Note d'experts

AI for Data Marketing: Challenges, Limits, and Adoption

Mathieu Lima
Mathieu Lima
Mis à jour : 26 nov. 20258 min read

Introduction

AI is transforming all professions, particularly those in data marketing. The promises are numerous: intelligent agents, automation, recommendations, content generation, business copilots...

But this excitement creates a lot of noise. Between tools, services, platforms, and integrated features, it becomes complex to know where to focus efforts. How can companies (large corporations, SMEs, or startups) structure their AI and data approach without getting lost in the race for tools and move forward efficiently and sustainably?

Many players launch AI marketing projects without a clear framework. They test tools, connect APIs, create workflows... but without an overall vision. The result: fragile data stacks, difficult to maintain, uncertain return on investment, not to mention risks related to security, GDPR compliance, and data confidentiality.

We help companies and their marketing/data teams see clearly on AI challenges, Data governance, and Performance Marketing. We support them in deploying the right Data architecture (Modern Data Stack), defining an adapted governance framework, and activating high-value use cases.

Solutions Overview

Solutions can be classified into four main categories:

1. No-code or Low-code Automations and Workflows

For example: Make, n8n, Zapier,...

  • Allow generating agents and/or automations without necessarily being a developer.
  • Useful for prototyping use cases, but often difficult to maintain or deploy at scale.
  • These solutions pose challenges for companies subject to strict security and compliance policies (GDPR, ePrivacy, cloud security).
  • These workflows proliferate on LinkedIn, with promises often too good to be true, difficult to implement and maintain.

2. Enterprise AI Platforms

For example: OpenAI, Google Vertex AI / Gemini Enterprise, Anthropic,...

  • Major market players offer integrated artificial intelligence platforms with Business / Enterprise plans to deploy LLMs and AI agents at organization scale.
  • They are increasingly customizable (creation of agents / instructions like GPTs / GEMs), integrate automations, and increasingly connect to external data tools (CRM, CDP, BI).

3. Internalized Models

For example: Mistral, Llama,...

  • For companies that want to control everything: security, sovereignty, and costs.
  • These projects are heavier and suited to large groups with major control, confidentiality, and data scalability challenges.

4. AI Integrated in Specialized Tools

For example: HubSpot, Notion, Salesforce,...

  • Most data / marketing tools on the market develop and offer AI features and integrations.

MCP Servers: The Key Technological Building Block

The emergence of this market standard for artificial intelligence architectures is, in our view, a very important step to accelerate AI adoption in companies, as it greatly simplifies how LLMs (Large Language Models) communicate with different tools.

Definition and Origin

The Model Context Protocol (MCP) is an open standard (open-source) initiated by Anthropic. It standardizes how an AI application connects to company data, tools, and workflows. The objective: avoid fragile ad hoc integrations and offer a stable interoperability layer between LLMs and the IS (Information System). MCP is a kind of universal plug that allows the LLM to connect to third-party service APIs, where it was complicated before because each API had its specificities. MCPs allow exposing to AI applications (LLM) the tools (executable functions) to act on the tool and therefore be able to retrieve data or give instructions.

Architecture and Roles

  • MCP Host / Client: the AI application (e.g., ChatGPT, Claude, Gemini,...) that opens a connection to one or more MCP Servers.
  • MCP Server: a service (local or remote) that exposes Tools/Resources/Prompts in MCP format for a given domain (e.g., GA, BigQuery, HubSpot, Snowflake, GitHub,...).

Why It's Key for Marketing and Data

  • Unified context: the LLM can read and act across your entire data marketing ecosystem (CRM, DWH, analytics, media tools).
  • Less complexity: we replace a mosaic of point-to-point automations with connectors, and it's the LLM that orchestrates and communicates with the tools.
  • Maintenance and scalability: we add or remove capabilities per MCP server, without rewriting the entire workflow.

The MCP ecosystem is still young. It was first deployed to run in a local environment (Gemini CLI, Claude Desktop), but MCPs are now being deployed on cloud environments (ChatGPT / Gemini with ADK, etc.). A large part of data & AI technology solutions have already created their cloud MCP server to be able to communicate directly with enterprise LLMs and AI agents.

EdgeAngel Recommendation: Deploy an AI Platform Within Your Organization

From our reading of AI and data marketing trends, our experience, and deployments in recent months, we are convinced that the new MCP standard and AI Platform features (ChatGPT Enterprise, Gemini Enterprise/Business + Vertex) should be at the center of your efforts. The focus should be on deploying these solutions and configuring specific AI agents for marketing, data, and analytics professions.

1. Deploy an Enterprise Plan

Deploy an Enterprise Plan adapted to your organization (ChatGPT Enterprise / Gemini Enterprise / Business) for your teams, work on adoption and training, while ensuring control of security parameters, GDPR compliance, and data confidentiality.

2. Define High-Value Use Cases

  • Build "simple" agents by profession and use case
  • Deploy more complex agents, connected to your data (Drive, Data Warehouse) and your tools (via MCP):
    • Very specific prompt.
    • Provide your company context (so the LLM fits your requests as closely as possible).
    • Implement data governance to continuously optimize these agents.
  • With Gemini Enterprise and ADK (Agent Developer Kit) to deploy enterprise AI Agents.
  • With ChatGPT (in developer mode for now).
  • Implement strict governance (attention to security, traceability, and cloud costs).

3. Automate

Agents can then be automated for repetitive tasks, where the agent will function autonomously (supervised by humans... or other AI agents)

  • Directly in the AI platform with embedded automation features (scheduled tasks).
  • Or via building "off-platform" agents with Agent Builders (OpenAI or Vertex AI) for more complex AI automations.

This organization allows the business to take control over agent creation, which will by definition be adapted to team needs, and allows the company (the administrator) to manage governance (role / authorized tools / monitoring etc.)

Note: this doesn't prevent using automation tools like n8n for a POC, or if your company has created the conditions for the tool to be maintainable and scalable. But this is not the approach we consider most suitable for promoting adoption at company scale.

Concrete example with the Google Analytics MCP server exposed in ChatGPT (business plan for companies):

  • Deployment of Google Analytics MCP in the cloud (+ customization + authentication management)
  • Creation of a master prompt to define anomalies and ask the LLM to handle data visualization.
  • The data analyst can in a few seconds identify problems, explore leads with the LLM, create a professional incident report, and send it to teams.

What This Implies

It's with good data and your own company context (DNA, history) that AI will be able to truly bring value to marketing and data teams.

This requires:

  • Robust tracking and data collection.
  • A Data Warehouse-oriented architecture (GCP BigQuery / Snowflake).
  • Solid data pipelines (ETL) to feed models and maintain data consistency.
  • Cloud Engineering to develop and maintain custom MCP servers (authentication, security, cloud costs) with total control.

EdgeAngel supports companies in this approach:

  • Strategic AI & data marketing consulting
  • Design of AI architectures (LLM + MCP + data pipelines)

With Capture, our tool, we enable deploying AI use cases at lower cost, by connecting data pipelines and specialized AI agents. You thus remain in control of your tools and data, and we take care of the plumbing.

AI is a strategic lever, and it must be aligned with each company's operational reality. Moving step by step, structuring before automating, and working with the right experts: that's what will make the difference between experimentation and sustainable transformation.

Discuss with an EdgeAngel AI & Data Expert

Want to assess your company's AI maturity, build your AI Platform, or identify relevant use cases?

👉 Schedule a discussion

Partnership with Silex (https://www.silex-labs.com): for complex AI missions and large-scale deployments, we have created a partnership with a top French player and work with their AI engineers (MVA ENS Cachan x École polytechnique) to bring maximum value to our clients.