Vivek Anand, Director of Advanced Analytics at a Fortune 500 Retailer, speaks with Derek Strauss, Chairman Gavroshe, and Editorial Board Member, CDO Magazine about data science and AI case studies across industries, and the two components for successful adoption of data science.

Anand turns the discussion into a case study treasure grove, as he gives multiple examples of data science and AI case studies across industries. He recalls how in B2B pricing, business segmentation done for price optimization was based on the expertise of business SMEs.

Further, the models were built in a way that had to be segmented based on customers’ opinions. Later, data science started becoming adaptable and people opened up to trying data science.

In the first case study, Anand mentions listing attributes from end customers and running an exploratory data analysis to find out how the attributes affected the target variable’s pricing. He shares that the business’ assumption that geography does not impact pricing changed after he started analyzing pricing in different states across the U.S.

This further made it possible to identify patterns that the business had previously dismissed as unimportant. Data science allows assessing such attributes to bring out the value.

Additionally, data science also impacts the scale and speed of it, says Anand. With new data science techniques, one can run a quick regression that will give a feature importance to the target variable’s price. For instance, from a laundry list of 100 variables, feature importance portrays the relative importance of the features which can then be plugged into the segmentation model.

The other aspect of data science is finding hidden patterns, says Anand. For example, he recalls working with a painting customer who segmented the business into three attributes having different prices for contractors, builders, and DIY customers.

In this case, when the salesperson’s pricing for builders did not match with the customer’s, Anand did affinity propagation on a set of data to find distinct clusters of customers in customer data. Through affinity propagation, it was clarified that the difference in pricing arose because there were two different kinds of builders whose building scale varied.

Consequently, machine learning and data science helped in identifying the different groups of builders and Anand could cleanse the data and bring in good attributes to add value.

Next, he states that the other use case of data science is forecasting. Sharing a gas station example, he mentions that there are five refiners selling gas at nearly the same price. The seller with the lowest price would sell out first and the one with the highest pricing would sell out last.

The objective, Anand adds, is to be priced high enough to sell out last, and that is a complex forecasting problem, forecasting demand and competitor pricing. Such problems have been resolved by building deep neural networks to precisely predict the competitor’s pricing and then fix a price based on that.

Such optimizations did not exist before data science, asserts Anand. The next example revolves around pricing analytics, wherein he advised a SaaS product builder on how to price it. Referring to the low-balling effect, Anand affirms that people usually quote a lower price for a product than what they could actually pay.

Therefore, he built Van Westendorp’s price sensitivity meter which would ask multiple questions regarding price to assess one’s buying capability. The questions include:

What price should make one think that it's cheap and of bad quality?
At what price would it be a good deal?
At which point would one feel it is expensive but still buy?
At what price point it would be too expensive to buy?

Now, it was evident that as prices keep increasing, the percentage of people considering it cheap go down, whereas, expenses go up, says Anand. That is how survey analytics can tell the range of acceptable prices where it is not too cheap and not too expensive.

When asked what helped in successfully overcoming the data science adoption challenges, Anand states that explainability comes first and then change management. Taking the oil and gas industry he worked with as an example, he notes that it did forecasting by using a relative time series model that accounts for seasonality and correlations.

But with the deep learning model or by using transformer architecture, it revolutionizes the way forecasting is done, says Anand.

The uniqueness of transformer architecture is that it has an attention mechanism. It can provide context to the content. Therefore, it can capture temporal patterns. So the forecasting model that Anand has built using deep neural networks, or transformer architecture, the forecasting will be far superior. ChatGPT is built on transformer architecture, he adds.

Shedding light on change management, Anand states that if the objective behind incentivizing salespeople and business objectives to implement data science analytics do not align, it will cause problems.

Illustrating one last example, he mentions that after building a pricing model for a customer, the business sponsor pointed out that the model was not being adopted well. Looking at it holistically, Anand found that salespeople who were on the lower end of the quartile were getting incentivized on the number of sales and not on profits.

Addressing that, Anand states that offering lower prices to get the deal will only make it a race to the bottom and the business objective is to maximize revenue. This is where change management comes in to guide the business through the change, he concludes.

CDO Magazine appreciates Vivek Anand for sharing his insights with our global community.

Scotiabank, Chief Data Officer: The Next Frontier in Data (Part 3)

Related Stories