The rise of big data and machine learning has created a booming market for solutions providers that can provide businesses with detailed analytics and insights. But how can solutions providers reliably deliver these solutions at scale? And how can these providers also ensure the rationale of their model deployment was comprehensible outside of engineering and data science teams, and thus secure sell-in from their clients?
These have been considerable questions facing Copenhagen-based Noitso, and ultimately drove the team to adopt Seldon’s technologies.
Founded in 2007, Noitso has enabled their customers to make vital business decisions accurately using their dynamic solutions to work with big data and AI. They are experts in data science, data collection and making data-driven predictive analysis of the future.
Their customers have use cases ranging from inferring budgets and credit rating, to enterprise data management and scorecards with machine learning used to capture high risk events.
What was the challenge?
Data Scientist Thor Larsen and Project Manager Emil Jensen work together in the data science team at Noitso. Their mission is to use data science and AI to provide their customers with credit ratings, scorecards and risk profiles. Model deployment needs to be accurate to ensure customer success and demonstrate value to potential customers.
While taking on this work Thor ran into some challenges. Models would take a long time to get to production and were a black box to users.
“Our models were simply built into C first and then added directly to the applications, it took time and there were too many errors in development.”
Using these existing processes, Thor and his team lacked some explainability and monitoring. They were unable to get data on when models needed to be retrained, the aim was to do it after a fixed period of time rather than when it was necessary. This was the only option to retain accurate predictions and prevent issues such as data drift.
Seldon gets involved
Thor’s plan was to use development best practices to ensure everything was right, manage code and collaborate with data savvy colleagues. They were developing in Python and needed a native environment in containers.
“There’s a lot of tools out there that we evaluated. But not having input transformers and a flexible infrastructure would be a huge loss.”
Upon introducing Seldon Deploy to their MLOps stack, Thor and Emil began to have faith in the way that they could deploy their machine learning models they wanted into production.
Existing customers who began using this new, faster and more accurate AI continue to be very impressed and Noitso are now rolling this out to more customers shortly. With this new speed to model deployment, Noitso will soon be able to prove their models in the POC phase and that they are reliable in production.
“With Seldon Deploy’s monitoring options utilising Alibi Detect, we can with greater faith run our models in production and see what is going on”
What’s in store for the future?
There are more challenges around the corner. Tighter industry regulations means that Noitso has to remain compliant and they must be able to explain how their models arrive at the decisions they make. This is where Seldon’s explainability features come into play, to future-proof organisations from a model’s potential risk.
As the leading open source MLOps provider, Seldon’s technology is perfect for consultancies like Noitso that want to help improve compliance and explainability, while also improving the speed and reliability that consultancies can leverage machine learning to provide cutting-edge insights for clients.
If you would like to read the full case study, you can download it below.
Want to hear from the Seldon team? You can request for a demo here.