SensorTowerSensorTower

Speeding Up the Back End with Graph Theory

Nov 26, 2019·4 min read

Summary

Sensor Tower improved a sluggish endpoint by analyzing its profiler data and benchmarking. The first optimization added a dirty flag to nodes, skipping re-encoding for unchanged parts, cutting time by 30%. The second optimization replaced linear tree traversal with precomputed paths, enabling logarithmic-time node access.

This required occasional background updates to maintain accuracy. Combined, these changes reduced latency from 300-500ms to about 80ms, a fourfold speedup, with negligible impact on user experience.

You Might Also Like

AppsFlyerAppsFlyer

Omnichannel Analytics Explained

Omnichannel analytics unifies customer behavior and campaign performance across mobile, web, CTV, and AI search into one identity, one attribution model, and one ROAS number. The core problem: a single customer seen on three surfaces becomes three users, inflating performance and corrupting attribution, LTV, and retention. Stitching channel reports into one dashboard is not unification—each channel keeps its own attribution logic, so conversions get double-counted and upper-funnel spend looks unproductive. For ad ops leaders, the takeaway is that identity resolution (deterministic plus probabilistic, privacy-safe) must happen at the data layer before any dashboard exists. Platform choice hinges on surface coverage, identity resolution, compliance, incrementality testing, and migration support.

Roi Tamir·Oct 5, 2026·12 min readRead article →
AppsFlyerAppsFlyer

Creative optimization: Why your best video ad fails as a static image

AppsFlyer’s 2026 creative data shows video and static image ads reward opposite creative qualities, so ad ops teams should plan, budget, and test them separately. Video spend is concentrating: gaming’s top 2% of video creatives captured 60.7% of spend, up from 53.5%, while image spend spread out. Top spenders are testing far more variations, widening the testing gap. Action/Strategy and Puzzle/Lifestyle show format-specific tone and social-proof reversals; Finance is the exception. Teams should use AI-assisted variation, analyze video winners deeply, keep broad image testing, and measure performance at attribute level.

Shani Rosenfelder·Oct 19, 2026·9 min readRead article →
AppsFlyerAppsFlyer

Pitfalls of modeling LTV and how to overcome them

LTV forecasting is challenging due to stale ML models and user heterogeneity. Effective systems combine ML, Bayesian methods, and secondary models to reduce bias and variance for accurate predictions.

Shani Rosenfelder·Jan 29, 2020·6 min readRead article →
AppsFlyerAppsFlyer

Cracking the 2026 Holiday Season: A Data-Backed Playbook for Marketing Leaders

Q4 2026 planning should treat peak shopping as a compressed event, not a week. Black Friday/Cyber Monday purchase lift runs 2–3x install lift, so ad ops must schedule by daypart and hour. Singles Day is not global: Brazil, UAE and Spain show real lift, while US/France barely react. CPI spikes hardest in supposedly cheap markets—Spain’s CPI rose nearly 50%. Remarketing already drives 65–92% of non-organic spend and rises during peak. Deep linking is critical: deep-linked users are ~2x more likely to purchase and spend nearly 3x more. Benchmark market CPI, shift budget to remarketing, and connect every entry point.

Stephanie Chevalier Naranjo·Oct 7, 2026·10 min readRead article →
TikTokTikTok

Most businesses underestimate how ready they are for TikTok | TikTok For Business Blog

TikTok's UK/Europe scale (30M+ UK, 200M+ EU) and content graph make category objections obsolete. For ad ops, success comes from relevance, not trends: proof-led demos, explainers, reviews, and problem-solving content can reach high-intent audiences without a large follower base. Case studies show material performance: Toyota CPA -38%, BCP 10x registrations, Randstad 2.81% CTR/17M impressions, Prep Kitchen +44% conversion and +374% orders with Smart+, PureGym CPA -77%, HydroClean cost per conversation -74%. Action: test utility/creator content, use Smart+ for scaled creative testing, layer TopView on trusted content, and target niche communities rather than broad reach.

Dan Madden UKI Independent·Oct 7, 2026·7 min readRead article →
AppsFlyerAppsFlyer

Web Gaming: The untapped path to lower CPI and boost ROAS

Mobile CPI and app store fees are eroding ROAS, making web gaming a viable acquisition channel. Full Unity-to-HTML5 ports now deliver 30-50% lower CAC and avoid 15-30% store revenue cuts via web payments. Measurement parity is here: AppsFlyer enables independent attribution, conversion signals, and unified reporting. Retention can improve; one strategy title jumped from 6-7 minutes mobile to 70+ minutes web. Best fit: deep RPG, strategy, puzzle, story games, not hypercasual. Action: audit catalog, pilot web ports, link accounts cross-platform, and use owned media to drive high-intent web users.

Adam Smart·Sep 24, 2026·7 min readRead article →
MetaMeta

Peak Season 2026: You Don’t Beat The Auction. You brief it.

The auction is a brief: optimisation event, budget, bid strategy, and attribution tell it which shopper to pursue, but it never asks clarifying questions—bad instructions execute perfectly. Results lag, so advertisers tinker, restart learning, and pay visible and invisible costs. Treat every campaign as a brief, use value-based optimisation where relevant, align attribution with finance, and consolidate changes. Maintain roughly 50 conversions per ad set weekly; below that, performance was ~28% costlier per purchase. Use the budget formula to cap ad sets. For peak, fix the largest campaign's optimisation, leave it alone, and measure against a holdout or lift study.

Oct 21, 2026·6 min readRead article →
MetaMeta

A New Way for Businesses and Personal Agents to Work Together

Meta and Sierra are previewing an open Personal Agent Protocol with Genesys, NiCE, Decagon, Rocket, Shopify, Stripe, and Walmart to standardize how personal agents like Muse interact with businesses. For ad ops, the key shift: high-intent customers will arrive via agents, not browsers. Blocking agents risks losing those customers. The protocol aims to cover authentication, user identity, business steering, APIs/connectors, and agent-to-agent communication. Brands should prepare for agent-readable product, inventory, booking, and support flows, plus measurement and consent. Expect less page-based browsing, more structured API and agent traffic; adapt bot policies, attribution, and customer journey controls now.

Oct 6, 2026·6 min readRead article →

More from SensorTower