Case Studies / Synthesis uses machine learning to speed up Absa – Barclays brand separation

When Barclays announced that it would separate from the Absa group, Absa’s brand team got busy. With a limited period to identify and remove all Barclays logos and mentions from key documentation, websites, contracts, database structures, marketing and communications, it needed an innovative alternative to man hours. Specialist software provider Synthesis stepped up with an agile approach to help deliver a solution using machine learning and a microservices cloud-based architecture.

Says Craig du Toit, who led the project at Absa: “Brand is unimaginably pervasive in technology. In June 2017 we were given 12 months to remove all Barclays mentions in South Africa. It was a massive project. By the end of 2017, we had identified 40,000 artefacts that needed to change across 300 systems.

“To meet our deadlines, our technology and brand team worked together to design an intelligent, innovative solution. Synthesis, with its strong microservices and cloud capabilities, brought the skills and insight we needed to get the new technology in place in record time, without forcing us to turn a complicated toolset into a high-cost long-term product or vendor commitment.”


The discovery process to identify artefacts had taken Absa’s internal team eight months. They built a powerful but complex manual system with multiple element checks to analyse each artefact, identify necessary changes, and create an audit trail in order to facilitate quality assurance and meet legal requirements. “The process was thorough but laborious,” notes Du Toit. “To identify the necessary changes, make the changes and validate them multiple times would take more than 18 months if done manually—time we did not have.

“We began to explore the creation of a brand learning system that would automate brand analysis. We knew we needed intelligent technology, the flexibility of a microservices architecture and the infinite scalability of the cloud to meet our deadlines.”


The Absa team’s first requirement was co-location and it fit Synthesis’ agile DevOps approach perfectly.

“The team understood quickly what needed to be done,” says Du Toit. “AWS provided the ingredients, Synthesis helped us create the recipe and put it together.”

“Synthesis had the experience and know how to implement a highly scalable solution that incorporated machine learning, how to deploy a microservices architecture, and how to rapidly build the technology stack required.”

Says Michael Shapiro of Synthesis: “Our solution was to take the varied Absa branded artefacts and run them through our AWS powered process engine. The process engine would analyse each file type, anchor the record in the AWS hosted database, then deconstruct that artefact, extract the images out of it, use artificial intelligence (AI) and machine learning (ML) to identify logos and text, and finally execute a search to find a brand match.

“Additionally, we compiled a register with all analysis results, providing auditors with documented and reported proof for each originating original anchored document, as well as changed document.”

An agile approach was critical to enabling the Synthesis team to meet the targeted deadline. “We used a lean Kanban approach with comprising of daily stand-up meetings with Absa, allowing the team to micro-pivot and course correct each day to deliver a minimum viable product (MVP) swiftly,” explains Shapiro. “With a team of software, cloud and marketing specialists, we delivered a business-usable solution in eight weeks.”


How well does the solution work?

Absa’s initial proof of concept system was able to assess 100 documents in a day. The Synthesis, AWS solution is virtually infinitely scalable, able to spin up many server instances simultaneously to handle a greater load. It was able to analyse 5,000 records in 30 minutes.

The technology

For machine learning, Synthesis used AWS SageMaker. It’s a fully-managed platform that enables developers and data scientists to quickly and easily build, train and deploy machine learning models at any scale. For image and text analysis, the team used AWS Rekognition.

Together, these solutions were highly effective. Says Shapiro: “Rekognition enabled word and image search and SageMaker enabled a context search. Soon, we began to understand tone and context and where documents ‘fitted’.

Adds Du Toit: “We needed the solution to be flexible and a key principle has been modularity. We have built many microservices to enable that. Going forward, these microservices can be consumed as-a-service to meet Absa rules and guidelines across brand items.”

The benefits

“This solution allowed us to complete the brand audit for South Africa by the end-of-June deadline, and has the potential to add great value to the brand in future.”

“For the Absa team, this was a journey to resolve a challenge we had not faced before. What has subsequently become clear is the potential and impact the machine learning elements of this solution may have on other areas within brand and marketing approaches in the future.

“Imagine the possibilities of using image recognition and sentiment analysis styled tools on a photograph of a branch, or to understand how your call centre agents are engaging with customers. Using machine learning to identify the tone and sentiment surrounding brand mentions on social media and elsewhere is an inevitable outcome of the technology and is being used extensively across the world — but the still unexplored possibilities of AI and machine learning will revolutionise brand strategies and brand management.”.

Synthesis’ Shapiro agrees: “We believe that with this solution a company can identify not just brand mentions but the sentiment around brand adding a very scientific approach to marketing analysis. This data-driven methodology to branding gives marketing professionals the potential to use empirical evidence rather than perception to drive decision-making. For Absa, this opens up many opportunities in the brand team to offer the rest of the bank this innovative ‘Brand-as-a-Service” solution going forward.”

“These are avenues we will certainly hope to explore,” notes Du Toit.

To learn more contact us.

Our Feed

  • Latest Tweets

    • Data with insight is competitive edge and customer satisfaction and to achieve this, data centralisation is needed… hours ago
    • Today’s problem is that there is too much data to handle and make sense of – noise. #data #bigdata #dataanalytics… hours ago
  • ⚠ Visit the COVID-19 Corona Virus South African Resource Portal here: