Indicia Worldwide and Data Reply Reduce Data Model Development and Deployment Time to Weeks Using AWS

Indicia Worldwide helps global brands improve their customers’ experience using data, machine learning, and AI to fine-tune marketing campaigns, improve retention, and drive new customer growth. It turned to AWS Partner Data Reply to help evolve its data science platform to a more robust, scalable, and efficient solution on AWS. This move cut the time spent building, testing, and deploying new data and machine learning models from months to weeks.

Indicia Worldwide works with the world’s top brands to activate their marketing insights, using machine learning to deliver growth and improve customer experience. It uses Amazon Web Services (AWS) to accelerate the development of bespoke machine learning models to help clients use their customer data so they ensure that their consumers receive the right messages at the right time.

The company needed to accelerate its machine learning capabilities to deliver relevant, omnichannel campaigns for its clients. Previously, a high level of manual input meant that models were built, deployed, and tested too slowly, resulting in frustration for internal teams. So, Indicia Worldwide made it a priority to keep up with advances in machine learning to ensure that complex models could be developed, tested, and brought to market quickly and successfully.

Indicia Worldwide knew it needed to make changes and researched the tools available. That research led to Amazon SageMaker. “But because this was such a critical project for us, we wanted a partner with the in-depth knowledge to work at the speed we required,” says Graham Lannigan, Head of Data Platform at Indicia Worldwide.

Indicia Worldwide chose AWS Partner Data Reply for its deep knowledge of AWS and Amazon SageMaker. “We really worked collaboratively, and it was far more than a supplier,” says Nil Patel, Global Data and Analytics Director at Indicia Worldwide. “The team rose to all of our challenges with creative and practical solutions that enabled us to get value from our efforts quicker.”

Amazon SageMaker can be used to build, train, and deploy machine learning models for any use case, with a fully managed infrastructure, tools, and workflows. This means that data scientists can forget about infrastructure and focus on customers. As a result, machine learning development can become a true part of an organization’s operations, through applying the principles and best practices of DevOps to machine learning—an approach known as MLOps.

Data Reply first carried out an MLOps capability assessment to get a clear view of what capabilities were already in place and what the different stakeholders expected. The assessment also reviewed existing documentation, technology, and development processes before making recommendations for a new operating model.

The key to getting the most value from machine learning is to use the best lessons and practices of modern software development. That means streamlining processes, automating as much as possible, and following DevOps best practices. The capability assessment helped everyone understand the work that needed to be done and the gaps in process and technology that needed to be filled.

Data Reply used AWS CodePipeline, which automates the build, test, and deploy phases of the release process every time there is a code change. “The platform that Data Reply built for us has massively reduced our time to value,” says Lannigan. “What used to take months, we can now do in weeks. It’s not just about the models. It’s about testing, deploying, and scaling, as well as integrating them into existing systems.”

Building the model is only the first step of a successful ML project. A model must also be deployed, tested, and integrated with existing systems. After that model begins to be used, it must then be scaled up. Using AWS and its infrastructure-as-code capabilities means that deployments can be automated and more easily replicated. This also simplifies monitoring and refreshing model performance over time. “Building the models is just the first hurdle,” says Patel. “We also have to test, deploy, and scale them fast. The flexibility and scalability of AWS lets us do that without thinking about infrastructure. We can build better models faster.”

Indicia Worldwide has much better visibility now that everything is running on one system. Data Reply has also automated many of the manual tasks required to deploy and scale Indicia Worldwide’s models. This is a repeatable automation, so processes can be reused across client projects.

Quality assurance—testing of both data and code at different stages of the pipeline—has also been improved, with issues automatically flagged as they arise. “We’ve also built-in better-quality assurance,” says Luca Piccolo, Manager at Data Reply. “We are testing both code and data at several stages of development to keep projects on track. This testing is all automated, which saves staff valuable time.” This capability helps overcome a common failure of machine learning: projects that never make it into production.

The changes have made working life better for the data team and moved it closer to a continuous integration and continuous delivery (CI/CD) development model. More rigorous version control, fewer outages, and less downtime have removed infrastructure frustrations. The improvements to the development and documentation process make work easier to manage if employees change roles and someone else has to pick up a project already in progress. The strength and stability of the system has also given the team more confidence to experiment and innovate.

Moving to AWS has given Indicia Worldwide a more transparent system, which makes it easier to continue to develop and improve the models that it builds. By working with Data Reply and using AWS, the company can now innovate faster and move toward offering customers real-time, reactive systems to maintain, train, and retrain models while they are in use.