Cloud Hosting

Ramana Satyavarapu, Chief Technology Officer, NiumIn a market such as Asia Pacific, the sheer volume of data and various emerging types of data create innumerable complexities for businesses that still require the adoption of data strategies from ground-up. For organisations that have understood the importance of data, they are yet to instil stronger data management practices in the current state of modern array. According to research revealed by Accenture, while only 3 of the 10 most valuable enterprises were actively taking a data-driven approach in 2008, that number has risen to 7 out of 10 today. All of it points to the fact that designing data-driven business processes is the only effective way to achieve fast-paced results and goals for organisations across sectors.

To further decode the nuances of the data landscape, with a special focus on the Asia Pacific region, we conducted an exclusive interaction with Ramana Satyavarapu, the Chief Technology Officer of Nium. Ramana is an engineering leader with a strong track record of delivering great products, organising and staffing geographically and culturally diverse teams, and mentoring and developing people. Throughout his career, he has delivered highly successful software products and infrastructure at big tech companies such as Uber, Google and Microsoft. With a proven track record of result-oriented execution by bringing synergy within engineering teams to achieve common goals, Ramana has a strong passion for quality and strives to create customer delight through technological innovation.

In this feature, he shares his outlook on the most relevant data management practices, effective data functionalities, building headstrong data protection systems, and leveraging optimal data insights for furthering business value propositions.

Ramana, what according to you are the most effective functions of data in the evolution of tech and innovation in Asia Pacific?

Data is becoming ubiquitous. Especially in Asia Pacific, because of the sheer number of people going digital. The amount of data available is huge. I will streamline its functions into three main areas:

First, understand the use case. Second, build just enough systems for storing, harnessing, and mining this data. For which, dont build everything in-house. Theres a lot of infrastructure out there. Data engineering has now turned into lego building, you dont have to build the legos from ground up. Just build the design structure using the existing pieces such as S3, Redshift and Google Storage. You can leverage all of these things to harness data. Thirdly, make sure the data is always encrypted, secure, and that there are absolutely robust, rock-solid, and time-tested protections around the data, which has to be taken very seriously. Those would be my three main principles while dealing with data.

How would you describe the importance of data discovery and intelligence to address data privacy and data security challenges?When you have a lot of data, reiterating my point about big datasets and their big responsibility, the number of security challenges and surface area attacks will be significantly higher. In order to understand data privacy and security challenges, more than data discovery and intelligence, one has to play a role in terms of two aspects - where we are storing it is a vault, we need to make sure the pin of the vault is super secure. Its a systems engineering problem more than a data problem. The second is, you need to understand what kind of data is this. No single vault is rock solid. Instead, how do we make sure that an intelligent piece of data is secure? Just store it in different vaults that individually, even if hacked or exposed - doesnt hurt it entirely. The aggregation of the data will be protected. Therefore, it must be a twofold strategy. Understand the data, mine it intelligently, so that you can save it not just in a single vault, but save it in ten different vaults. In layman terms, you dont put all your cash in a single bank or system. Therefore, the loss is mitigated and no one can aggregate and get ahold of all the data at once. Also, just make sure that we have solid security engineering practices to ensure the systems are protected from all kinds of hacks and security vulnerabilities.

The interpretative value of data provides immense scope for evaluating business processes. What role does data analytics play in the evolution of business success?There is a natural point where the functional value proposition that can be added or given to the customer, will start diminishing. There will be a natural point where data will be the differentiator. Ill give a pragmatic example which everybody knows - the difference between the Google search and Microsoft Bing search, both of which are comparably similar kinds of algorithms. But the results are significantly different! That's because one adopts fantastic data engineering practices. Its all about the insights and the difference that they can provide. At one point, the value addition from the algorithm diminishes and the quality and insights that you can draw from the data, will be the differentiator.

Twofold advantages of data insights or analytics. One, providing value to the customer beyond functionality. Like in the context of say Nium, or payments, or anyone whos doing a global money movement, weve identified an xyz company doing a massive money movement on the first day of every month - say to the Philippines or Indonesia. Instead of doing it on the first day of every month, why dont you do it on the last day of the previous month. That has been historically proven to be a better interchange or FX rate. At the end of the day, its all about supply and demand. Doing it one day before can save you a huge FX rate conversion which will benefit the business in many ways by one quantifiable amount, that is very powerful. Those kinds of insights can be provided to the customers by Nium. Being a customer-centric company, its our strong value proposition - we grow when the customer grows. Those insights, in addition to the business intelligence that can be drawn from it. Offering a new value proposition to the customer and just improving their processes is important.

For example, we are seeing that on an average, these customers transactions are taking x days or minutes, or this customer's acceptance rate is low, then we can improve the value, the reliability, and the availability of the system using analytics. We had a massive customer in the past, none other than McDonalds. We were looking at the data and we observed that theres a very specific pattern of transaction decline rate. However, independently, youll look at it and notice that only a few transactions are being declined. But if you look at it on a global scale, thats a significant amount of money and customer loss. When we analysed it further, we identified that this is happening with a very specific type of point of sale device in the east coast at the peak hour. We sent a detailed report of it to McDonalds saying we are identifying this kind of a pattern. McDonalds then contacted the point of sale device manufacturer and said that at this peak, these kinds of transactions, your devices are failing. That would have saved them hundreds and thousands of dollars.

Saachi, the whole idea is having a clear strategy of how we are going to use the data and we need to demystify this whole data problem space. There are data lakes, warehouses, machine learning, data mining, all of which are super complex terms. At the end of the data, break it down, and its really not that complex if you keep it simple.

In a world continually dealing with new-age data, mention some best data management practices for tech leaders. Again, theres no one set of practices that can determine that this will solve all your data problems. Then youd have to call me the data guru or something! To keep it simple, the three main aspects that I talked about - collection, aggregation, and insights - there are specific management practices for each of these strategies.

First, when it comes to data collection, focus on how to deal with heterogeneity. Data is inherently heterogeneous. From CSV files to text files to satellite images, theres no standard. Find a good orchestration layer and a good reliable, retry logic, with enough availability of ETLs to make sure this heterogeneous data is consistently and reliably collected. Thats number one. Im a big believer of: that what cannot be measured, is whats not done. Measure, measure, measure. In this case, have some validators, have some quality checks on consistency, reliability, freshness, timeliness, all the different parameters of if the data is coming to us in an accurate way. Thats the first step.

Second is standardisation. Whether its web-crawled data or Twitter information or traffic wave information or even satellite images, there was a dataset where we were measuring the number of sheep eating grass in New Zealand - so we were using image processing techniques to see the sheep. And why is that useful? Using that, you can observe the supply of merino wool sweaters in the world. If the sheep are reduced, the wool is less, and therefore the jacket will be costly. How do we store such data, though? Start with a time series and a standard identification. Every dataset, every data row, and every data cell has to be idempotent. Make sure that every piece of data, and the transformations of it, are traceable. Just have a time series with a unique identifier for each data value so that it can be consistently accessed. Thats a second.

Third, start small. Everyone presents people with machine learning or advanced data mining. Those are complex. Start with linear regressions and start identifying outliers. Start doing pattern matching. These are not rocket science to implement, start with them. Machine learning, in my opinion, is like a ten pound hammer. Its very powerful. But you want to have the right surface area and the right nail to hit it. If you use a ten pound hammer on a pushpin, the walls going to break. You need to have the right surface area or problem space to apply it. Even with ML, start with something like supervised learning, then move onto semi-supervised learning, then unsupervised learning, and then go to clustering, in a very phased manner.

That would be my approach on dividing it into the collection - having good validators or quality checks on it to ensure reliability, standardisation in the form of a time series, and then pattern recognition or simple techniques, wherefrom you can progress gradually onto how we want to mine the data and provide the insights.

To summarise, keep the data problem simple. Make sure you have a clear understanding of it - what is the use case that we are aiming to solve before we attempt to build a huge data lake or data infrastructure? Being pragmatic about the usage of data is very important. Again, data is super powerful. With lots of data, come lots of responsibilities, take it very seriously. Customers and users are entrusting us with their personal data, and that comes with a lot of responsibility. I urge every leader, engineer, and technologist out there to take it very seriously. Thank you!

Continued here:

Asia Pacific will lead the new wave of transformation in data innovation: Nium's CTO Ramana Satyavarapu - ETCIO South East Asia