Scribd offers access to over a million titles – books, audiobooks, documents, and magazines – through a monthly subscription. One of the leading literature apps in the US and internationally, it had no trouble attracting users. The 30-day free trial did that reliably. The problem was what happened at the end.
Too many iOS and Android users who had genuinely used the product during the trial were not converting to paid. They had opened books, listened to audiobooks, browsed the library, and still left when the free period ended. The value was there. Something in how it was being communicated wasn't landing.
Scribd came to Epom with a specific goal: don't bring in more trial users, help convert the ones already there.
The Gap Between Free and Paid
Free trials create a particular kind of conversion challenge. The user has already committed enough to sign up. They've used the product. The gap you're closing is not awareness or intent; it's the moment of decision when the trial ends and paying feels like a new commitment.
For a subscription app, that gap is where growth lives or dies. Scribd's library was strong. The trial experience was generous. But without the right message, at the right moment, reaching the right user – the conversion didn't follow.
The team needed to understand who was converting and who wasn't, test what messaging moved the needle, and deliver it without making the app feel like it was selling at them.
Building a Foundation With Attribution Data
Before testing anything, Epom integrated with Appsflyer to track installs and user events across iOS and Android. This was the foundation on which everything else would be built.
The integration gave both teams clear visibility into how users moved through the subscription journey: which actions preceded conversion, which engagement patterns predicted drop-off, and where the decision moment actually occurred.
Decisions that had previously been educated guesses became decisions grounded in actual user behavior. That shift changed the quality of everything that followed.
45 Creatives, One Question: What Actually Converts?
With the attribution data in place, the next step was creative. Not one approach tested against another – a genuine search across 45 variations to find what worked.
Epom ran A/B tests on Smart Banner creatives across native, banner, and interstitial placements. The goal wasn't just clicks. It was conversions – finding the specific combination of format, message, and placement that moved a trial user toward a paid subscription.
Forty-five variations is a real commitment to not guessing. Some formats that looked promising in theory underperformed in practice. Some that seemed unremarkable outperformed everything else. The testing surfaced those differences rather than assuming them.
Once the highest-performing variants were identified, they were scaled. Spend moved toward what the data showed was working.
"A/B testing at this scale only produces value if you're willing to act on what it tells you, including killing variants you expected to work. Scribd's team understood that, and it's why the creative phase produced a clear answer rather than just more data."
Machine Learning on Mobile Profiles
Identifying the right creatives was one problem. Delivering them to the right user at the right moment was another.
Epom's machine learning system analyzed Scribd's mobile profiles in real time and optimized impression bidding accordingly. Each user received messaging based on their individual behavior patterns, content preferences, and likelihood to convert – not a segment approximation of those things, but a per-user optimization running continuously.
| Optimization area | Traditional approach | With Epom |
|---|---|---|
| Audience targeting | Broad segments | Real-time, per-user personalization |
| Bid management | Manual adjustments | Automated, continuous |
| Ad relevance | Same message for all | Matched to individual behavior |
| Campaign updates | Periodic | Ongoing |
A user who had been reading legal documents saw different messaging from a user who had spent their trial on audiobooks. The platform made that distinction automatically, without the team having to define every rule in advance.
Within a week, the new approach was already hitting Scribd's target CPA.
Ads That Felt Like Part of the App
Subscription growth driven by intrusive advertising creates a specific problem: users who convert under pressure churn faster. The goal was not just to increase conversions – it was to increase conversions from users who genuinely saw the value of staying.
That meant the creative had to fit the environment. Ads were designed to feel visually integrated into the Scribd experience – unobtrusive, relevant, and aligned with what the user was already doing in the app. A reader deep into a book during their trial shouldn't feel like the app was pivoting to sell at them. They should see something that made paying feel like the natural next step.
The combination of personalized machine learning targeting and carefully integrated creative meant the ads worked without friction. Users engaged with them because they were relevant, not because they were inescapable.
What Three Months of Optimization Produced
The results built quickly and kept building.
Within 40 days, new subscriptions had increased by 25%. By the end of three months, that figure reached 27%. The campaign achieved a 12.5% ROAS across the full period.
| Metric | Result |
|---|---|
| New subscriptions – first 40 days | +25% |
| New subscriptions – 3 months | +27% |
| Creative variations tested | 45 |
| ROAS | 12.5% |
The 40-day number matters because it reflects how quickly the combination of attribution data, creative testing, and real-time optimization began to work. The three-month figure shows it wasn't a spike. It was a sustainable shift.
For a subscription business, sustainable is the number that matters. Anyone can drive a conversion spike. Converting trial users into subscribers who stay – that's a different outcome entirely.
