AI Google Ads Performance Analyst

A multi-account reporting engine and live chat agent that turn raw Google Ads data into prioritized, plain-English recommendations — with every number computed in code, never guessed by the model.

How it works

The problem

Agencies managing dozens of ad accounts under one manager account spend hours every week pulling reports by hand, and answering a simple 'how is this campaign doing?' means opening the platform and digging through tabs.

The approach

1. 1

   A scheduled workflow pulls performance data across every sub-account under the manager account via the Ads API, filtered to the campaigns that matter.

2. 2

   Budget pacing, quality scores, and search-term hygiene are computed deterministically in code — the LLM never invents a number, it only explains what the code already calculated.

3. 3

   Claude turns the structured findings into a prioritized, plain-English report per account and emails it automatically, skipping accounts with nothing new to report.

4. 4

   A separate conversational agent, wired to the same underlying data through callable tools, answers ad-hoc questions instantly in a chat interface instead of waiting for the next scheduled report.

Highlights

• Numbers are computed in code and only narrated by the LLM — no hallucinated metrics.
• Scales across a multi-account manager structure without per-account setup.
• Two ways in: scheduled email reports and an always-on chat agent for instant answers.

Built with