Operations Research as a Data Science Problem.

Operations Research as a Data Science Problem.

By Matthew Eric Bassett,

Data Science extracts knowledge from data. And most of this decade has been a’buzz with discussion on how businesses must learn to extract value from their data or be left behind.
Yet many businesses (and even entire industries) are not seeing the expected return on investment from data science.

There are, broadly speaking, two general approaches to data science:

1) To make a data product (or a data-feature of an existing product, exempli gratia a data-driven recommendation engine to suggest which film a viewer should stream next).


2) To visualize and communicate the data to inform or guide some decision.

Many readers can come up with examples of the first case, where the extracted knowledge is baked into some product for mass consumption. But examples of the second type are more difficult to find, which is unfortunate, because many industries can only apply data science in this way.

In these second examples, knowledge requires context – a framework for understanding business, inventory, process, and customers – to become valuable.

Yet, there’s another name for this idea that advanced analytical methods, like markov-chain monte carlo or neural networks, should help drive better decisions, and it’s a bit older than the big data hype. It’s operations research, and it can provide that context.

Operations Research got started during the second world war. In fact, some of its early successes resemble many of today’s “data-driven” discussions. For instance, the Royal Navy Coastal Command’s Operational Research Section did A/B testing to determine the most effective color of anti-submarine aircraft (to approach surfaced u-boats in the daytime) and even the trigger-depth for depth charges (apparently, they had more enemy kills if the depth charge detonated in shallower waters).

After the war, many people became aware that one could find marginal-to-massive improvements in procurement, logistics, infrastructure, et cetera from a careful analysis of the data involved. Indeed, in modern times, Operations Research is defined as the application of analytics to make better decisions – suggesting that it should walk hand-in-hand with the latest trends in data science.

So exactly how should they work together? Operations research should be a “metal detector”, guiding one to the right area of the business, whereas data science is the shovel to dig through the data and get to the value.

For instance, consider the operations of a movie cinema, where a manager needs to book films and schedule their screenings in a way that turns neighbors into guests and maximizes cinema admissions. In fact, three years before the Harvard Business Review declared that “data science is the sexiest job of the 21st century”, operations researchers already identified that cinema scheduling could account for a 3.1% increase in cinema admissions [1].
This increase came from a scheduling methodology and demand forecast based solely on past data.
A “big data” approach might try real-time demand forecasting from social media, traffic, and other factors.
And this demand forecasting need not stop with the schedule – a smart-phone-and-app equipped workforce could be clued into changes, and management could more accurately predict staffing needs. It could go further still, and data scientist could match local audience preference with film sentiment, and use the knowledge gained not just make a schedule, but to decide which films should be booked in the first place. [2]

Much of the room for operations research in a cinema comes from scheduling, booking films, and staffing, as the guests have only a few interaction points while they flow through the cinema. But consider instead a theme park. Guests here spend their entire day interacting with the business, and the internet of things and wearables allows data scientists to predict guests’ needs before the guests themselves do. Indeed, Disney does just that, and its another area where data science and operations research meet. [3]

[1] Eliashberg, et al, Demand-driven scheduling of movies in a multiplex,
International Journal of Research in Marketing (2009)

[2] The author works on these problems, and others, in his company Gower Street Analytics, in London.

[3] https://medium.com/re-form/welcome-to-dataland-d8c06a5f3bc6

You may also like...