Indian clients see analytics as a commodity when it should be seen as value addition: Bill Roberts

A recommendation engine is not a plug-and-play tool; solving business problems require creative thinking, Bill Roberts tells Ankita Rai

Click here to connect with us on WhatsApp

Recommendation engine is one of the most sought offer practices in big data analytics today. Can it be used in cases of new product launches or instances where data is sparse?

The Deloitte recommendation engine can work in situations where data is sparse. In fact, a lot of inference can be drawn even if not enough data is available. For instance, if you have a new customer for whom you don't have any data, the recommendation engine can look at similar profiles of customers and see their product ratings and accordingly find matching recommendations.
 
Consider the case of Netflix with 17,000 movies and 5,00,000 customers. Four years ago even that was a big data problem as only 3 per cent of possible combinations were available. Also, some clients have a lot of data but they don't know what to do with that. The key to a recommendation engine is to use transaction data- how do customers actually behave - rather than using demographic information. The strength lies in using actual data and then layering it with social data.

What are the key parameters you look at when working on a recommendation engine? Which sectors can benefit the most by leveraging recommendation engines?

The first step is understanding the business problem. The second is looking at what data the client has that can be leveraged to solve the data problem. Third is, using the right model that solves the problem. Not just product-based companies but even companies in fields like oil and gas are using recommendation engines now.

Generally the algorithms can solve a mathematical problem but does not always solve a business problem. There has to be a creative force to understand what problem the recommendation engine is solving - whether it is about driving customer satisfaction or profit. Solving a business problem also needs creative thinking. It is not simply a plug and play tool.

It can be applied in any sector. For instance, telecom can use it to reduce customer churn. It can be used to infer whether the promotion or offer would work for a particular set of customers or not. As of now the e-commerce, BFSI, telecom sectors are the early adopters.

What are the three ways in which recommendation engines can boost investment returns for companies?

Recommendation can help in improving customer engagement, reduce churn and increase sales. Sales are the top matrix for most clients. Second, focus on increase in margins. Consider this: You can easily give 5 per cent discount to all your customers. But are you profitable by doing that? The idea is how do you build in optimisation. So if you know some customers are anyway going to buy it, there is no point giving discounts to them. That's where algorithm comes into picture. You have to identify those customers who are sensitive to discounts and give better experience to loyal customers. There is an interesting example of an online insurance company in the US. The company has an algorithm that helps identify bad-risk customers. The moment it identifies that the customer planning to buy its policy is a bad risk, it gives out ads of competitors selling same insurance at a cheaper rate.

Between collaborative filtering and context-based recommendation engine, which do you prefer?

Collaborative filtering is a general name used for algorithms, which recommends 'if you like that you may like this too'. It is more of retargeting. Product recommendations cannot be simply based on the previous purchase history and buying behaviour or what others are buying or watching. It has to be contextual. The recommendation engine needs to use the right filters.

What are the key problems specific to the Indian market that recommendation engines can solve?

In the Indian scenario, a recommendation engine can be useful in reducing churn, improving engagement and also in cases where you have good products that are not selling. The big problem here is the large customer base. Processing huge data on a real-time basis and posting recommendations is a challenge. Here the efficiency of the algorithm becomes more important. Many Indian companies are at the initial stages and don't have much transaction data. That is cold start problem. The results will refine over time. But even the early stage players can incorporate recommendation engines. The momentum will build on. But don't wait for the data. The experience of the first-time buyer can determine whether that customer will comeback or not.

Also, it is important that the company investing in deploying a recommendation engine should know what the return on investment is. For example, we can quickly tell the clients what percentages of recommendations are actually looked at by customers and how much additional revenue is coming from them. But this type of analytics requires significant investment. That's how Indian clients should approach analytics: Look at it as a business problem first. Do a proof-of-concept study and commit investment only after seeing the business outcome. Recommendation alone cannot solve the business problem. Define the problem first.

Also many Indian clients see analytics as a commodity. On the contrary it should be seen as a service, a value addition.