[{"data":1,"prerenderedAt":808},["ShallowReactive",2],{"article-alternates":3,"article-\u002Fde\u002Fgaming\u002Fbayesian-fiyat-optimizasyonu-mobilf2p":12},{"i18nKey":4,"paths":5},"gaming-002-2026-07",{"de":6,"en":7,"es":8,"fr":9,"it":10,"ru":11},"\u002Fde\u002Fgaming\u002Fbayesian-preisoptimierung-mobile-f2p","\u002Fen\u002Fgaming\u002Fbayesian-price-optimization-mobile-f2p","\u002Fes\u002Fgaming\u002Foptimizacion-bayesiana-de-precios-f2p-mobile","\u002Ffr\u002Fgaming\u002Foptimisation-tarifaire-bayesienne-f2p-mobile","\u002Fit\u002Fgaming\u002Fbayesian-price-optimization-mobile-f2p","\u002Fru\u002Fgaming\u002Fbayesovskaya-optimizatsiya-tseny-v-mobilnom-f2p",{"_path":13,"_dir":14,"_draft":15,"_partial":15,"_locale":16,"title":17,"description":18,"publishedAt":19,"modifiedAt":19,"category":14,"i18nKey":4,"tags":20,"readingTime":26,"author":27,"body":28,"_type":802,"_id":803,"_source":804,"_file":805,"_stem":806,"_extension":807},"\u002Fde\u002Fgaming\u002Fbayesian-fiyat-optimizasyonu-mobilf2p","gaming",false,"","Mobile F2P'de Bayesian-Preisoptimierung","IAP-Preislisten mit Posterior-Estimation und Segment-basierter Modellierung optimieren. Datengesteuerte Preisstrategien für Mobile Games.","2026-07-22",[21,22,23,24,25],"f2p-monetisierung","bayesian-optimierung","iap-preisgestaltung","mobile-gaming","datengesteuerte-preisgestaltung",9,"Roibase",{"type":29,"children":30,"toc":791},"root",[31,39,46,51,56,61,67,94,99,104,111,116,122,136,141,162,618,623,629,648,653,658,664,669,674,723,729,734,739,744,750,755,760,776,780,785],{"type":32,"tag":33,"props":34,"children":35},"element","p",{},[36],{"type":37,"value":38},"text","In mobilen F2P-Spielen basieren Preisgestaltungsentscheidungen häufig auf Vermutungen oder \"am Markt üblichen Referenzpreisen\". $0,99 Starter-Paket, $4,99 Mid-Tier, $99,99 Whale-Bundle — diese Preisstaffeln sind in den meisten Spielen statisch. Dabei unterscheiden sich Cohort-Struktur, Geo-Mix und Value Perception jedes Spiels. Bayesian Price Optimization modelliert diese Unterschiede über Posterior-Wahrscheinlichkeitsverteilungen und findet für jedes Segment den optimalen Preispunkt. Statt klassischer A\u002FB-Tests kann ein kontinuierlich lernendes System Ihre IAP-Conversion-Rate um 15–40 % verbessern.",{"type":32,"tag":40,"props":41,"children":43},"h2",{"id":42},"warum-der-bayesian-ansatz-ab-tests-überlegen-ist",[44],{"type":37,"value":45},"Warum der Bayesian-Ansatz A\u002FB-Tests überlegen ist",{"type":32,"tag":33,"props":47,"children":48},{},[49],{"type":37,"value":50},"Der klassische A\u002FB-Test arbeitet mit einer fixen Hypothese: $4,99 vs. $5,99 werden verglichen, man wartet auf 95 % Konfidenz und wählt den Gewinner. Dieses Vorgehen hat zwei Probleme: Erstens wird der Traffic während des Tests halbiert, und die schlecht performende Variante wird weiterhin Nutzern gezeigt (Opportunitätskosten). Zweitens erfährt man nach Test-Ende nur \"A oder B\" — über Zwischenwerte oder Segment-spezifische Unterschiede lernt man nichts.",{"type":32,"tag":33,"props":52,"children":53},{},[54],{"type":37,"value":55},"Bayesian Optimization startet mit einer Prior Distribution (z. B. \"Preis zwischen $3–$7, gleichmäßig verteilt\"), fügt jedes Conversion-Ereignis zur Posterior hinzu und aktualisiert die Wahrscheinlichkeitsverteilung kontinuierlich. Thompson Sampling und ähnliche Algorithmen lenken Traffic dynamisch zur gewinnenden Variante — während des Tests wird die Gesamtrevenue maximiert. Bei einem 10-tägigen Test produziert der Bayesian-Ansatz 8–12 % mehr Revenue, weil schlechte Preispunkte nur minimalen Traffic erhalten.",{"type":32,"tag":33,"props":57,"children":58},{},[59],{"type":37,"value":60},"Zudem liefert das Bayesian-Modell nicht nur \"welcher Preis gewinnt\", sondern auch \"dieser Preis ist mit 87 % Wahrscheinlichkeit optimal\". Diese Konfidenz-Intervalle beschleunigen Iterationen: Bei 60 % Konfidenz können Sie einen Preis bereits live nehmen und einen neuen Test starten, da die Posterior bereits ausreichend Information trägt.",{"type":32,"tag":40,"props":62,"children":64},{"id":63},"segment-basierte-prior-konstruktion-bei-iap-preisstaffeln",[65],{"type":37,"value":66},"Segment-basierte Prior-Konstruktion bei IAP-Preisstaffeln",{"type":32,"tag":33,"props":68,"children":69},{},[70,72,78,80,85,87,92],{"type":37,"value":71},"In F2P-Spielen sind nicht alle Nutzer gleich. Die korrekte Definition von Spender-Segmenten verstärkt die Prior des Bayesian-Modells. Typische Segmentierung: ",{"type":32,"tag":73,"props":74,"children":75},"strong",{},[76],{"type":37,"value":77},"Minnows",{"type":37,"value":79}," (Lifetime Spend \u003C$10), ",{"type":32,"tag":73,"props":81,"children":82},{},[83],{"type":37,"value":84},"Dolphins",{"type":37,"value":86}," ($10–$100), ",{"type":32,"tag":73,"props":88,"children":89},{},[90],{"type":37,"value":91},"Whales",{"type":37,"value":93}," (>$100). Jedes Segment hat unterschiedliche Preiselastizität — Minnows konvertieren selbst bei $0,99, Whales kaufen $99,99-Bundles ohne Preisbeachtung.",{"type":32,"tag":33,"props":95,"children":96},{},[97],{"type":37,"value":98},"Um die Prior-Distribution segment-basiert aufzubauen, benötigen Sie historische Daten. Wenn in Ihrem Minnow-Segment die durchschnittliche IAP-Conversion zwischen $0,99 und $1,99 bei 3,2 % liegt, verwenden Sie $1,49 als Prior-Mean und $0,50 als Sigma (unter Normalverteilungsannahme). Im Whale-Segment bleibt die Conversion zwischen $49,99–$149,99 nahezu flach — hier ist eine Uniform Prior sinnvoller, um die Hypothese \"Whales sind preisunempfindlich\" im Modell abzubilden.",{"type":32,"tag":33,"props":100,"children":101},{},[102],{"type":37,"value":103},"Der Vorteil einer segment-basierten Prior ist, dass Cross-Segment-Lernen verhindert wird. Der klassische A\u002FB-Test mischt alle Nutzer in einem Pool, und die hohe Conversion von Whales in der niedrigpreisigen Variante kann den optimalen Minnow-Preis überlagern. Das Bayesian-Modell aktualisiert jedes Segment getrennt, sodass $1,49 für Minnows und $79,99 für Whales als segment-optimale Preise entstehen.",{"type":32,"tag":105,"props":106,"children":108},"h3",{"id":107},"geo-spezifische-prior-anpassung",[109],{"type":37,"value":110},"Geo-spezifische Prior-Anpassung",{"type":32,"tag":33,"props":112,"children":113},{},[114],{"type":37,"value":115},"Tier-1-Länder (US, UK, JP) und Emerging Markets (BR, TR, IN) haben massiv unterschiedliche Kaufkraft. In den USA wirkt $4,99 \"günstig\", während der gleiche Betrag (₺150 in der Türkei) als Mittelklasse-Preis wahrgenommen wird. Um Prior-Distributionen geo-basiert zu normalisieren, nutzen Sie lokale ARPU-Daten. Wenn US-Durchschnitt $0,42 täglich ist und TR $0,18, skalieren Sie die Prior-Mean (0,18\u002F0,42 = 43 %) entsprechend. Das Modell testet dann dieselbe relative Preisstaffel in jeder Geo, mit Absolutwert-Unterschieden in der Prior.",{"type":32,"tag":40,"props":117,"children":119},{"id":118},"posterior-estimation-und-thompson-sampling",[120],{"type":37,"value":121},"Posterior Estimation und Thompson Sampling",{"type":32,"tag":33,"props":123,"children":124},{},[125,127,134],{"type":37,"value":126},"Das Runtime-Engine des Bayesian-Modells ist die Posterior Estimation. Bei jedem IAP-Impression wird ein Sample aus der aktuellen Posterior Distribution gezogen (z. B. mit ",{"type":32,"tag":128,"props":129,"children":131},"code",{"className":130},[],[132],{"type":37,"value":133},"np.random.beta(alpha, beta)",{"type":37,"value":135}," bei Beta Distribution). Der entsprechende Preis wird dem Nutzer gezeigt. Bei Kauf wird alpha += 1, bei Skip beta += 1 — die Posterior wird aktualisiert.",{"type":32,"tag":33,"props":137,"children":138},{},[139],{"type":37,"value":140},"Thompson Sampling nutzt diesen Mechanismus für Traffic-Verteilung. Für jede Variante wird ein Reward-Expectation aus der Posterior gezogen; die höchste Reward gewinnt. In den ersten Tagen erhalten alle Varianten gleichen Traffic (Exploration), später konzentriert sich Traffic auf die Gewinner-Variante (Exploitation). Das Balance wird nicht durch Epsilon, sondern durch Posterior-Varianz gesteuert — Varianten mit niedriger Varianz (hohe Konfidenz) erhalten mehr Traffic.",{"type":32,"tag":33,"props":142,"children":143},{},[144,146,152,154,160],{"type":37,"value":145},"Für praktische Implementierung können Sie ",{"type":32,"tag":128,"props":147,"children":149},{"className":148},[],[150],{"type":37,"value":151},"scipy.stats.beta",{"type":37,"value":153}," oder ",{"type":32,"tag":128,"props":155,"children":157},{"className":156},[],[158],{"type":37,"value":159},"pymc3",{"type":37,"value":161}," nutzen. Ein simpler Code-Block:",{"type":32,"tag":163,"props":164,"children":168},"pre",{"className":165,"code":166,"language":167,"meta":16,"style":16},"language-python shiki shiki-themes github-dark","import numpy as np\nfrom scipy.stats import beta\n\n# Prior: alpha=1, beta=1 (uniform)\nalpha_a, beta_a = 1, 1  # Variante A ($4,99)\nalpha_b, beta_b = 1, 1  # Variante B ($5,99)\n\ndef select_variant():\n    sample_a = np.random.beta(alpha_a, beta_a)\n    sample_b = np.random.beta(alpha_b, beta_b)\n    return \"A\" if sample_a > sample_b else \"B\"\n\ndef update_posterior(variant, converted):\n    global alpha_a, beta_a, alpha_b, beta_b\n    if variant == \"A\":\n        if converted:\n            alpha_a += 1\n        else:\n            beta_a += 1\n    else:\n        if converted:\n            alpha_b += 1\n        else:\n            beta_b += 1\n","python",[169],{"type":32,"tag":128,"props":170,"children":171},{"__ignoreMap":16},[172,200,223,233,243,278,308,316,336,353,371,416,424,442,456,484,498,517,530,547,560,572,589,601],{"type":32,"tag":173,"props":174,"children":177},"span",{"class":175,"line":176},"line",1,[178,184,190,195],{"type":32,"tag":173,"props":179,"children":181},{"style":180},"--shiki-default:#F97583",[182],{"type":37,"value":183},"import",{"type":32,"tag":173,"props":185,"children":187},{"style":186},"--shiki-default:#E1E4E8",[188],{"type":37,"value":189}," numpy ",{"type":32,"tag":173,"props":191,"children":192},{"style":180},[193],{"type":37,"value":194},"as",{"type":32,"tag":173,"props":196,"children":197},{"style":186},[198],{"type":37,"value":199}," np\n",{"type":32,"tag":173,"props":201,"children":203},{"class":175,"line":202},2,[204,209,214,218],{"type":32,"tag":173,"props":205,"children":206},{"style":180},[207],{"type":37,"value":208},"from",{"type":32,"tag":173,"props":210,"children":211},{"style":186},[212],{"type":37,"value":213}," scipy.stats ",{"type":32,"tag":173,"props":215,"children":216},{"style":180},[217],{"type":37,"value":183},{"type":32,"tag":173,"props":219,"children":220},{"style":186},[221],{"type":37,"value":222}," beta\n",{"type":32,"tag":173,"props":224,"children":226},{"class":175,"line":225},3,[227],{"type":32,"tag":173,"props":228,"children":230},{"emptyLinePlaceholder":229},true,[231],{"type":37,"value":232},"\n",{"type":32,"tag":173,"props":234,"children":236},{"class":175,"line":235},4,[237],{"type":32,"tag":173,"props":238,"children":240},{"style":239},"--shiki-default:#6A737D",[241],{"type":37,"value":242},"# Prior: alpha=1, beta=1 (uniform)\n",{"type":32,"tag":173,"props":244,"children":246},{"class":175,"line":245},5,[247,252,257,263,268,273],{"type":32,"tag":173,"props":248,"children":249},{"style":186},[250],{"type":37,"value":251},"alpha_a, beta_a ",{"type":32,"tag":173,"props":253,"children":254},{"style":180},[255],{"type":37,"value":256},"=",{"type":32,"tag":173,"props":258,"children":260},{"style":259},"--shiki-default:#79B8FF",[261],{"type":37,"value":262}," 1",{"type":32,"tag":173,"props":264,"children":265},{"style":186},[266],{"type":37,"value":267},", ",{"type":32,"tag":173,"props":269,"children":270},{"style":259},[271],{"type":37,"value":272},"1",{"type":32,"tag":173,"props":274,"children":275},{"style":239},[276],{"type":37,"value":277},"  # Variante A ($4,99)\n",{"type":32,"tag":173,"props":279,"children":281},{"class":175,"line":280},6,[282,287,291,295,299,303],{"type":32,"tag":173,"props":283,"children":284},{"style":186},[285],{"type":37,"value":286},"alpha_b, beta_b ",{"type":32,"tag":173,"props":288,"children":289},{"style":180},[290],{"type":37,"value":256},{"type":32,"tag":173,"props":292,"children":293},{"style":259},[294],{"type":37,"value":262},{"type":32,"tag":173,"props":296,"children":297},{"style":186},[298],{"type":37,"value":267},{"type":32,"tag":173,"props":300,"children":301},{"style":259},[302],{"type":37,"value":272},{"type":32,"tag":173,"props":304,"children":305},{"style":239},[306],{"type":37,"value":307},"  # Variante B ($5,99)\n",{"type":32,"tag":173,"props":309,"children":311},{"class":175,"line":310},7,[312],{"type":32,"tag":173,"props":313,"children":314},{"emptyLinePlaceholder":229},[315],{"type":37,"value":232},{"type":32,"tag":173,"props":317,"children":319},{"class":175,"line":318},8,[320,325,331],{"type":32,"tag":173,"props":321,"children":322},{"style":180},[323],{"type":37,"value":324},"def",{"type":32,"tag":173,"props":326,"children":328},{"style":327},"--shiki-default:#B392F0",[329],{"type":37,"value":330}," select_variant",{"type":32,"tag":173,"props":332,"children":333},{"style":186},[334],{"type":37,"value":335},"():\n",{"type":32,"tag":173,"props":337,"children":338},{"class":175,"line":26},[339,344,348],{"type":32,"tag":173,"props":340,"children":341},{"style":186},[342],{"type":37,"value":343},"    sample_a ",{"type":32,"tag":173,"props":345,"children":346},{"style":180},[347],{"type":37,"value":256},{"type":32,"tag":173,"props":349,"children":350},{"style":186},[351],{"type":37,"value":352}," np.random.beta(alpha_a, beta_a)\n",{"type":32,"tag":173,"props":354,"children":356},{"class":175,"line":355},10,[357,362,366],{"type":32,"tag":173,"props":358,"children":359},{"style":186},[360],{"type":37,"value":361},"    sample_b ",{"type":32,"tag":173,"props":363,"children":364},{"style":180},[365],{"type":37,"value":256},{"type":32,"tag":173,"props":367,"children":368},{"style":186},[369],{"type":37,"value":370}," np.random.beta(alpha_b, beta_b)\n",{"type":32,"tag":173,"props":372,"children":374},{"class":175,"line":373},11,[375,380,386,391,396,401,406,411],{"type":32,"tag":173,"props":376,"children":377},{"style":180},[378],{"type":37,"value":379},"    return",{"type":32,"tag":173,"props":381,"children":383},{"style":382},"--shiki-default:#9ECBFF",[384],{"type":37,"value":385}," \"A\"",{"type":32,"tag":173,"props":387,"children":388},{"style":180},[389],{"type":37,"value":390}," if",{"type":32,"tag":173,"props":392,"children":393},{"style":186},[394],{"type":37,"value":395}," sample_a ",{"type":32,"tag":173,"props":397,"children":398},{"style":180},[399],{"type":37,"value":400},">",{"type":32,"tag":173,"props":402,"children":403},{"style":186},[404],{"type":37,"value":405}," sample_b ",{"type":32,"tag":173,"props":407,"children":408},{"style":180},[409],{"type":37,"value":410},"else",{"type":32,"tag":173,"props":412,"children":413},{"style":382},[414],{"type":37,"value":415}," \"B\"\n",{"type":32,"tag":173,"props":417,"children":419},{"class":175,"line":418},12,[420],{"type":32,"tag":173,"props":421,"children":422},{"emptyLinePlaceholder":229},[423],{"type":37,"value":232},{"type":32,"tag":173,"props":425,"children":427},{"class":175,"line":426},13,[428,432,437],{"type":32,"tag":173,"props":429,"children":430},{"style":180},[431],{"type":37,"value":324},{"type":32,"tag":173,"props":433,"children":434},{"style":327},[435],{"type":37,"value":436}," update_posterior",{"type":32,"tag":173,"props":438,"children":439},{"style":186},[440],{"type":37,"value":441},"(variant, converted):\n",{"type":32,"tag":173,"props":443,"children":445},{"class":175,"line":444},14,[446,451],{"type":32,"tag":173,"props":447,"children":448},{"style":180},[449],{"type":37,"value":450},"    global",{"type":32,"tag":173,"props":452,"children":453},{"style":186},[454],{"type":37,"value":455}," alpha_a, beta_a, alpha_b, beta_b\n",{"type":32,"tag":173,"props":457,"children":459},{"class":175,"line":458},15,[460,465,470,475,479],{"type":32,"tag":173,"props":461,"children":462},{"style":180},[463],{"type":37,"value":464},"    if",{"type":32,"tag":173,"props":466,"children":467},{"style":186},[468],{"type":37,"value":469}," variant ",{"type":32,"tag":173,"props":471,"children":472},{"style":180},[473],{"type":37,"value":474},"==",{"type":32,"tag":173,"props":476,"children":477},{"style":382},[478],{"type":37,"value":385},{"type":32,"tag":173,"props":480,"children":481},{"style":186},[482],{"type":37,"value":483},":\n",{"type":32,"tag":173,"props":485,"children":487},{"class":175,"line":486},16,[488,493],{"type":32,"tag":173,"props":489,"children":490},{"style":180},[491],{"type":37,"value":492},"        if",{"type":32,"tag":173,"props":494,"children":495},{"style":186},[496],{"type":37,"value":497}," converted:\n",{"type":32,"tag":173,"props":499,"children":501},{"class":175,"line":500},17,[502,507,512],{"type":32,"tag":173,"props":503,"children":504},{"style":186},[505],{"type":37,"value":506},"            alpha_a ",{"type":32,"tag":173,"props":508,"children":509},{"style":180},[510],{"type":37,"value":511},"+=",{"type":32,"tag":173,"props":513,"children":514},{"style":259},[515],{"type":37,"value":516}," 1\n",{"type":32,"tag":173,"props":518,"children":520},{"class":175,"line":519},18,[521,526],{"type":32,"tag":173,"props":522,"children":523},{"style":180},[524],{"type":37,"value":525},"        else",{"type":32,"tag":173,"props":527,"children":528},{"style":186},[529],{"type":37,"value":483},{"type":32,"tag":173,"props":531,"children":533},{"class":175,"line":532},19,[534,539,543],{"type":32,"tag":173,"props":535,"children":536},{"style":186},[537],{"type":37,"value":538},"            beta_a ",{"type":32,"tag":173,"props":540,"children":541},{"style":180},[542],{"type":37,"value":511},{"type":32,"tag":173,"props":544,"children":545},{"style":259},[546],{"type":37,"value":516},{"type":32,"tag":173,"props":548,"children":550},{"class":175,"line":549},20,[551,556],{"type":32,"tag":173,"props":552,"children":553},{"style":180},[554],{"type":37,"value":555},"    else",{"type":32,"tag":173,"props":557,"children":558},{"style":186},[559],{"type":37,"value":483},{"type":32,"tag":173,"props":561,"children":563},{"class":175,"line":562},21,[564,568],{"type":32,"tag":173,"props":565,"children":566},{"style":180},[567],{"type":37,"value":492},{"type":32,"tag":173,"props":569,"children":570},{"style":186},[571],{"type":37,"value":497},{"type":32,"tag":173,"props":573,"children":575},{"class":175,"line":574},22,[576,581,585],{"type":32,"tag":173,"props":577,"children":578},{"style":186},[579],{"type":37,"value":580},"            alpha_b ",{"type":32,"tag":173,"props":582,"children":583},{"style":180},[584],{"type":37,"value":511},{"type":32,"tag":173,"props":586,"children":587},{"style":259},[588],{"type":37,"value":516},{"type":32,"tag":173,"props":590,"children":592},{"class":175,"line":591},23,[593,597],{"type":32,"tag":173,"props":594,"children":595},{"style":180},[596],{"type":37,"value":525},{"type":32,"tag":173,"props":598,"children":599},{"style":186},[600],{"type":37,"value":483},{"type":32,"tag":173,"props":602,"children":604},{"class":175,"line":603},24,[605,610,614],{"type":32,"tag":173,"props":606,"children":607},{"style":186},[608],{"type":37,"value":609},"            beta_b ",{"type":32,"tag":173,"props":611,"children":612},{"style":180},[613],{"type":37,"value":511},{"type":32,"tag":173,"props":615,"children":616},{"style":259},[617],{"type":37,"value":516},{"type":32,"tag":33,"props":619,"children":620},{},[621],{"type":37,"value":622},"Diese simple Loop konvergiert nach 10.000 Impressionen gegen die echte Conversion-Rate mit einer Fehlerquote von ~2 % (unter korrekter Beta-Prior-Annahme). In Production aktualisieren Sie BigQuery + Airflow täglich die Posterior-Parameter und starten neue Cohorts mit der aktualisierten Distribution.",{"type":32,"tag":40,"props":624,"children":626},{"id":625},"multi-armed-bandit-vs-vollständiges-bayesian-modell",[627],{"type":37,"value":628},"Multi-Armed Bandit vs. vollständiges Bayesian-Modell",{"type":32,"tag":33,"props":630,"children":631},{},[632,634,639,641,646],{"type":37,"value":633},"In der Bayesian-Preisoptimierungs-Literatur gibt es zwei Hauptansätze: ",{"type":32,"tag":73,"props":635,"children":636},{},[637],{"type":37,"value":638},"Multi-Armed Bandit (MAB)",{"type":37,"value":640}," und ",{"type":32,"tag":73,"props":642,"children":643},{},[644],{"type":37,"value":645},"vollständige Bayesian-Regression",{"type":37,"value":647},". Der MAB-Ansatz ist das oben beschriebene Thompson Sampling — diskrete Preisvarianten (z. B. 5 Preispunkte) werden als Arms definiert, die Posterior wird für jeden Arm separat geführt. Vorteil: Implementierung ist einfach, Runtime ist leicht, Echtzeitentscheidungen sind möglich.",{"type":32,"tag":33,"props":649,"children":650},{},[651],{"type":37,"value":652},"Vollständige Bayesian-Regression modelliert den Preis als kontinuierliche Variable und bindet die Conversion Probability durch logistische Regression oder Gaussian Processes an den Preis. Dieser Ansatz ist flexibler — z. B. kann er nicht-lineare Beziehungen wie \"Conversion fällt exponentiell mit Preis\" lernen. Nachteil: Model Training erfordert BigQuery + Python Stack, Echtzeitentscheidungen sind nicht möglich (Batch Prediction).",{"type":32,"tag":33,"props":654,"children":655},{},[656],{"type":37,"value":657},"Für F2P-Spiele reicht MAB normalerweise aus, da die Preisstaffel ohnehin diskret ist ($0,99, $2,99, $4,99, $9,99). Vollständige Bayesian-Regression kommt zum Einsatz, wenn Sie Dynamic Pricing machen möchten (unterschiedliche Preise pro Nutzer) — aber das wird von den meisten App-Store-Richtlinien als Diskriminierung unterbunden. Mittelweg: MAB pro Segment, innerhalb jedes Segments vollständige Bayesian-Regression. So finden Sie für das Whale-Segment kontinuierlich den optimalen Punkt zwischen $79,99–$149,99.",{"type":32,"tag":40,"props":659,"children":661},{"id":660},"revenue-uplift-und-cohort-ltv-effekt",[662],{"type":37,"value":663},"Revenue-Uplift und Cohort-LTV-Effekt",{"type":32,"tag":33,"props":665,"children":666},{},[667],{"type":37,"value":668},"Der echte ROI der Bayesian-Preisoptimierung zeigt sich im Cohort-LTV. Die Conversion-Rate steigt in der ersten Woche um 8 %, aber das D30-LTV dieser Nutzer ist 15–20 % höher. Warum? Der optimale Preispunkt sitzt exakt auf dem Value Perception des Nutzers — weder zu niedrig (Wertabschwächung) noch zu hoch (Reibung). Diese Nutzer kaufen nach dem ersten IAP mit höherer Wahrscheinlichkeit das zweite Paket.",{"type":32,"tag":33,"props":670,"children":671},{},[672],{"type":37,"value":673},"Beispiel: Ein Mid-Core-RPG ändert seinen $4,99-Starter-Pack durch Bayesian-Modell auf $3,49 (Minnow-Segment, US-Geo). Die Conversion steigt in Woche 1 von 22 % auf 28 % (+27 % relativ). D7-Retention bleibt gleich (42 %), aber D30-ARPU steigt von $2,18 auf $2,51 (+15 %). Warum? Der $3,49-Preis senkt die Hürde \"Ich kann in dieses Spiel investieren\", Second-Purchase-Friction sinkt. Gesamt-Cohort-LTV steigt von $8,90 auf $10,20 (+15 %).",{"type":32,"tag":33,"props":675,"children":676},{},[677,679,685,686,692,693,699,700,706,707,713,715,721],{"type":37,"value":678},"Zur Messung ist Cohort-Analyse erforderlich. In BigQuery tracken Sie ",{"type":32,"tag":128,"props":680,"children":682},{"className":681},[],[683],{"type":37,"value":684},"user_id",{"type":37,"value":267},{"type":32,"tag":128,"props":687,"children":689},{"className":688},[],[690],{"type":37,"value":691},"install_date",{"type":37,"value":267},{"type":32,"tag":128,"props":694,"children":696},{"className":695},[],[697],{"type":37,"value":698},"first_iap_price",{"type":37,"value":267},{"type":32,"tag":128,"props":701,"children":703},{"className":702},[],[704],{"type":37,"value":705},"d7_revenue",{"type":37,"value":267},{"type":32,"tag":128,"props":708,"children":710},{"className":709},[],[711],{"type":37,"value":712},"d30_revenue",{"type":37,"value":714},". Flag den Bayesian-Test-Variant als ",{"type":32,"tag":128,"props":716,"children":718},{"className":717},[],[719],{"type":37,"value":720},"experiment_group",{"type":37,"value":722},", vergleichen Sie LTV-Kurven mit Kontrollgruppe. Significance-Tests in den ersten 7 Tagen sind früh; D30 gibt Konfidenz.",{"type":32,"tag":40,"props":724,"children":726},{"id":725},"missverständnisse-und-tradeoffs",[727],{"type":37,"value":728},"Missverständnisse und Tradeoffs",{"type":32,"tag":33,"props":730,"children":731},{},[732],{"type":37,"value":733},"Das Missverständnis \"Bayesian Optimization gewinnt sofort\" ist verbreitet. Realität: Posterior Convergence benötigt mindestens 5.000–10.000 Impressionen pro Segment. Bei niedrig-Traffic-Spielen (DAU \u003C50k) dauert der Test 4–6 Wochen. Während dieser Zeit muss die Data Pipeline (Impression Logging, Conversion Tracking, Posterior Update) stabil laufen — ein einzelner Bug zerstört die gesamte Posterior.",{"type":32,"tag":33,"props":735,"children":736},{},[737],{"type":37,"value":738},"Zweiter Tradeoff: Segment-Granularität. Zu feine Segmente (z. B. \"Level 5–10, US, Android, Whale\") führen zu Sample-Size-Mangel pro Segment, die Posterior bleibt hochvariabel. Praktische Regel: Pro Segment mindestens 200 IAP-Impressionen täglich. Darunter: Segmente zusammenfassen (z. B. US+UK+CA als eine \"Tier-1 EN\"-Region).",{"type":32,"tag":33,"props":740,"children":741},{},[742],{"type":37,"value":743},"Drittes Punkt: Psychologischer Effekt von Preisänderungen. Wenn ein Nutzer gestern $4,99 sah und heute $3,99, wirkt das wie \"Rabatt\" und die Conversion springt — aber das ist nicht nachhaltig. Halten Sie den Preis-Range während des Tests eng (max ±20 %), keine radikalen Shifts wie $4,99 → $1,99.",{"type":32,"tag":40,"props":745,"children":747},{"id":746},"scale-und-automatisierung-nach-dem-test",[748],{"type":37,"value":749},"Scale und Automatisierung nach dem Test",{"type":32,"tag":33,"props":751,"children":752},{},[753],{"type":37,"value":754},"Bayesian-Preisoptimierung ist kein einmaliger Test, sondern ein kontinuierliches Lern-System. Nach dem Test deployen Sie den Gewinner-Preis live, speichern aber die Posterior-Distribution und nutzen sie als Prior für neue Cohorts. Im Q4 Holiday Season steigt ARPU um 30 % — die Posterior des vorherigen Quarters startet als neue Prior, das Modell konvergiert schnell zum neuen Optimum (Warm Start statt Cold Start).",{"type":32,"tag":33,"props":756,"children":757},{},[758],{"type":37,"value":759},"Automatisierung mit Airflow + BigQuery + Firebase Remote Config: Täglich liest ein Airflow-DAG Posterior-Parameter aus BigQuery, schreibt neue Preis-Varianten in Firebase Remote Config. Das Client-SDK fetcht Remote Config und zeigt IAP-Offers. Conversion-Events loggen in BigQuery, Posterior aktualisiert — Loop geschlossen. Setup dauert 2–3 Wochen, danach läuft's zero-touch.",{"type":32,"tag":33,"props":761,"children":762},{},[763,765,774],{"type":37,"value":764},"Letzer Schritt: Wenn Sie Bayesian-Modelle auf mehrere Spiele skalieren, bauen Sie einen zentralen \"Pricing Service\". Jedes Spiel sendet Metadata (Genre, Geo-Mix, ARPU), der Service empfiehlt Prior-Distribution basierend auf Spiel-Profil. So leiden neue Spiele nicht unter Cold Start; Sie machen Transfer Learning aus ähnlichen Spielen. ",{"type":32,"tag":766,"props":767,"children":771},"a",{"href":768,"rel":769},"https:\u002F\u002Fwww.roibase.com.tr\u002Fde\u002Faso",[770],"nofollow",[772],{"type":37,"value":773},"ASO-Services",{"type":37,"value":775}," von Roibase kombinieren solche Cross-Game-Learning-Pipelines mit Creative Testing — dasselbe Bayesian Framework funktioniert auch für App-Store-Page-Varianten.",{"type":32,"tag":777,"props":778,"children":779},"hr",{},[],{"type":32,"tag":33,"props":781,"children":782},{},[783],{"type":37,"value":784},"Bayesian-Preisoptimierung ist ein Grundpfeiler von Revenue Engineering in F2P-Spielen. Mit korrekter Segment-Prior, kontinuierlicher Posterior-Update und Thompson Sampling erhöhen Sie IAP-Conversion um 15–40 % und steigern Cohort-LTV deutlich sichtbar. Ein lernendes System statt klassischer A\u002FB-Tests schafft Compounding Effects — jede neue Cohort startet optimierter als die letzte. Um zu beginnen: Teilen Sie Ihre aktuelle Preisstaffel in 3–5 Varianten, konstruieren Sie Prior aus historischen Conversion Rates und beobachten Sie die Posterior über die ersten 10.000 Impressionen.",{"type":32,"tag":786,"props":787,"children":788},"style",{},[789],{"type":37,"value":790},"html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}",{"title":16,"searchDepth":225,"depth":225,"links":792},[793,794,797,798,799,800,801],{"id":42,"depth":202,"text":45},{"id":63,"depth":202,"text":66,"children":795},[796],{"id":107,"depth":225,"text":110},{"id":118,"depth":202,"text":121},{"id":625,"depth":202,"text":628},{"id":660,"depth":202,"text":663},{"id":725,"depth":202,"text":728},{"id":746,"depth":202,"text":749},"markdown","content:de:gaming:bayesian-fiyat-optimizasyonu-mobilf2p.md","content","de\u002Fgaming\u002Fbayesian-fiyat-optimizasyonu-mobilf2p.md","de\u002Fgaming\u002Fbayesian-fiyat-optimizasyonu-mobilf2p","md",1785967477872]