Filling in the Gaps with Fractal Analytics

Published By : Datanami

Srikanth Velamakanni, the founder and CEO of Fractal Analytics, has a pet peeve. “Over the last 15 years I’ve been in this business, the one thing that frustrates me the most is companies make so many mistakes using faulty analytics,” he says. “People use analytics for political reasons, or they make some very stupid mistakes in terms of how they’re thinking about the problem, and make huge errors in their decisions.”

Just as Hippocrates encouraged medical doctors to do no wrong, Fractal aims to help companies avoid errors in analytics. That business is booming in that regard is not surprising, Velamakanni says.

“My role has been to work with companies to make sure they’re setting it [analytics] up correctly,” Velamakanni says in a phone interview last week. “But over the years, as data volumes have increased and it’s become harder and harder to put all of these things together, I think that people are definitely losing control over the quality of their decisions with the analytics. There are the exceptions. But for most companies, it’s fair to say they are currently really struggling to pull the piece of data together in an industrious manner and make decisions.”

When he founded Fractal Analytics 14 years ago, the term “big data” was just a gleam in a marketer’s eye. “We started doing analytics way before it was really popular or it was called big data or whatever,” Velamakanni says. Over the years, he has created a successful business by building and deploying predictive analytic tools for some of the biggest companies in the world, including Fortune 100 firms in the consumer packaged goods (CPG), financial services, and retail industries.

Today, Fractal Analytics employs more than 600 workers around the world from its base in Jersey City, New Jersey, giving it more than 3,000 person-years of analytics experience. The company’s 300 or so customers come to Fractal not only for the technology and industry-specific expertise it can provide in the areas of customer analytics, pricing, marketing analytics, risk analysis, and supply chain analytics, but also for the way it can bring experts from diverse backgrounds like statistics, machine learning, auditing, sociology, and psychology to bear on a business analytic problem.

The company’s data harmonization product, called Concordia, can help the analytics process by cleaning up the data. After all, without good data, even the most powerful and precise analytic tool is essentially worthless.

Concordia brings together capabilities you might find in ETL and master data management (MDM) products. “It’s a bit like ETL in the sense that it’s solving exactly the same problem,” Velamakanni says. “But we are looking at data that’s very, very unclean, much more unclean that a traditional ETL tool would handle. We get lots of flat files with little or no information about it.” The software also has the capability to work with data in more than 100 languages, which is something a traditional ETL would struggle with.

Concordia is used by companies that find large gaps in their data is hurting their capability to perform analytics. “A company might want to understand and analyze their data on a weekly basis, but some [subsidiaries in other] countries are only reporting on a monthly basis. How do you convert monthly information to a weekly basis? Traditional ETL technology tech doesn’t do any of that stuff. But we use a lot of algorithms to intelligently interpolate, extrapolate, or divide the data in such a way that it can answer the question that is required to be answered. There are a lot of interesting algorithms in the product that do this job quite well.”

The company’s other main product, Customer Genomics, is used to deliver a multi-dimensional view about an organization’s customers and their views, interests, and activities. Just like Concordia, the Customer Genomics product can run on Hadoop, if the volume of data demands tremendous scalability. (TIBCO’s Spotfire product is also brought in as a visualization layer for the products.) And just as Concordia can be used to cover over bad patches of data, Customer Genomics can keep analytic projects from veering off into fantasy land.

According to Velamakanni, the traditional customer segmentation methods used for decades have almost no bearing on reality. “Essentially the idea that we can standardize all the people in the world into 10 or 20 neat boxes and they’ll behave that way for all purposes is very faulty,” he says. Fractal tackles the problem of accurate customer portrayals by getting lots of data, and then running its “secret sauce” algorithms against it.

“The idea is, how do we uniquely label customers so that we understand them on a multi-dimensional basis, and not on just one axes in a segmentation schema,” Velamakanni says. “That’s what Customer Genomics does. It looks at all the transactions from a customer–what they see on social media, they’re browsing behavior, the devices they use–and starts to put various kinds of labels against them.  Maybe you’re an expert in photography, or you like organic foods, or you have kids, or you’re thin. I can find out all these things based on your transactions, and using this information, we can then hyper-customize the customer experience.”

But don’t mistake this for another episode of “Algorithms Gone Wild.” Without human supervision, the machines can’t be trusted to generate accurate answers against new data. Just as you wouldn’t make a $1-million bet on the hunch of a human analyst with little experience in her field, machines need training, too. And despite what some in the industry have said, the “bigness” of big data is no substitute for knowing what to look for, and what to ignore. “If you’re looking at new data sources and choosing them for the first time, you don’t know what to expect, so you don’t have a sense of what’s valid and what’s not,” Velamakanni says.

Fractal Analytics’ technology essentially is the synthesis of years of human experience solving business analytic problems in CPG, financial service, and retail. It’s solving the old “garbage in, garbage out” quandary, at big data scale. At the same time, it’s paving a way for machines to take over the jobs they are good at, while simultaneously keeping humans and their often flawed thinking away from the analytics as much as needed.

The goal of Fractal Analytics is to make analytics error free and less dependent on people and more dependent on processes and technology. “That’s the big drive we’re seeing,” Velamakanni says. “We never like to [have humans] do things that machine are capable of doing better. That’s what we’re seeing all over the in analytic world. When you’re playing with a lot of data, I don’t think any company in the world has a foolproof way of fixing errors that creep into analytics. This is a way in which we’ve addressed the problem. By creating a visual workflow and by automating it, we’re reducing the variability and the human error that creeps into any analytics exercise.”