI-native creative formats produced and tested. Live across 5 channels on iOS and Android.

WHAT ENTERPRISE TEAMS CAN TAKE AWAY

A repeatable growth pattern, không a one-off win.

A repeatable growth pattern, không a one-off win.

AI creative production can out-compete both internal teams and outside agencies on performance, not just speed. Testing multiple formats in parallel — not just one — is what surfaces the highest-IPM concept in a crowded casual games market.

Frequently Asked Questions

Frequently Asked Questions

Appvertiser AI is a recommendation AI for mobile user acquisition. Every morning it reads every campaign, ad set, ad, keyword and country in your ad accounts, judges each one against your own goal, and writes a short list of recommendation cards: raise a budget, lower one, pause an ad, raise a target ROAS, exclude a country, replace weak assets. It's a symbolic, graph-based decision model that reasons the way a senior UA manager does, so every card can say exactly why. Your team decides, and Apply makes the change in one click.
No. Your data is used to make decisions for your account only.
No. Every read and write is scoped to your account.
Yes, in a multi-tenanted, safe and secure environment.
Ad networks: Google Ads, Meta, TikTok, Apple Search Ads, Unity, AppLovin, Mintegral and Moloco, for spend, budgets, bids and delivery status. MMPs: AppsFlyer, Adjust, Singular, or your own BigQuery warehouse, for installs, revenue, in-app events and cohort revenue by install day.
Every night, for every app, day and metric, our figures are compared with the partner's own. Spend must match within 2%, installs and revenue within 3%. A day is checked once it has settled: one day for spend, two for installs, three for revenue and events. A card that includes a day that failed the check says so.
If a network's data hasn't refreshed, the AI writes no cards for that network that night. A failed or partial read is never treated as zero, and the AI makes no ROAS call where more than a tenth of the spend sits on days the data doesn't state.
No. Decisions come from a symbolic, graph-based decision model. No language model writes card text. It's a deterministic model rather than a probabilistic one.
The AI holds your account as a connected structure: network, operating system, campaign, ad set, ad, keyword and country, each linked to its parent and siblings. A campaign's verdict decides what happens inside it, a fix inside a campaign comes before pausing the whole campaign, and an ad is judged beside the other ads in its ad set.
Every sentence on a card is built from figures it read, so a card can't state a figure it didn't read.
It holds. A missing figure stays missing and leads to no verdict.
Yes. Each card shows the finding, figures, target, reason and exact change.
Yes. It remembers every change and grades every call; your team's feedback adjusts its confidence.
Every card is graded on the 14 days after it was made, once that data has settled, and the grade is never changed. Cards that can't be tested (missing data, no target, the campaign changed mid-window) are marked untestable, not scored. For each client, the AI weighs each kind of card by how often that kind of call has held up.
Each decline reason teaches something different. "I disagree with the call" lowers confidence in that kind of card for your account. "Not now" snoozes it for a week. "This campaign is off-limits" keeps the AI away from it for a year. "The numbers look wrong" goes to our team as a data question and doesn't count against the AI.
No. Nothing changes until someone with permission presses Apply.
Yes: budget caps, change-size limits, cooldowns, optional second approver and a kill switch.
Safety checks run before any change; every action is logged and graded. Mistakes can be corrected by giving feedback through the card, using the Decline button and adding an explanation.
Anything the network's API allows: campaign budgets, bids, target ROAS and status; ad set budgets and status on Meta and TikTok; ad set status on Google and Apple Search Ads; ad status on Meta and TikTok; keyword bids and status on Apple Search Ads; and country exclusions on Google, Meta and Apple Search Ads. Where the network refuses the change, the card shows Agree instead.
A preview runs every safety check and stops just before the network. You get a few seconds to undo, then the checks run again for real, the change is made and logged with before and after values, and the result is tracked in Active Decisions.
After a budget raise, the next raise waits until the first has matured: seven days plus your ROAS window, so 14 days on a D7 client. After a cut, the next raise within 30 days is half the usual size, and there's no second cut while the last one is still settling.
It judges on matured cohort days; projections only come from tested curves, and a projection never raises a budget alone. It also judges on matured early KPIs.
Your MMP for installs and revenue; spend from the network that charged it.
Nightly checks against partner figures; when data doesn't match, the card says so or the AI holds.
The one you set a target for: D1, D7 or D30. Most clients judge on D7. Targets can be set per app, operating system, network, country or campaign, and the most specific one wins. The AI never uses a platform default as your target.
Install days younger than your ROAS window are projected from the point they've reached (D1 or D3) on the campaign's own revenue curve. A curve only projects after 28 install days of evidence, with 70% of past projections in range, a median error of 15% or less and a bias of 10% or less.
Your risk profile sets the lines. On Balanced (default), a campaign under 85% of target gets a 10% budget step down, and under 65% a pause. Conservative uses 95% and 70%; Aggressive, 75% and 60%. A campaign under target keeps running while its operating system's total still meets target.
When a campaign is at or above target in its matured and newest days, spends at least 90% of its budget, and has no earlier raise still settling: a 10% raise. Raises are funded from your budget plan's headroom and from money that night's cuts free up.
If MMP installs drop sharply in the last two or three days while the network's own clicks and installs hold steady, the AI reads it as attribution still arriving and lowers its confidence in any cut.
Never. New kinds of change are proven on our own ad accounts first.
Every attempt, made or refused, is logged with before/after values. Access: details pending.
No. One night's pauses can never stop more than half of a network's active spend, each network gets at most ten cards a night, and new campaigns aren't cut before they've had their ROAS window plus a week to show their return.

Sẵn sàng khi bạn cần

See the AI growth workforce in action

See the AI growth workforce in action

Book a walkthrough of how Appvertiser AI turns campaign signals into decisions, execution, and measurable growth.

Book a walkthrough of how Appvertiser AI turns campaign signals into decisions, execution, and measurable growth.

Book a demo

Explore all case studies

Explore all case studies

→

Keep reading

Chiến lược tăng trưởng

US, UK

Skillz: Multi-Channel UA Strategy Driving +107% D180 ROAS

Skillz: Multi-Channel UA Strategy Driving +107% D180 ROAS

Appvertiser AI managed a $2.1M+ multi-channel UA operation for Skillz across 13 ad networks and DSPs over 3 months (August–November 2023), achieving +107% average D180 ROAS and -6.3% CPA reduction in the US and UK markets.

Appvertiser AI managed a $2.1M+ multi-channel UA operation for Skillz across 13 ad networks and DSPs over 3 months (August–November 2023), achieving +107% average D180 ROAS and -6.3% CPA reduction in the US and UK markets.

+107%

+107%

Average D180 ROAS

-6.3%

-6.3%

CPA Reduction

Read story

→

Chiến lược tăng trưởng

Hoa Kỳ

Dave: Scaled Monthly UA Spend from $120K to $2.5M

Dave: Scaled Monthly UA Spend from $120K to $2.5M

Appvertiser AI partnered with Dave, a digital banking app with 10M+ users, to scale monthly UA spend from $120K to $2.5M — a 1,983% increase — through bid strategy optimization across Google Ads, Apple Search Ads, TikTok, and Snapchat. Dave went from a unicorn startup to a $4B valuation and a successful IPO during the engagement.

Appvertiser AI partnered with Dave, a digital banking app with 10M+ users, to scale monthly UA spend from $120K to $2.5M — a 1,983% increase — through bid strategy optimization across Google Ads, Apple Search Ads, TikTok, and Snapchat. Dave went from a unicorn startup to a $4B valuation and a successful IPO during the engagement.

$2.5M

$2.5M

Chi tiêu hàng tháng

+198%

+198%

Chiến lược tăng trưởng

Read story

→