Case Study

ML-powered brand intelligence that generated 4.8 million designs

Digital Scientists built an AI-driven content ecosystem for Mailchimp's Creative Assistant, enabling small businesses to produce professional branded campaigns in under 10 seconds.

Machine Learning Python Computer Vision NLP Microservices
Mailchimp Creative Assistant Site Analyzer scanning a website and extracting brand elements in real time

4.8M+

Designs Created

<10s

Per Campaign

4

ML Microservices

14+

Content Categories

Overview

Building the AI backbone behind Mailchimp's Creative Assistant

Mailchimp engaged Digital Scientists to build the machine learning infrastructure powering their Creative Assistant design tool. The engagement spanned multiple interconnected projects, codenamed Magpie, Raven, and Rhesus, each responsible for a critical layer of the content intelligence pipeline.

Magpie is a website scraper service that uses advanced scoring mechanisms to capture brand and campaign assets from any URL. Raven is the content intelligence engine that analyzes, categorizes, and pairs captured content for specific campaign intents. Rhesus provides image analysis using ML models for logo detection, saliency mapping, and marketing image classification.

Together, these services enable Mailchimp users to generate professional, branded multichannel campaigns with the click of a button, eliminating the manual effort of uploading brand assets and selecting campaign content.

Mailchimp logo

Client

Mailchimp (Intuit)

Industry

Marketing Technology, SaaS

Services

Machine Learning R&D, Product Design, Microservices Architecture, Computer Vision

Engagement

2019 – 2021

Method

Capabilities

The Challenge

Small businesses struggle to create professional marketing content

Mailchimp's small business users needed professional-quality branded designs for email, social media, and ads, but lacked the design skills, time, and tools to produce them. The biggest barrier to adoption of Mailchimp's Creative Assistant was the manual setup process: users had to upload logos, pick colors, choose fonts, and source campaign imagery before they could generate a single design.

Manual Brand Setup

Users had to manually upload logos, colors, and fonts before generating any designs, creating friction that killed adoption.

No Content Intelligence

Campaign content had to be scraped each time with no understanding of context, intent, or relevance to specific marketing goals.

Disconnected Channels

Generating campaign content across email, social media, and ads required separate workflows with no unified brand intelligence layer.

Jobs To Be Done

What Mailchimp's small business users needed to accomplish

Seamlessly integrate their existing website brand into Mailchimp without manual uploads

Generate professional branded designs for multichannel campaigns in seconds

Receive intelligently paired campaign content recommendations based on marketing intent

Produce ready-to-send email, social media, and ad campaigns from a single content source

Accurately abstract a user's brand from existing digital assets without any user input

Get campaign content recommendations that improve over time based on engagement data

Project Magpie

Intelligent website scraping that captures brand DNA instantly

Magpie started as an effort to increase adoption of Mailchimp's ecommerce solution by helping users migrate their existing product data. It quickly grew into a comprehensive website scraper service that uses advanced scoring mechanisms to capture brand and campaign assets from any URL.

A user simply enters their website URL into Mailchimp's Content Studio, and Magpie along with its companion services automatically detects, analyzes, and classifies all discoverable brand elements. The service captures logos, brand colors, background/text color pairings, fonts and font styles, brand images, and a complete website style guide including button styles, navigation patterns, and background treatments.

Magpie testing interface showing URL input field, brand profile and product scraping options for capturing website brand assets Click to zoom

What Magpie captures

Logos Brand Colors Typography Color Pairings Brand Images Site Style Guide Product Listings Social Links Campaign Content Metadata
Project Raven

AI-powered content intelligence that understands marketing intent

Raven is the content intelligence engine that analyzes data acquired through Magpie to provide intelligent campaign content pairings. Its mission: accelerate a marketer's efforts by intelligently recommending content for their next campaign.

The system operates in four stages: User Identification classifies the business by industry and website type. Collection & Analysis uses text and image ML models to collect, store, and categorize content elements. Content Curation groups categorized elements into content pairings based on intent, channel, and campaign type. Performance Analysis feeds engagement data back to refine future recommendations.

How Raven analyzes a website

When a user enters their website URL, Raven identifies marketing triggers such as popup offers, promotional banners, and calls to action. It then extracts and categorizes content elements across multiple dimensions.

Raven detecting a marketing popup trigger on the Tattly ecommerce website, identifying the 20% off signup offer as primary campaign content Click to zoom

Raven identifies marketing triggers like signup popups to determine campaign intent

Content element categories

Raven searches for specific content elements based on context, enabling accurate categorization and intelligent pairing for campaign use.

Hero images
Product data
Sale/reward info
Call-to-action text
About content
Reviews/quotes
Hashtags
Event data
Magpie content pairings interface showing extracted product images, headers, subheaders, and CTAs with ML confidence scores Click to zoom

Content pairings with ML confidence scores for campaign relevance

Campaign Output

From website to multichannel campaigns in seconds

Raven's content pairings feed directly into Mailchimp's Creative Assistant (SAWA), which transforms brand elements and campaign content into professionally designed assets for email, social media, and ads. Each channel receives optimized content with platform-specific formatting, imagery, and copy.

AI-generated branded email campaign with hero image, logo, brand colors, about copy, tagline, and promotional offer automatically composed from website content Click to zoom

Email Campaign

Auto-generated with hero image, brand colors, about copy, and current promotion

AI-generated Facebook ad campaign with contextual product imagery and marketing copy extracted from the brand's website Click to zoom

Facebook Ad

Contextual imagery with human figures, optimized for engagement

AI-generated Twitter post with product-focused imagery, promotional text, and auto-generated relevant hashtags Click to zoom

Twitter Post

Product imagery with stylistic backgrounds and auto-generated hashtags

See It In Action

Creative Assistant demo

The Process

From website analysis to branded designs

The brand elements are instantly assembled into a brand profile and sent to Creative Assistant, which transforms them into beautiful, professionally branded designs.

These designs, in PNG or JPEG format, can be used as ads, heroes, or resized for social media. The entire process generates a multi-channel marketing campaign in less than 10 seconds.

Through the AI-driven pipeline, Mailchimp customers can import a full range of content features to generate custom multichannel designs with the click of a button.

Step 1: Analyze source website for brand elements including logos, colors, and typography
Step 2: Extract and classify brand assets using ML models
Step 3: Assemble brand profile from all detected elements
Step 4: Generate custom branded designs for multiple channels
Technical Architecture

A microservices ecosystem for content intelligence

Digital Scientists designed and built four interconnected microservices that form the content intelligence backbone of Mailchimp's Creative Assistant platform.

Magpie

Website scraper service using advanced scoring mechanisms to capture brand assets, product data, and campaign content from any URL.

Raven

Content intelligence engine that analyzes, categorizes, and pairs campaign content based on marketing intent and channel requirements.

Rhesus

Image analysis service providing logo detection, saliency mapping, and marketing image classification using computer vision ML models.

Hummingbird

Data adapter service that bridges Magpie's output to Creative Assistant (SAWA), translating scraped assets into the design generation format.

Workstream 1

Website Classification

Ecom, Content, Portfolio, Business Card

Workstream 2

Structure Analysis

Pinpoint content to capture and analyze

Workstream 3

Content Analysis

Context identification, parsing, quality

Workstream 4

Content Pairing

Intent-based campaign content assembly

Content Extraction

Deep analysis of every brand touchpoint

When Magpie scrapes a webpage, it breaks the site into identified structures using parsing and scoring logic. These structures are sent to Raven, where they are prioritized and analyzed to yield the most valuable and relevant campaign content. The top content pairings, including images, header text, subheader text, CTAs, and more, are sent to Creative Assistant for design generation.

Product listing extraction showing individual products identified with names, prices, and sale indicators from an ecommerce website Click to zoom
Brand typography extraction showing detected font families, weights, and styles including Proxima Nova from the analyzed website Click to zoom
Copy extraction showing about text, taglines, product descriptions, and mission statements captured from the brand's website for use in campaign content Click to zoom
Results

Measurable impact at scale

Within eight months of the beta launch in April 2020, Mailchimp users created nearly five million instant designs, a testament to the power of removing friction from the creative process.

By combining machine learning with thoughtful product design, Digital Scientists helped Mailchimp make professional-quality branded design accessible to millions of small businesses. Site Analyzer eliminated the biggest barrier to adoption, manual asset uploading, and Creative Assistant turned brand intelligence into beautiful, ready-to-use marketing content.

Increased adoption of Creative Assistant by eliminating manual brand setup

Enabled Brand Playground concept for new user acquisition

Achieved 90%+ accuracy for scraped brand elements across content sources

Expanded Rhesus image analysis to serve multiple Mailchimp platform services

4,849,372

designs created

within 8 months of beta launch

Ready to build AI-powered products?

Digital Scientists helps companies like Mailchimp build machine learning products that scale to millions of users.