Ilmari
Ihalainen

Problem

Return on investment was too small.

Solution

Hotel Hideaway’s UA KPIs were bad. I couldn’t figure out why. I worked through the variables one at a time: geos, user demographics, Android against iOS, creatives. None of them moved the number meaningfully, which ruled out targeting and creative as the cause.

Campaigns were at the time optimising for installs. Real purchases were too rare and came too late in the player’s life to train on, and while the ROI was slightly higher with the purchase optimisation, cost per install was too high to sustain. The platform bidding had no way to tell a future payer from a tourist. No real proxy events.

I collaborated with live-ops and development and created a new early-funnel €0.99 purchase event. It was cheap enough that a meaningful share of new players bought it in their first sessions, which made it frequent and early enough for the ad platforms to optimise against, while still identifying someone willing to spend. Those who convert early retain for a long time, and typically purchase again and again. I moved the campaigns to target that event and concentrated budget on the highest-value market and OS combination, US on iOS.

Outcome

ROI tripled in February 2024 and stayed up. The improvement landed while the UA budget was being renegotiated, and it carried the case for keeping Sulake’s parent company on the agreed high-budget deal. Without it we would have been scaling UA down instead.

Problem

Poor campaign performance, lack of quality creatives, and lack of in-house creative resources to support scaling on paid ad platforms.

Solution

Like a primal monkey touching a monolith and discovering tools and fire, I touched the TikTok Creative Challenge and discovered that User Generated Content on TikTok works very well for Hotel Hideaway.

Stepping into the role of managing user acquisition in September ‘23 came with many unsolved issues: previous months of poor campaign performance, out-of-date creatives, and small resources for in-house video ad production (just me, lol).

My account managers introduced me to the TikTok Creator program, and since we were already doing paid UA on TikTok, it felt natural to give TTCC a shot — it could solve both problems.

TTCC lets brands submit a text brief and gameplay visuals, which creators then turn into ad videos that can be used in paid campaigns. Recording interesting gameplay and writing video briefs for the creators felt natural to me — I had all these ideas but no resources for execution.

Outcome

Following the initial campaign setup, the new UGC videos quickly outperformed any old video ad inventory items we had previously used, resulting in significant improvements in CPI and ROI in September 2023.

Parental company of Sulake, Azerion, recognised the potential for product growth through paid UA and approved a higher user acquisition budget going forward.

Problem

Allocating all of the budget to the high-return market resulted in a sharp decline in total volume of users acquired, and it had the risk of hurting the game’s user base long-term.

The downside of acquiring high-retention, paying iOS-US users was that they were expensive. Hotel Hideaway is a social game: user retention and conversion to payers go hand in hand (long-term users end up paying, early payers end up playing long-term). The biggest metric we could tie to retention was whether the user was able to make friends. Those who interacted with other users and made a friend on their first day were over 3× more likely to stick around, and eventually convert to payers.

A large decline in monthly active users would be detrimental to the game’s success and would affect retention and paying behaviour of all existing users (less users = less friends = less reason to stay and pay). Upper management wanted this problem solved.

Solution

For UA this meant finding a medium where we were still acquiring high-retention paying users, but they had to be cheaper.

After months of testing, the eventual best-performing user cohort was EEA Android users. Not US, not iOS — but Android EEA.

It was initially surprising how Android users could result in higher ROI than iOS, but it started to make sense when inspecting global phone trends. iOS makes up a majority of mobile devices in the US, and there’s a certain stigma to Android, whereas in the EU there are 2× more Android users than iOS users. The average buying power of an EU Android user wasn’t quite the same as a US Android user, but the CPA was significantly lower — resulting in the best ROI on the Android EU cohort.

Having discovered the best user cohort for the time being, I was also able to test 1000+ TTCC ads and different ad genres — for example ‘relationship drama’ themed ads vs ‘cutesy’ ads or ‘dress-up’ style ads — and evidently came to the conclusion that drama ads continued to have the highest click-through rates and best ROAS.

The user funnel was further refined by multiple iterations of the store landing pages, app icons, and improvements in the game’s early user monetisation.

Outcome

Through broad marketing efforts in 2024 we surpassed 100% total annual ROAS for the first time in the history of Hotel Hideaway, while also addressing the problem of decreased MAU by targeting a cheaper audience cohort.

To achieve this, I received a lot of help from TikTok’s side and agreed to co-write a case study with them about the business success Sulake had experienced through maximising the use of TikTok’s Creative Challenge system.

Problem

Late 2024, Hotel Hideaway paid user acquisition KPIs started to trend downwards — and none of the new geo-tests, creative tests, ASO tests, bid changes or conversion-event experiments seemed to help in a meaningful way.

Due to none of the previous tests improving performance, I determined the underperformance would either be down to creative quality, platform fatigue, or product. The easiest way for me to have an impact was to experiment using alternative ad platforms, to see if the platform was the issue.

This was not easy to transition to, because almost all of the high-performing creatives were through TikTok’s Creative Challenge system, and thus exclusive to TikTok.

Solution

The first step was determining which alternate channels to try, and platform configuration. Google Ads already had telemetry configured, so out of convenience it made most sense to try that one first.

Where TikTok ran smoothly on TTCC portrait creatives (and I never ran out of quality videos), Google Ads heavily incentivised using multiple video ad resolutions (9:16, 16:9, 1:1) and images to maximise ad-group strength and delivery.

I had access to old Hotel Hideaway ad inventory from before my time, and I had previously determined the ad quality to be very poor by using it on TikTok and comparing it against TTCC creatives. Since the old inventory included ads we had the rights to use on any platform, and some videos were landscape, they would suffice for the initial test.

The initial test was a failure. The CPI on the platform was bad and the returns non-existent. I figured the creatives were the problem, considering they had performed much worse than the TTCC videos ever did. The overall monthly KPIs were still spiralling down and I was under pressure to come up with a solution.

I wanted to know if high-quality UGC ads on Google would be my saving grace. The process of getting these videos was unknown to me outside the TTCC funnel. It would take weeks. It would cost a lot. It would require cross-team communication with finance, legal and marketing, and it wasn’t even guaranteed to work — and I was expected to solve the issue as soon as possible.

After careful consideration, I took a rogue marketing approach.

I downloaded a few of the highest-performing TikTok Creative Challenge videos, uploaded them to YouTube as unlisted videos, and ran a 1-day minimum-budget Google Ads campaign with them. The results were like nothing I had ever seen before. 50% Day 1 ROI. 80% on day 2.

It would be unheard-of to reach 100% ROI even within the first 30 days of a campaign maturing, let alone less than a week. The sample size was small, but the numbers were big enough to justify pursuing this further. Do ends justify means? I had struck gold. Google Ads UGC really was the fix we needed. Now I needed to scale it.

Finding the creators

I needed high-quality ads. A handful of TTCC creators really understood Hotel Hideaway and consistently produced super high-performing videos for us, so it made sense to try to contact them.

TikTok does not reveal the name of the creator to the advertiser. It’s within TikTok’s best interest to keep me — the advertiser — on their platform. But it was no longer mine.

I sharpened my magnifying glass and started a fierce digital-detective research to find out who the creators were. AI + search terms: nothing, just dead ends. Google reverse image search on the creator’s face in one of the videos? I found one of my top 2 creators — a girl COD streamer.

I was still missing a Texan couple that had a giant handheld TV. I wanted these 2 creators specifically, as they had by far the best track record of consistently creating high-performing ads.

I could not find their identity for the longest time — until TikTok released a new creator-marketplace feature: TikTok One. TikTok One was essentially TikTok Creative Challenge rebranded, with a couple of UI changes. One of the changes was that each video had the profile picture of the creator next to it, along with their name blurred out like this: U*********s.

Hallelujah. Right-click image, open in a new tab, search Google for this picture — a match. My second creator I had been chasing was a Minecraft YouTuber. I had identified my high-value creators.

Negotiating rights

I found the creators’ emails/Discord servers through the social media links and sent out inquiries, asking if they would be interested in video commissions. Some weeks later I had received 12 portrait- and 8 landscape-videos.

I collaborated with Azerion’s legal department in creating a legal document where the creator delivers us new landscape videos, grants full non-TikTok usage rights to select previous winning portrait videos, permits us to edit said videos, and assigns perpetual rights to all commercial use.

‘Permits to edit’ could be considered the small print here. What this meant in essence was that I took the deliverables and created 1:1 resolution variants to match Google Ads’ recommendations. Having videos of all desired resolutions helped improve campaign strength. In addition I was able to create more variants by changing the music and by adding text elements to the hook.

Having negotiated permit to edit allowed us to get much more use out of the ad videos, improved testing, and helped greatly in fighting against ad fatigue.

Outcome

Transitioning Hotel Hideaway’s UA efforts to Google Ads elevated the D30 ROAS to an all-time high, while at the same time bringing T1-country user CPI to an all-time low — improving MAU, hitting two bull’s eyes with a single dart.

The goal was reached with the help of commissioned UGC ad creatives. Having negotiated edit permits improved the longevity of the campaign through added ad variants to fight ad fatigue.

Problem

Our user acquisition was ineffective and we did not know why users dropped off. An external partner provided the data visuals. They did not support decision making or development, and they cost six figures a year.

Solution

I used Claude Code and BigQuery to build our own dashboard. It pulled the numbers from every ad platform and from Google Analytics into one place, across all sixteen steps from ad impression to published game.

Dropping paid B2B tools and building them in-house is the trend of 2026.

Outcome

Product and acquisition decisions for the next year came off this data. No more six-figure external bill either. The initial dashboard took a day to build, and any further changes took minutes instead of a week.

How it worked

Every change to the external dashboards took a meeting and one to two weeks. In-house, a change takes minutes.

Almost every click on bitmagic.ai sent a signal to Google Analytics, so we could draw a full picture of how each individual user interacted with the site: how far they got before they bounced, how likely they were to engage with the product, and which parts of the site needed the most refining. We used that to improve conversion rates over a year and several iterations of the product.

None of the underlying data was new. It was the same numbers the ad platforms and GA4 already had. Anything the charts did not answer I asked Claude Code to write the query for, so the dashboard had no fixed set of questions. Asking Claude directly was good for quick one-off questions. For the whole picture the dashboard was easier.

We used BIHMD™ (Bitmagic In-house marketing dashboard) to find real bottlenecks, evaluate landing page A/B tests, and work on the steps with the most friction.

One example: prompting a registration very early in the funnel made a lot more people bounce, but the people who did register were more likely to reach our primary conversion event, submitting a prompt in Bitmagic Creator. Thanks to the insights we kept the login step early, despite it cutting total visitor volume and raising CPC.

I have no single improvement percentage for the whole run. The product changed three times in that period: an Unreal Engine Steam game, then a Unity web embed, then a WebGL app, and finally a Claude and Codex CLI add-on. Their funnels are not comparable to each other. The friction points moved every time, and the tool is what let us find the new ones.

No new data. The same numbers we already had, in one place and in a format we could make decisions on.

Problem

Bitmagic was in closed beta, so we let people in a batch at a time. Every batch took 15 to 30 minutes of manual work from copying user data from Google Sheet, generating unique access codes to each with a GitHub action, creating a CSV, creating a new email list and campaign and sending the codes to the users.

Solution

People applied through a Google Form, which writes into a response sheet. I built one tool that automates everything after that:

• Reads the new applicants and their answers from the response sheet through the Google Sheets API
• Creates a contact for each of them through the Brevo API, with the form answers mapped to contact attributes
• Fires the GitHub Actions workflow that generates a unique access code per person, through a workflow_dispatch call on the GitHub REST API
• Builds the Brevo list, clones the previous campaign as the template, and merges each person’s code in
• Sends the campaign

Three APIs behind one button. Keys live in the environment, not in the tool.

Outcome

15 to 30 minutes of manual work per batch became one click.