Connecting with your customer

ML Ops in action

Nigel Cummings

At our 2022 Annual Conference (OR64), Dr Bettina Schirrmeister. Head of Data Science Marketing at MoneySuperMarket (MSM), gave a plenary talk focusing on optimising the effectiveness of MSM’s marketing strategies and designing an action plan had significantly improved MSM’s business.

Schirrmeister related that in 2001, William Cleveland designed an action plan for expanding data science in the field of statistics. He saw a need for companies “to have a lot of statisticians” working for them in their data science departments, but statisticians did not have the tools to utilise the data and the computer scientists did not have the statistical knowledge needed to analyse the data. So, he proposed a teaching strategy that could transform statisticians and computer scientists into data scientists, equipped to understand data and to derive insight from it.

By 2012, considerable interest had grown in data science, everyone, it seemed, wanted to do it. The luminary Tom Davenport dubbed being a data scientist as having the “sexiest job of the 21st century”.

Moving to the present-day, Dr Schirrmeister’s timeline presentation on data science referred to increasing use of cloud- based and edge technologies, and an increased emphasis on modelling data to gain insight. She said she wanted to explain why there was a need for ML Ops.

The typical data science workflows are: research and data exploration, data processing, data modelling, training, model evaluation and “digesting and understanding” the business problem. Once appropriate modelling have been selected, it is then necessary to continually monitor the  business problem and be prepared to use its outputs to help upscale the business processing as required.

Data science projects can fail because sometimes it is not possible to get permission to use the data required to solve the problem. Sometimes the data might not be accessible, or it may not be clean data because it has been manipulated on other user’s machines and then resaved in modified formats.

As if the data issues are not a sufficient problem what also has to be considered is, deployment and scaleabilitity. There are challenges in maintaining ML lifecycles cross functionally. Many dependencies required constant re-evaluation. There can be cross functional language difficulties too, where different ‘languages’ are used across teams. It is important too, to maintain trust within and across teams.

It is also important to connect with your customer – to create incentives with them; to facilitate their customer journey; to keep them engaged; to retain them as customers and; to see that the product you are offering them is relevant to their needs and pricing investment.

Bettina spoke of how MoneySuperMarket uses Bayesian inference techniques and regression models to estimate returns on investment. It also has an investment modelling system which can help determine customer segments where business opportunities have been missed.

Dr. Bettina E. Schirrmeister’s presentation will be available on The OR Society’s YouTube channel very soon.