Data Science at OnDemand Agility

Demystifying Data Science

Data Science has become a blanket term for all manner of technological and, often, aspirational ideas that fall in the general bucket of advanced data analytics and insight. In order to put this exciting and truly innovative new toolset (and it is just that – a toolset) to good use, it is important to understand what it is and, perhaps more importantly, what it is not.

Any data process seeks to perform the same fundamental operation – convert data into information, information into insight and insight to action. As an example,

• A list of a website’s customers and their orders in separate excel sheets are data.
• To convert this into information we must clean and de-duplicate, associate customers with orders and create summaries of which customer ordered what and how much etc.
• Insight is achieved by plotting trend lines between the regions (group customers by addresses) and time of the year so we can make statements of the sort “customers in the state of Nebraska buy twice as much product in November compared to any other month”
• This insight can then be converted into an action if it can achieve a business objective. E.g. one of our objectives can be avoiding bad customer experience by running out of stock, so the action we take is making sure we stock up extra for November in Nebraska

Traditionally, Data Warehouses and BI technologies have helped businesses get insight into their data. However, these technologies are limited in a very crucial manner, they can only answer questions that we are already aware of. In the above example, we knew that we wanted to see sales trends by region and time, therefore we constructed a data warehouse (or a simple MS Access database) to serve that specific reporting need.

This is where Data Science enters the arena as a game changer – it has the ability to answer questions that we don’t yet know to ask! Sounds like magic – doesn’t it? Rest assured that it is anything but. At the heart of the Data Science toolset are sophisticated statistical modeling paraphernalia that can discover correlations between data sets hitherto unknown and therefore, obviously, not reflected in the structure of the data at all.

To continue our example, let us say we choose to correlate (don’t ask why) how the sales in Nebraska were related to the average daily temperature. As it so happens, we hit a positive correlation that was previously unknown. We now see that the reason customers are buying more product in Nebraska in the month of November is because temperatures have been dipping precipitously in that month for the last 3 years – which is coincidentally the length of time we have been in business. We also see that this is not true of any other state in that same time period. There would have to be more work done, of course, before we could declare this insight actionable – we would have to create test models to confirm, perhaps look at past or alternate data sets for validation but at the end of it, if this insight turns out to be accurate we can refine our action for even better business outcomes – not only will we stock up for Nebraska but for any other state as well where similar weather patterns may manifest. In fact, we may decide to not stock up for Nebraska if the weather this year is reliably predicted to be mild and we would save a ton on inventory carrying costs.

Data Science as a discipline

Data Science, then, is a discipline that allows us to discover new insights about data that were not available to us through traditional data processing techniques and convert these insights into actions. It requires a masterful blend of three distinct skill areas

• The business domain – we must know enough to be able to imagine where to let our modeling toolset loose (in our example, we at least needed to be aware that we sell weather ware and that climate is likely a factor)
• Statistical modeling and the math that goes with it – this piece is critical. Without a solid basis in statistical modeling theory, we would have no idea what modeling techniques are appropriate to our situation – from the many options available. Also, we would have no ability to interpret the model outputs
• Technology – how do we implement the models we dream up? Again there are a variety of choices ranging from R/Python to cloud based pre-packaged APIs and services from AWS, Google etc. It is important to know which is appropriate given our unique set of requirements.

Data Science at ODA

As an organization we define ourselves as a domain-led technology solutions provider. We insist on a deep understanding of the businesses that we serve – be it the broad range of financial services i.e. consumer, private, investment banking, capital markets, asset / wealth management etc. or specific core areas like trading, core banking systems, risk management and regulatory control. Being a technology firm we are well versed in the third leg of the stool, as it pertains to data science. We have made active investments, through strategic partnerships and internal ramp-up, to build up our expertise on the statistical modeling side. This enables us to unlock this truly remarkable toolset for our clients and bring to bear the promised value – be it through more efficient operating models or enhanced revenue opportunities.

For any of our case studies, feel free to reach out to us at IP@ondemandagility.com

-Sandeep Singh Ghuman, Chief Technical Officer