Our talk at AI Congress 2018

Our talk at AI Congress 2018

Our head of machine learning Eddie Bell spoke at this year's AI Congress, which took place at the 02 Arena in London earlier this month. Bell spoke to a packed audience on the functionality of micromodels, the types of fraudulent transactions that are flagged at Ravelin and the importance of applying software engineering principles to ML model deployments.

One of the key points that Eddie focused on was the power of scalability and how, at Ravelin, we’ve been able to grow our customer base with businesses scaling quickly and develop new models without drastically growing our team. 

“Our clients have grown exponentially but we can make our team remain small. This is done with machine learning.”

ravelin-team-growth

The happy path

At Ravelin, we like to deploy the happy path for our clients which is are a set of simple, interpretable metrics that cover all failure cases.

Bell spoke about the model around the path: “If a model passes these tests then it’s on the happy path and can be deployed automatically. It can be relative to the previous model - for example if the performance has increased.”

“Humans are only involved when things break.”

ravelin-ai-congress-talk

Eddie discussed the importance of the happy path, along with the most important things we’ve learnt:

  1. happy path deploys are easy

  2. the more you deploy, more models stay on the happy path

  3. the easier it is to deploy the more you will deploy

Bell also spoke about some of the most frequent transactions we see and stop and Ravelin. Could you guess what the least fraudulent meals to order were?

Bell concluded his talk by giving the audience advice on how to scale your business efficiently without impacting team growth.

"If you want to scale your business then deploy often, quantify what good looks like, automate everything and choose simplicity over performance."

For more content, follow us on Twitter.

Subscribe to the Ravelin Blog

For the latest in fraud prevention, machine learning, artificial intelligence and graph databases, subscribe today.