Achieving Analytics at Scale & Speed
Vivek Chaturvedi
Leader in Advanced Analytics and Data Science | Retail, SME, Corporate, Treasury, Transaction Banking & Wealth Management | India & MENA | Thought leader | Public speaker | IIM Bangalore
Many BFSI organizations want to adopt data analytics and use it increase the bottom line. However, it is imperative to understand that Data Analytics is a strategic investment for any enterprise and it will yield really good results only over a period. In order to achieve those results, organizations need to make the right investment in right time. Otherwise, they will lose in the highly competitive business environment for banks as exist in India and many other geographies. In this article I will try to lay down a path that banks can use to achieve the ultimate dream – Analytics at Scale & Speed. A word of caution is necessary at this stage. This article is intended to guide you in the right direction and the details of execution require multiple small decisions which will vary from one organization to another and the right answers to those questions will also vary.
The holy trinity – there are three basic components of any analytics-based solution. These are often referred to as “layers.” These three layers are: data, intelligence, and execution.
Data - Financials services as sector has maximum number of transactions per rupee of its income. Banks, capital markets and insurance firms can conduct business in a very small denomination and multiple times a day. A typical individual or enterprise makes multiple payments in a single day ranging from a few rupees to buy office supplies to a single payment of few crores. Contrast this with a car company which sells a few thousand cars a month[1] with one car costing a minimum of 4 lakh INR.
Banks have multiple non-monetary transactions in a day. There are thousands of employees marking their attendance, applying for leaves etc. There are customers who are raising request for issue of cheque books. Investors converting the saving account balance to fixed deposits and so. Value can be derived from data when you combine data from various sources. In the language of business these are called Source Systems. A more technical name is Online Transaction Processing System (OLTP). A typical bank can have 100 to 300 such systems and each one of them has data of past 2 to 10 years.
It is therefore very important that the bank has a good Data Warehouse and has stored data in such a manner that they can retrieve it quickly and efficiently whenever needed. Data is the first building block in the analytics strategy of a bank and I have seen analytics teams struggle for weeks to get data. This data is shared with them as MS Excel files and is often inconsistent with similar data from other sources. Because of this teams spend even more time in data reconciliation. Thus, it makes the entire process of data gathering extremely slow.
Banks should have list of all Source Systems with the details of the tables and the fields to decide which fields to save and of what duration. Something called a Data Dictionary. This may sound simple, but it is a very time-consuming process and needs wise judgement. Thankfully the cost of data storage has gone down in the last few years. But as I wrote in the beginning its an investment. Actions of today will give results tomorrow.
Secondly, if the data size is big, as it should be, one cannot run the code in personal laptop or desktop. Typical laptop has a 16 GB RAM. However, if the data is stored on cloud, the RAM can be of 64 GN or 128 GB, which increase the processing power multi fold.
Intelligence – People have often asked me what is the scope of data analytics of data science. In order to answer that question, I always present data analytics as a subset of decision sciences. Data analytics allows humans to take correct decisions and those decisions can be of a wide variety. I have listed a few questions below as examples and these are in no way exhaustive. In fact, these are far from exhaustive and the questions which can be answered will keep on evolving with the evolution of data science.
1.?????? Who should we give loan to?
2.?????? What price should we charge for a particular product being sold to a particular customer?
3.?????? Which employee is likely to leave?
4.?????? Should we try to retain them? If so, how?
5.?????? Which business segments we should grow and how?
Other dimension of the decisions is the extent to which they can be automated. On one extreme can be a simple reporting which is used by humans to take decisions which are macro and nonspecific and on the other, those which are completely automated and implemented for a specific question or issue. A good framework to follow is categorization of analytics as: descriptive, diagnostic, predictive or prescriptive[2]. This ability to answer questions and take decisions is what I refer to as the Intelligence layer.
So, how to build that layer? Banks need to build a team of business analysts and data scientists to explore the applications and develop those. The trick is to hire the right skill set at the right time. As data analytics is an evolving industry, different organizations may use different nomenclature for similar work and same nomenclature for strikingly different work. These articles provide a good insight into such roles[3].
领英推荐
Let us leave the nomenclature aside for a moment and define the skill sets that are needed to man this layer. Banks that are setting up this division need someone who can: understand products, interact with various departments like product, sales, credit etc., work with data to summarize it, seeks patterns and perform tasks like model development, clustering etc., present the results of analysis to people across hierarchy and ensure that it serves their queries and concerns and if accepted deploy the models/ rules for use by a large number of people. Needless to say, these are too many skills for one person to have and thus, banks should draw permanent or temporary staff from various departments.
These people should try to identify use cases of data analytics which solve their immediate problems. I have previously posted about how to identify data analytics use cases for business banking[4]. The same framework can be used for use case identification in other areas of banking and other financial institutions. The key is to keep on identifying newer areas, trying those with the help of data stored in the data warehouse and check if it works with the help of implementation mechanism (described in the next part of this article). Using this approach, the teams can build up an inventory of useful solutions over a period and achieve analytics at Scale.
Implementation – as they say the proof of the pudding is in eating. Having identified and developed a use case, the bank should start a pilot implementation in a select region or a customer segment. Management of the bank needs to give a serious thought to – “Why should the users experiment with the solution and give feedback?” For the analytics team and the project team members, it is their core job to design and implement use cases. Not so for the larger audience. If you want them to adopt the solution then you need to create two things – incentive and ease of use.
Many a time good projects fail because leads were shared with relationship managers on email in excel sheets and project team could not keep a track of what happened to those leads. Banks should invest in a good Customer Relationship Management (CRM) platform or a digital platform (in case the solutions are designed to be used by the end customers). Such a platform will allow to capture the user feedback which is very crucial for improvement of the solution.
As I have been highlighting that these activities are investments and that means that the results will accrue over a period and if the solution is implemented as scale, everyone will benefit. The questions why should a few people in the team spend efforts for something which will benefit them after a few months or years, and, it will benefit everyone – while a few put the effort in refining it.
This creates the need for instant gratification of the users in the pilot project. And this gratification should be substantial to more than compensate for the opportunity they will lose if they spend time in an experiment. Various teams have various incentives, the most common ones are money and recognition.
Conclusion
The three layers of analytics strategy can also be viewed as a virtuous cycle. Once a solution has been implemented, it allows the bank to gather more Data and that in turn improves the Intelligence.