Helping Wunderman Thompson Build a Cutting-Edge AI Tool for Checking the Quality and Consistency of 100,000s of Brand Assets

partnership with STX Next
July 2021 - June 2022
Marketing and Advertising
United States

About Wunderman Thompson and Brand Guardian

Founded in 2018, Wunderman Thompson is a global marketing communications agency based in New York. It was formed when their parent company WPP plc merged holding company J. Walter Thompson with digital agency Wunderman.

WPP Open Brand Guardian, created by Wunderman Thompson, is an AI-driven platform that gives marketing, creative, and production teams total content quality assurance at scale.

Brand Guardian complements any corporate asset assurance system with an AI-driven custom rules engine that reduces the time necessary for manual asset checking. Wunderman Thompson’s tool promises:

  • Brand governance introducing consistency across your digital product
  • Inclusion through diversity and accessibility assurance
  • Compliance with legal, industry, and market requirements
  • Creative excellence by applying best practices to boost your content performance

In short, Brand Guardian reviews creative assets—images, video, and text—and provides valuable and immediate insights. It can be plugged into the existing production process and the current toolset.

Business challenges and our cooperation goals

  • Supporting the design of a machine learning (ML) system that can adapt to various customer requirements in the context of their individual policies, compliance regulators, accessibility assurance of advertising assets, packaging, etc.
  • Assisting and guiding Wunderman Thompson through the process of implementing a thorough methodology leading them to become an AI-driven company

Technical challenges and the client’s needs

  • Developing a solution architecture and scalable deploying of computer vision models
  • Accelerating the processing of marketing assets, allowing users to serve multiple customers at the same time
  • Appropriating code design to be modular and reusable, whose goal was to follow the users’ “rules” and reduce the work for future customers
  • Developing generic solutions for generalized computer vision problems (e.g. object detection)
  • Implementing the latest deep learning techniques (object detection, automatic accessibility check-in, segmentation, etc.)
  • Developing a methodology for working with low or insufficient data quality and quantity

Our machine learning and data engineering solution for Wunderman Thompson

The main task for us was to introduce a robust ML solution that will help assess content quality at scale and accelerate the process of onboarding new clients. We identified the main bottlenecks: the data gathering process, ease of introducing new evaluation rules, and speed of asset processing.

Our work focused on redesigning the architecture, allowing existing models and algorithms to be reused and run on a large scale. Part of the solution was introducing ML-focused microservices with better-defined responsibilities. By doing so, we were able to more easily auto-scale with an independent group of computing endpoints in Kubernetes.

We addressed the challenge of detecting undefined, custom image parts (such as logos, fonts, and particular objects). In addition, we created a framework for generating new training datasets (using data augmentation and synthetic data). The framework accelerates training and provides agile and robust solutions for client requirements. New, artificially created data solves the problem of detection.

We also suggested a messaging system for running most machine learning and computer vision tasks by multiple Celery workers in parallel. It allows computing each “rule” as a separate task and processing the entire asset analysis pipeline, significantly improving performance.

“The collaboration was based on complete openness and trust in our people and their skills. As a result, we were able to jointly develop a number of new approaches and significantly improve the implementation process of the models and the architecture of the overall solution.”

Krzysztof Sopyła

Head of Machine Learning and Data Engineering @ STX Next

Developing Brand Guardian: Creating AI microservices

  • Collaborating with Wunderman Thompson on image processing pipelines for images to simplify AI-targeted tasks
  • Collaborating on an AI module that will handle all AI-related operations for the rule engine in the core platform
  • Working together on moving the AI logic from the core application (Java) into multiple AI services (microservices in FastAPI)
  • Handling all heavy-lifting tasks inside Celery workers with async support
  • Introducing good software engineering practices (especially in Python), writing unit tests, and adding Python tooling (formatting, linting, static type checking, etc.)
  • Creating a prototype of a report generator tool for creating endpoint summaries (complete ML result with preprocessing and postprocessing)

Developing Brand Guardian: Building a smart detector framework

  • Creating the first iteration of an artificial dataset generator for logo detection and segmentation (multiprocessing, background downloader and crawler, modular architecture, command line interface, etc.)
  • Writing scripts for Detectron2 to be run inside Azure Machine Learning Studio
  • Writing utils-scripts to work with unbalanced or low-quality data
  • Collaborating on creating a custom inference solution for an AI service

Developing Brand Guardian: Setting up AI-related client rules

  • Collaborating on logo detection (sift, template matching, etc.) with STX Next introducing deep learning to the solution
  • Collaborating on logo clear space calculation (logo visibility according to the background and adjacent elements)
  • Implementing font size and baseline detection based on OCR and classical computer vision algorithms
  • Implementing font type and family detection using Siamese CNN
  • Collaborating on detecting objects of interest using classic computer vision algorithms (edge detection, dilation, contour detection, flood fill, binary mask area count, etc.)

Services we’ve provided to Wunderman Thompson

  • Machine Learning
  • Software Engineering
  • Cloud Deployment
  • Architecture Design
  • Expert Consulting
  • Microservices
  • Computer Vision
  • Deep Learning
  • Cloud Services

Can we help you the way we’ve helped Wunderman Thompson?

If you’re planning to build a software product, but don’t have the skills or capacity to do it—we can help you just like we’ve helped Wunderman Thompson.

We’ve created marketplaces, internal deployment tools, community platforms, and much more for clients like Mastercard, Unity, or Decathlon.

We have:

  • 18+ years of market experience
  • 800+ projects delivered under our belt
  • 300+ clients served from all over the world
  • 500+ professionals on board (including 350+ developers)

We can take your product from a vague concept all the way to launch, then keep growing it with you using industry best practices and technological innovations.

Start small with just two developers or hire a full team from the start—we’re ready for any challenge that comes our way.

Head over here to contact us and tell us about your project!

Tell us about your project

Speed up work on your software projects and outpace the competition.

marta blazejewska director of sales

Marta Błażejewska


matthew harris photo

Matthew Harris

vp of sales

Our customers love to work with us