–40%
CPA after implementation
Project Description
Ad Campaign Control Ecosystem
Yandex Audit Platform is an enterprise SaaS for marketers, agencies, and in-house teams. The platform combines traditional ad management tools with generative AI: from quick campaign audits to complex scenarios like "generate an A/B test, launch it, and shut down the losers in a week".
The AI agent plays a key role: it understands screen context, knows campaign IDs, account structure, API status, and independently calls the right functions. For users, it feels like a "smart employee" embedded in the interface.
The platform collects data from search ads, analytics, Webmaster, Business, and Audiences to provide a panoramic view of ROI/ROMI. Large exports are processed by Celery workers and an engine that handles 50,000+ search query rows, builds CPA/CPC distributions, detects anomalies, and generates action plans.
The result: a shift from manual Excel analysis to an automated optimization cycle that becomes the central marketing decision hub.
Features
Interfaces That Work While You Sleep
01
Unified Dashboard
Accounts, campaigns, statuses, budgets, balance, and API quotas — all in one glass panel. Audits and optimization scenarios launch from here.
DashboardRealtime
Summary Screen
02
AI Assistant with Tool Calling
A floating agent that understands context (campaign page / A/B / semantics) and executes complex commands in natural language.
AI AgentNLU
Tool Calling
03
Deep Query Audit
An Excel export is loaded into the system, which calculates distributions, identifies "draining" N-grams, builds charts, and suggests negative keywords with estimated savings.
N-gramAnomaly Detection
Semantic Audit
04
Campaign Management + A/B
Flexible status and budget management, auto-generated creatives, and a "loser killer" — automatic shutdown of underperforming ads.
AB testsAutopilot
A/B Center
05
Analytics & Time Targeting
Demand heat maps, cross-channel analytics, device and geo segmentation. Bid adjustment recommendations.
HeatmapROI
Time targeting
06
Semantics, audiences, reputation
Shared Sets of negative keywords, RFM segments from Audiences, review monitoring and AI responses from Yandex.Business. Everything integrated into common bus.
RFMShared sets
RFM + reputation
What's inside platform
Modules that collect advertising universes into one control center
Platform covers full cycle: from raw log collection to AI agent actions. Each module is independent service, but all synchronized through unified data bus.
- Neural auditors of traffic quality and search phrases.
- Task orchestrator: Celery workers that parse backups and Excel exports.
- Realtime KPI showcase with SLA and API quota alerts.
- Tool Calling agent that initiates scenarios without human participation.
Work scenarios
How agent leads marketer by hand
Diagnostics
Search query log upload → CPA/CTR statistics, outlier detection, traffic map building.
Drain identification
N-gram engine marks garbage patterns, recommends negative phrases and shows savings forecast.
Optimization and tests
AI assistant generates ad variants, launches A/B and monitors "loser killer".
KPI control
End-to-end ROI/ROMI analytics, time heatmaps, PDF/HTML reports and action log for compliance.
Value
What client gets
Budget optimization. AI finds "leaks" and collects negative keywords, saving 15-40% spend without losing leads.
Execution speed. Routine tasks "analyze campaign / find problem keywords" take 2-5 minutes instead of hour.
Enhanced A/B testing. Creative generation and auto-kill of weak ads gives +20-35% CTR.
End-to-end analytics. Funnel from impression to revenue with segmentation by devices, geo, RFM and real ROMI.
Reputation under control. Auto-responses to Yandex.Business reviews and unified rating monitoring.
Junior accessibility. Natural language makes complex operations accessible even to junior PPC specialists.
Technologies
Stack and integrations
Backend: Python 3.11, FastAPI, SQLAlchemy, Pydantic, Celery + Redis, Pandas, NumPy, SlowAPI, Alembic.
Frontend: HTML5, CSS3, Vanilla JS (ES6+), Chart.js, NProgress, WebSockets.
AI & ML: OpenAI GPT-4o / GPT-4o-mini, Responses API, LangChain patterns, custom N-gram engine, Z-score & IQR analysis.
Yandex integrations: Direct API, Metrika API, Audiences, Webmaster, Business, OAuth, JSON-RPC + REST.
Infrastructure: Docker Compose, Nginx, PostgreSQL/SQLite, Redis, python-dotenv, pre-commit.
Security: JWT, rate limiting, retry with backoff, input validation, structured logging.
Architecture
Yandex Audit Platform project structure
Visual auditYandex tree shows where agents, API, analytics and infrastructure live — so team saves onboarding hours and finds needed module faster.
Results
Implementation metrics
Economics: CPA -25...-40% thanks to smart negative keywords and budget redistribution; spend savings 15-30%.
Operations: Campaign audit takes 5-10 minutes instead of 4-6 hours; processing 50,000 queries without performance degradation.
Strategy: Transition to proactive management: system warns about risks before budget drain and scales to dozens of accounts.
Team: Reduction of human errors, training through AI recommendations and knowledge bases, transparent action log.
Want your own AI marketing center?
We'll build platform that will manage ads, data and processes — as beautifully and flawlessly as this case.