Level Up Your App Marketing: Measurement, Brand-Building, and Taking Risks — Sherina Khalidi, Deezer

Around 2015, after realizing ads alone would never make the freemium model profitable, Deezer went very far in shutting down features and frustrating free users to push them to subscribe.

Reworked the free experience and how Deezer works with advertisers so free users are profitable on their own, while subscriptions stay the main revenue and freemium is treated as a conversion tool. Also opening the product and sharing features to people on other streaming services.

Indirect monetization through bundle deals, which is how Deezer started (telcos). Each deal is ad hoc, tailored to the partner's business objectives to find a win-win: partners get emotional connection, brand value and access to younger audiences. Examples: Sonos (Deezer's US presence), Mercado Libre in Latin America (a bundle of Disney+ for video and Deezer for audio). Users get a lower package price; Deezer gets reach and retention. Being smaller than Spotify/Apple/Amazon lets Deezer dedicate more resources to partners, who also want to compete with the tech giants.

Massive TV and out-of-home campaigns in France whose messaging only reinforces brand positioning (everything is about living the music; the user at the core), with no price, trial or download call to action. No immediate ROI expected; measured with brand trackers, post-tests and quantitative/qualitative studies. The performance team runs campaigns that amplify whatever the brand team is doing to give users who are ready the final push to download and subscribe.

Uses several 'sources of truth' and looks for them to agree. Brand: every quarter/semester a panel is asked if Deezer knows music, understands them, feels close, and especially whether they'd consider paying; the upward trend matters more than the level. Performance: last-click attribution via MMP, a multi-touch model in progress (biased by web vs app tracking differences), and as many incrementality tests as possible: geo lift (shut down matched regions, with Google Ads tooling), Meta conversion lift, and brand lift with control vs exposed groups. In-house data science builds a media mix model (spend vs daily/weekly traffic, baseline, offline and YouTube brand campaigns). Next step, inspired by L'Oréal: turn incrementality results into correction scores for over-attributed (e.g. search) and under-attributed (e.g. social) channels and feed them into the MMM. Considering a hybrid with an external consultant auditing the MMM.

Runs upper-funnel digital campaigns optimized for installs (not paying subs) or for reach and frequency, and uses incrementality tests to find the right investment level and their effect on conversion campaigns. Tests showed these campaigns lift the performance campaigns far more than expected, so investment continues despite no short-term measurable return. Brand campaigns are not judged by on/off switching because their effect is slow.

Team is expected to test constantly (new channels, audiences, campaigns), report failures honestly and extract learnings; failure is never blamed, not trying is. Tests are tightly framed so total risk is limited, and success is judged on the overall LTV:CAC: as long as the whole program is profitable, losing money on individual tests is acceptable. Year-end review compares biggest fails with biggest wins; wins always outweighed fails. Required countering a French cultural fear of failure.

The logged-out ('unlocked') homepage was unclear and nobody liked it. Built two alternatives: a branding page with key brand and product messages, and a player-like page with limited features that prompted subscription when you tried to stream. ABC-tested them against the control; neither beat it. Because the control didn't convey the desired messages and wasn't user-friendly, the team rolled out the branded version anyway.

The equalizer logo (2007-2023) had weak brand attribution. Asked users to draw the current logo to check recall, and tested the existing logo against two alternatives in studies. The purple heart, which the team doubted because it was so disruptive (maybe like a dating app), won by far across all target audiences and KPIs, and was adopted in 2023.