Yirla Agentic Query Engine (The Assistant Logic)

Yirla Agentic Query Engine (The Assistant Logic)

Overview: Data Synthesis vs. Static Reporting

The Yirla Assistant is not a basic chatbot; it is an Agentic Query Engine designed to perform real-time data synthesis across siloed ad accounts. It solves the "Data Fragmentation" problem by interpreting natural language intent and executing complex multi-step analysis to provide strategic answers, not just raw tables.

The "Reasoning Loop"

When a user asks a question, the Assistant follows a proprietary reasoning chain to ensure accuracy and groundedness:

1. Intent Decomposition: Breaking compound questions (e.g., "Why is my spend up but leads down?") into focused sub-queries for Google, LinkedIn, and CRM data.

2. Contextual Retrieval: Pulling real-time metrics while maintaining account-specific memory (e.g., knowing that "last month" refers to a specific billing cycle seen in the Portfolio Oversight).

3. Logic Application: Applying Yirla’s Performance Framework (Quality > Volume) to filter the results.

Core Capabilities & Agentic Tools

Based on our UI analysis, the engine utilizes specific "Tools" to answer complex B2B queries:

Cross-Platform Comparison Tool

* Function: Analyzes performance parity between Google Search and LinkedIn Ads.

* Logic: Instead of comparing raw CPC, it synthesizes Lead Quality Scores from the CRM to determine which platform has a higher "Pipeline Velocity".

* Example Query: "Is LinkedIn performing better than Google for our Mid-Market segment?"

2. Zero-Impression Diagnostic Tool

* Function: Troubleshoots campaign delivery issues.

* Logic: The agent cross-references three distinct layers:

* Campaign Status: Is it active?

* Data Availability: Is the pixel/API firing?

* Structure Audit: Are the bidding and targeting constraints too narrow?

3. Creative Impression Analysis

* Function: Identifies creative fatigue or "Zero Reach" scenarios.

* Logic: Correlates Headline Patterns and Emotional Tones with impression share to find where the "Creative Decay Index" is impacting the auction.

Executive Synthesis: The "Top 3" Logic

The Assistant is programmed to move from "Finding Data" to "Actionable Advice". For every query, it attempts to provide:

* The Insight: What happened (e.g., "Mid-Market CPL rose by 20%").

* The Why: The underlying cause (e.g., "Competitor Brex shifted to a 'Relatable' tone, winning the auction").

* The Action: What to do next (e.g., "Pivot Mid-Market creative to Benefit-Led hooks").