This thesis investigates the marketing mix of an indie video game with a specific emphasis on influencer marketing, addressing the measurement challenge faced by small studios that operate with limited budgets and incomplete user-level tracking. The study’s objective is to quantify the incremental contribution of major pre-launch channels to demand formation, using daily Steam wishlist additions (“Adds”) as the key performance indicator, observed over a 160-day period. A time-series Marketing Mix Model (MMM) is developed using a log-linear regression framework in which daily Adds are explained by time-varying marketing exposures across five channels (YouTube, Twitch, Reddit, TikTok, and Press/Media). To reflect carryover dynamics, exposures are transformed using adstock, and the specification includes controls for a baseline time trend, day-of-week seasonality, and two high-impact product beats (demo release and playtest). Estimation relies on Newey–West (HAC) standard errors to account for autocorrelation and heteroskedasticity typical of daily marketing time series. Results from the aggregated MMM show positive and statistically significant elasticities for YouTube (β≈0.09), Twitch (β≈0.10), Reddit (β≈0.10), and Press/Media (β≈0.12), while TikTok exhibits a negative coefficient (β≈−0.09) that is interpreted cautiously as likely reflecting timing and collinearity rather than a true negative causal effect. Modeled attribution highlights Press/Media and YouTube as the largest explained shares of incremental Adds. Product events dominate short-term dynamics: the demo release is associated with an approximately 7.5× multiplicative uplift versus baseline, and the playtest with roughly +33%. Model comparisons reveal the classic trade-off between granularity and robustness: an influencer-level decomposition substantially improves in-sample fit but fails to generalize (overfitting), whereas congruence-based segmentation models improve holdout performance. However, the congruence analysis finds no statistically significant differences in effectiveness between “fit” and “non-fit” influencer groups (indie-fit and genre-fit), suggesting that, within the gaming domain, broader influencer outreach can be as effective as niche-aligned targeting for wishlist conversion.
Questa tesi analizza il marketing mix di un videogioco indie, con un focus specifico sul ruolo dell’influencer marketing, in un contesto in cui i budget sono limitati e la misurazione user-level è spesso impraticabile. L’obiettivo è stimare il contributo incrementale dei principali canali digitali nella fase pre-lancio, utilizzando come metrica di outcome le aggiunte giornaliere alla wishlist su Steam (Adds), osservate su un orizzonte di 160 giorni. La ricerca adotta un Marketing Mix Model (MMM) basato su regressione log-lineare su serie storiche, in cui le Adds sono spiegate da esposizioni time-varying su cinque canali (YouTube, Twitch, Reddit, TikTok e Press/Media), opportunamente trasformate tramite adstock per rappresentare effetti di carryover, e controllate per trend temporale, stagionalità settimanale e due eventi ad alto impatto (rilascio della demo e playtest). La stima utilizza errori standard Newey–West (HAC) per gestire autocorrelazione e eteroschedasticità. I risultati del modello aggregato indicano effetti positivi e significativi per YouTube (β≈0,09), Twitch (β≈0,10), Reddit (β≈0,10) e Press/Media (β≈0,12), mentre TikTok mostra un coefficiente negativo (β≈−0,09), da interpretare con cautela. In termini di attribuzione modellata, Press/Media e YouTube rappresentano le quote maggiori dell’incremento spiegato. Inoltre, gli eventi prodotto risultano determinanti: il rilascio della demo genera un uplift moltiplicativo di circa 7,5× rispetto al baseline, e il playtest un incremento di circa +33%. Confrontando specifiche alternative, la decomposizione “influencer-level” migliora il fit in-sample ma soffre di overfitting, mentre modelli che segmentano YouTube per congruenza influencer–prodotto migliorano la generalizzazione out-of-sample senza però evidenziare differenze statisticamente robuste tra influencer “fit” e “non-fit”. Nel complesso, la tesi mostra la fattibilità di un MMM per campagne indie e propone indicazioni operative su allocazione canali e pianificazione di burst coordinati attorno a beat di prodotto.
Marketing mix modelling for the gaming industry: quantifying influencer and earned-media effects on steam wishlist formation
KALEEV, ARTEM
2025/2026
Abstract
This thesis investigates the marketing mix of an indie video game with a specific emphasis on influencer marketing, addressing the measurement challenge faced by small studios that operate with limited budgets and incomplete user-level tracking. The study’s objective is to quantify the incremental contribution of major pre-launch channels to demand formation, using daily Steam wishlist additions (“Adds”) as the key performance indicator, observed over a 160-day period. A time-series Marketing Mix Model (MMM) is developed using a log-linear regression framework in which daily Adds are explained by time-varying marketing exposures across five channels (YouTube, Twitch, Reddit, TikTok, and Press/Media). To reflect carryover dynamics, exposures are transformed using adstock, and the specification includes controls for a baseline time trend, day-of-week seasonality, and two high-impact product beats (demo release and playtest). Estimation relies on Newey–West (HAC) standard errors to account for autocorrelation and heteroskedasticity typical of daily marketing time series. Results from the aggregated MMM show positive and statistically significant elasticities for YouTube (β≈0.09), Twitch (β≈0.10), Reddit (β≈0.10), and Press/Media (β≈0.12), while TikTok exhibits a negative coefficient (β≈−0.09) that is interpreted cautiously as likely reflecting timing and collinearity rather than a true negative causal effect. Modeled attribution highlights Press/Media and YouTube as the largest explained shares of incremental Adds. Product events dominate short-term dynamics: the demo release is associated with an approximately 7.5× multiplicative uplift versus baseline, and the playtest with roughly +33%. Model comparisons reveal the classic trade-off between granularity and robustness: an influencer-level decomposition substantially improves in-sample fit but fails to generalize (overfitting), whereas congruence-based segmentation models improve holdout performance. However, the congruence analysis finds no statistically significant differences in effectiveness between “fit” and “non-fit” influencer groups (indie-fit and genre-fit), suggesting that, within the gaming domain, broader influencer outreach can be as effective as niche-aligned targeting for wishlist conversion.| File | Dimensione | Formato | |
|---|---|---|---|
|
kaleev thesis final.pdf
accessibile in internet per tutti
Dimensione
634.02 kB
Formato
Adobe PDF
|
634.02 kB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/252995