yann lecun

The marketing tech stack has changed considerably across the last decade or so. Generally I like to frame this evolution as follows:

Gen 1: Siloed Marketing Tools (Pre-CDP Era, Pre-2015) – Data was fragmented across various tools, requiring manual integration and leading to inefficiencies
Gen 2: The Rise of Customer Data Platforms (CDPs) – CDPs attempted to unify customer data and streamline integrations, but they introduced high costs and redundancy
Gen 3: The Warehouse-Centric Approach – Companies now store all customer data in a centralized data warehouse and use Reverse ETL tools to sync data back into business applications

Below is a deeper dive into the tech stacks of each era.

Gen 1: Siloed Marketing Tools (Pre-CDP Era)
Before CDPs emerged, marketing teams relied on a collection of specialized tools for different functions, but these tools did not communicate with each other effectively. The result was fragmented data, manual workflows, and a reliance on engineering teams to move data between systems.

The typical Gen 1 marketing stack consisted of the following components:

• Event Tracking & Analytics: Tools like Mixpanel, Amplitude, Heap, and Google Analytics captured user interactions such as page views, button clicks, and session times.
• Marketing Automation & CRM: Platforms like Marketo, HubSpot, Eloqua, and Salesforce stored customer and lead information and enabled marketing campaigns.
• Advertising & Attribution: Solutions such as Google Ads, Facebook Ads, AppsFlyer, and Adjust tracked ad performance and attribution data.
• Customer Engagement & Personalization: Messaging platforms like Braze, Customer.io, Intercom, and LiveChat enabled communication via email, push notifications, and chat.

Pain Points of Gen 1
This fragmented approach led to several issues:

• Data silos: Each tool had its own version of the truth, making it difficult to get a unified view of customer behavior.
• Engineering dependency: Marketers needed help from engineers to integrate systems, pull reports, and clean data.
• Manual data movement: Extracting and uploading data between platforms was time-consuming and error-prone.
• Inconsistent and incomplete data: Discrepancies often arose between different tools due to variations in tracking methods and storage formats.
• Difficulty in real-time activation: Because data had to be manually moved between tools, responding to customer actions in real time was nearly impossible.

Gen 2: The Rise of Customer Data Platforms (CDPs)
To address the problems of Gen 1, CDPs like Segment, mParticle, RudderStack, and Tealium emerged. These platforms aimed to unify customer data across various sources and make it accessible to marketing, sales, and analytics tools without requiring heavy engineering involvement.

How CDPs Changed the Stack
In the CDP-driven model, companies no longer had to integrate each tool separately. Instead, all event data was sent to the CDP first, which then distributed the data to various downstream applications. For example, a “Signup Completed” event would be collected once by the CDP and then automatically sent to:

• Amplitude for behavioral analysis
• Braze for triggering a welcome email
• Facebook Ads for retargeting campaigns
• Salesforce for sales follow-ups

Problems CDPs Solved
CDPs eliminated data silos by centralizing customer data. They reduced some of the engineering workload since they no longer required integration buildouts into each of the analytics/marketing tools. They also enabled better real-time data activation, since data could now flow automatically between systems.

Why CDPs Were Not a Perfect Solution
Despite their benefits, CDPs introduced new problems that limited their effectiveness:

• High costs and redundant storage: Because CDPs stored customer data separately from a company’s main database, businesses effectively paid for duplicate storage.
• Limited data transformation capabilities: CDPs collected and routed data but did not allow for complex data transformations, such as merging behavioral data with transactional data from financial systems.
• Data accuracy issues persisted: If the original event data was messy or incomplete, the CDP simply passed along the bad data to other tools.
  These limitations led companies to rethink their approach to data management, resulting in a shift toward a warehouse-centric model.

Gen 3: The Warehouse-Centric Stack & Reverse ETL
The most recent evolution of the marketing tech stack is centered around the data warehouse, where all customer data is stored, cleaned, and enriched before being sent to business applications. Instead of relying on a CDP to store and distribute customer data, companies now use data warehouses like Snowflake, BigQuery, and Redshift as their single source of truth.

How the Gen 3 Stack Works
Data Collection: Event tracking tools such as Snowplow, RudderStack, or Segment collect user behavior data and send it directly to the warehouse instead of storing it in a CDP.
Data Warehouse as the Source of Truth: All customer interactions, transaction records, and support logs are stored in a central warehouse, where they can be cleaned and enriched.
Reverse ETL for Data Activation: Instead of sending data directly from a CDP to marketing tools, Reverse ETL platforms like Hightouch and Census extract structured data from the warehouse and sync it back into tools like Salesforce, Braze, and Marketo.

Why This Model Is More Effective
No more redundant data storage: The warehouse is the single source of truth, eliminating the need for a separate CDP.
More control over data quality: Data is cleaned and enriched before being sent to marketing tools, ensuring consistency.
Reduced costs: Companies avoid the expensive data storage fees associated with CDPs.
Greater flexibility: Data from multiple sources (event tracking, sales, finance, support) can be combined and activated based on business needs.
By using a warehouse-first approach, companies can ensure that all teams—marketing, sales, customer success—are working with the same high-quality data without the inefficiencies of previous generations of the marketing tech stack.

The Shift from CDPs to Reverse ETL, and now Composable CDPs
CDPs were designed to solve the problem of fragmented customer data, but they introduced new inefficiencies. Now, we have a new term on the market which is a “composable CDP,” which is essentially reverse ETL plus a marketer-friendly interface where teams can drag and drop actions such as defining customer segments (e.g., “high-value users who haven’t purchased in 30 days”), syncing audiences in real-time to ad platforms and email automation tools, etc. This makes it easy for non-technical users to create audience segments without writing SQL or relying on data teams.

What are your thoughts on the marketing tech stack? Where do you see this space going? Comment your thoughts.