Guru ｜ Auto Ad Placement Agent
Problem & Demand
Guru Game runs over 50 mobile game titles across Google Ads, Meta Ads, and Apple Ads — each needing fresh creatives daily. Design teams spent days producing ad variations. Ad ops teams spent entire days manually uploading creatives platform by platform, title by title. The math simply didn't work.
They needed a system that could discover trending topics, generate matching creatives, place ads across every platform, and learn from performance data — all autonomously. And they needed it fast.
Solution
We designed and deployed a 4-stage end-to-end agent that transforms Guru Game's entire ad operations into a continuous improvement loop.
Stage 1 — Hot Topic Discovery. The agent continuously filters RSS feeds across the web, surfacing trending topics and creative angles for each game's target audience. These signals flow into ggbond, our internal discovery engine that matches trends to specific game titles, generating actionable creative briefs around the clock. What used to be a creative team guessing at what might work is now a data-driven pipeline that finds what's already working and adapts it.
Stage 2 — Creative Generation. From each brief, the agent produces high-quality ad creatives at scale — hundreds of variations ready for A/B testing in minutes. Work that previously consumed the design team's entire week now runs automatically.
Stage 3 — Cross-Platform Placement. The agent connects to ad platforms through MCP (Model Context Protocol) servers, automating campaign creation and creative upload across Google Ads and Meta Ads for all 50+ titles simultaneously. No more logging into three different dashboards and repeating the same flow for every single title.
To complete the operational picture, we integrated the agent with Feishu (Lark) for internal notifications and workflow coordination — built in the pre-lark-cli era by directly interfacing with Feishu's backend API.
Stage 4 — Analytics & Feedback. Performance data from every platform, every title, every creative flows into GF — our private deployment of Google Mixboard. The analytics engine identifies winning patterns and routes insights back to Stage 1, closing the loop. Every ad dollar spent makes the next one smarter.
We didn't just ship code remotely. We flew to the Guru Game office for on-site FDE (Foundational Deployment Engineering), ensuring the agent meshed with their real infrastructure, real workflows, and real constraints.
The impact, measured:
ggbond became the idea engine for the entire creative team — serving ~30 people who now start every campaign with data-backed creative briefs instead of blank-page brainstorming.
GF Mixboard replaced Google Mixboard as the creative team's primary image generation platform, covering ~20 people. Creative generation efficiency: 10x.
The Guru Agent saves each ad optimizer 3 hours per day on creative uploads — time now spent on strategy, not data entry. Meanwhile, the system monitors the entire data platform in real time, flagging traffic anomalies 24/7 before a single dollar is wasted.
End result: the team's ad campaign cycle compressed from 7 days to 2 days. Not a marginal gain — a structural change in how fast a 50-title portfolio can move.