/
AI Media Planner: Technical Overview
Edit

AI Media Planner: Technical Overview

This document is for INTERNAL KNOWLEDGE ONLY and is proprietary to our technology. Do not share ANY of this information externally

This document is designed for internal use to help our team understand the "how" behind the AI Media Planner engine. It provides a look under the hood at the technology that powers our recommendations, moving from initial advertiser insights to a final product strategy.

Table of Contents

Overview: The Power of Generative AI

Our recommendation tool is powered by Large Language Models (LLMs)—sophisticated AI systems trained on vast amounts of data to understand context, reason through problems, and generate human-like text.

By integrating these models directly into our proposal workflow, we’ve shifted from a manual "search and select" process to an "insights-driven" experience. The engine doesn't just suggest products; it analyzes the advertiser’s digital footprint to explain why those products are the right fit.

Our Secret Sauce

The "textbook" used in this process is a proprietary document built by the internal team. It contains our logic on product bundling, budget minimums, and strategy.

  • Privacy: This document is internal-only and is not visible to partners or clients.

  • Updates: If you notice a recommendation that feels off or needs adjustment, contact the Product Team. We manage the supporting documentation that feeds the RAG to ensure the AI stays aligned with our latest sales strategies.

Advertiser Insights: The Digital Scan

The process begins with a website URL. Our engine uses the Google Gemini API to perform a live scan of the advertiser's website.

  • What it does: It "reads" the homepage and sub-pages to identify the business category, core products/services, and brand voice.

  • The Output: Instead of you having to research the client, the AI provides a concise overview of who the advertiser is and what they do, ensuring everyone on our team starts with the same foundational knowledge.

Geo-Locations: Intelligent Targeting

After understanding what the business is, the AI determines where they should target.

  • What it does: The engine scans the website for locations (cities, states, zips, DMAs, etc) based on the information. It then uses the Gemini API to aggregate market-specific data for those regions.

  • The Output: A list of recommended geographic targets along with aggregated market insights to justify the selections

    • If users adjust locations, these locations are sent back to the API to recalculate

Profiles: Generating Ideal Personas

The AI uses the gathered data to visualize the "who"—the ideal customer for this advertiser.

  • What it does: The LLM analyzes the business type and services to identify likely pain points, goals, and behaviors of its customers. The data points entered then get aggregated into three distinct archetypes.

  • The Output: Three "Ideal Customer Profiles" that provide a narrative on who the campaign should target and why they are a fit.

Product Recommendations: Proprietary RAG Technology

The most critical part of the tool is the final recommendation. To ensure the AI suggests the right digital products (Programmatic, SEO, Social, etc.), we use a framework called RAG (Retrieval-Augmented Generation). This is the core of the engine, where data is turned into a specific advertising strategy.

  • What it does: We use RAG (Retrieval-Augmented Generation). Think of this as giving the AI a private, "internal-only" textbook. The AI retrieves our proprietary product rules, augments them with the advertiser's specific data, and generates a recommendation.

  • The Output: A tailored mix of digital products (e.g., SEO, Social, Programmatic) backed by our internal sales logic and best practices.

What is RAG?

Think of a standard LLM as a brilliant student who has read every book in the public library but doesn't know anything about our specific company or products. RAG is like giving that student a specialized, private textbook to refer to before they answer a question.

  • Retrieval: The AI "retrieves" the most relevant sections of our internal product documentation.

  • Augmented: It adds that specific info to the user's current request.

  • Generation: It "generates" a recommendation based on our proprietary rules.

Internal FAQ

We primarily utilize Generative AI powered by Large Language Models (LLMs).

  • Large Language Models (LLMs)

    • We use the same category of advanced technology found in tools like Google Gemini.

    • Think of the LLM as a highly sophisticated "Reasoning Engine." It doesn't just look for keywords; it understands context. This is what allows the tool to "read" a website, summarize a business's core values, and write out detailed customer personas.

  • While the LLM provides the "brainpower," the RAG provides the "knowledge." It forces the AI to look at our specific, proprietary product documentation before it makes a recommendation.

The Bottom Line: We use Generative AI to handle the heavy lifting of research and writing, and RAG to ensure that every recommendation is grounded in our company’s specific expertise and product offerings.

RAG stands for Retrieval-Augmented Generation. Standard AI models are trained on general internet data, but they don't know our specific business rules. RAG allows us to "plug in" our own private data. When you ask the tool for a recommendation, the AI first retrieves the relevant parts of our internal documentation and uses that specific info to generate the answer. This ensures the AI stays on-brand and follows our product strategy.

The tool does take into account product information (rates, minimum requirements, etc.) and will make recommendations accordingly. For example, the tool will not recommend a product that has a 6-month requirement if the user has a 3-month proposal.

No. Because of the RAG framework and product restrictions we have built into the engine, the AI is "grounded" in our own documentation. It is forced to refer to our approved product catalog before making any recommendation.

Strictly no. This document contains our internal sales "secret sauce." It is ring-fenced within the code and is never exposed to external users.

If you find the AI is consistently missing a specific strategy, contact the Product Team. We can investigate updates to the "internal textbook" to refine the AI's logic, but this is not guaranteed.