Enhancing Digital Marketing Insights: AI Agent Case Study

Scenario 1

How are our Google Ads performing week over week?

AI Agent: I'll help you retrieve the spending data for last week and compare it with the four-week average.

(Gathering comprehensive data...)

Here's what I found,

• Four-week average spend $245,678.90
• Last week spend $200,345.67
• Spending shrunk by 22% 📉
• Conversion volume shrunk by 15% 📉
• Cost per acquisition increased by 25% 📈

I thought our campaigns were doing better over the past few months. What happened?

AI Agent: Let me look into the emails and marketing team's Slack conversations.

I'll also search our curated news channel.

(Gathering comprehensive data...)

Novak Djokovic suffered a straight-sets defeat to Italy's Matteo Arnaldi, marking his third consecutive loss and second first-round exit in recent tournaments. The incident sent shockwaves in the sporting industry.

There are a few conversations on Slack about pausing the tennis coaching ads, which could be the reason for the drop in numbers.

Do you want me to compare the performance of other platforms like Meta, ASA and Bing Ads for the same period?

Scenario 2

Suggest some creative copy/themes that could perform well in our upcoming Valentine's Day campaign.

AI Agent: Let me analyze the performance of our Valentine's Day campaigns from the previous years.

Next, I'll search our Google Drive for relevant files.

Here are a few themes that have performed well in our previous Valentine's Day campaigns,

1. Fall in Love. With Fitness. With Pilates.
2. Partner Pilates. Strengthen your bond.
3. Mat about you.
4. Soulmates on the mat.
5. You had me at Roll Up.

Do you want me to suggest more themes similar to these?

AI Assistant for Digital Marketing Teams

Over the past five months, I developed an AI Chat Agent for a US client, overcoming technical challenges and delivering strong results.

The AI Chat Agent behaved like a digital marketing expert, answering questions from their internal team about ad performance, factors that may have influenced it, ideas for creative assets, and decisions that live hidden in communication channels.

At the heart of this system is their data – reports from Google Ads, Meta Ads, Bing Ads, Apple Ads, Spotify Ads, etc., some dating back to 2015-16. We could query this rich dataset dynamically, leveraging Vanna AI.

My primary responsibility was to build RAG pipelines for the following sources: Email, Google Drive, Slack and SimilarWeb. The solution consisted of,

• Typer CLI commands to fetch data, slice them into smaller chunks, create vector embeddings using Azure OpenAI and store them in the database.
• Azure Container App Jobs running the CLI commands at regular intervals.
• Azure Monitoring with OpenTelemetry for logging and tracing.
• Azure Monitor Dashboard to visualize the tokens utilized every day.

Once the RAG component was ready, I started integrating it with the web APIs. The solution consisted of,

• LangChain and LangGraph tools that search and retrieve information from the embeddings. Claude AI to digest the data from tools and answer the queries (with thread-level memory).
• FastAPI to provide chat API endpoints.
• VueJS to power the chat interface.
• Sentry to monitor 500 errors raised in production.

Looking back, three things made the project click:

1. Ruthless scoping. We kept things simple and tried not to "boil the ocean". We worked on one knowledge stream at a time and made it sing before adding anything else. That restraint saved us from weeks of integration headaches.
2. Observability from day one. The OpenTelemetry traces weren't a luxury; they were the safety net. Any spike in token usage or lag in embedding jobs came on the Azure Monitor dashboard within minutes. Instead of poring over logs at 11 p.m., we could spot-fix misbehaving jobs while the coffee was still hot.
3. A stubborn focus on the UX. Marketers don't care whether LangGraph is orchestrating eight tools behind the curtain—they want answers faster than they can type "/ga-report." We kept the chat interface clean, let Claude handle follow-ups in the same thread, and reserved the nice-to-have items (like drill-down charts) for later.

Where to next?

(These are my ideas and not necessarily on the roadmap yet)

1. Streaming insights to Slack. Instead of waiting for someone to ask, the agent will drop a morning digest: spend deltas, CPA anomalies, and a "why it moved" blurb with links to the raw data.
2. Auto-generated briefs. Feed the agent a campaign goal and a budget cap; get back media mix suggestions, draft copy, and a polished deck—ready for the client's inbox.

Five months ago, this was a whiteboard sketch. Today, it is the teammate that never sleeps, never ghosts a Slack mention, and happily explains why last week's tennis ads face-planted—all before the morning stand-up.

The AI revolution is real and is happening right now.