Do you remember what it was like the first time you got your hands on an iPhone? When you realized that all the things that you used to have to do on separate devices now could be accomplished on one single device? Well, the minds behind LogicBlox would like you to feel the same way about its foundational technology that collapses multiple technology stacks into a unified smart database environment that aims to enable enterprises to create sophisticated and easily iterated applications in one place.
Transaction and analytics co-exist in the platform, with the system utilizing a single declarative language with extensions – such as Machine Learning capabilities and statistical relational models – to support prescriptive and predictive analytics. Users can leverage LogicBlox’ full-blown database functionality to train Machine Learning models, for use in solving forecasting or optimization problems, for instance. It all runs in the Cloud to take advantage of unlimited computing power on demand at low cost, handy for the spikes that can come with big Machine Learning runs.
LogicBlox was in development starting back in the mid-2000s, says Rafael Gonzalez Caloni, the company’s CAO and EVP of Marketing, and long before the Cloud was an accepted enterprise platform – back when Big Data was not quite even a whisper. The idea was to do “paradigm-changing stuff,” he says, not only by leveraging the Cloud, but also by turning on its head the concept of building apps that were glued together conglomerations of multiple technology components, drawing from multiple databases and layers, statistical or optimization tools, and programming languages and skill sets. It could take years to build sophisticated systems that were then hard to change because of fears that their very brittleness would break them. “That stifles innovation in many respects,” he says.
Given its plans to be game-changing, though, the company needed to prove its point. That led to the birth of Predictix, a separate company but jointly owned with LogicBlox. The goal was to pick a vertical that relied on large scale sophisticated applications to drive a lot of value, and the decision was to start with retail. “It’s a fantastic industry to prove we can do something very different with this underlying technology,” Gonzalez Caloni says. Using the LogicBlox platform, Predictix bet on bringing applications to market that help retailers handle challenges like forecasting demand for products, particularly when it comes to promotions and particularly when it’s something they’ve never promoted before, and handling assortment planning across potentially thousands of stores carrying hundreds of thousands of products.
Learning from All the Data
“When dealing in huge companies like [some retailers], the idea that people are going to get their heads around every intricacy is not realistic,” he notes. Compared to traditional methods, Predictix has been able to deliver a 25 to 50% higher accuracy with its Machine Learning approach and forecasting app built on its smart database. Traditional methods include, for instance, time-series based forecasting, which basically is a way of gleaning trends based on looking at historical data.
That’s fine for simple ongoing product forecasts – for instance, stores can easily gauge from their sales history that they sell more bottled water in summer than in winter. But the situation changes when there’s a new product to be promoted and no history to hark back to. You need to be able to extrapolate a promotional forecast from a lot of different data points – not about the particular product that the retailer has never promoted in this or any other way before, but about that product from data on thousands of other promotions of other products in stores across regions and seasons, brought to buyers’ attention via different methods from emails to in-store flyers.
“Our approach says let’s see all that data,” he says. “We let the machine run through every possible scenario of every promotion you’ve done and their multiple attributes, and find a set of patterns to tell you that this product being promoted in this way will likely do this kind of business.”
Say the new product is a blue item, for instance, and that the retailer has promoted many different kinds of blue products before. Machine Learning will determine if color will be a relevant attribute for the new product so that can be considered in forecasts, and if it determines it isn’t a relevant attribute, it won’t. “But it’s a more sophisticated way of understanding a problem and providing a solution for it,” he says.
With LogixBlox as the platform for Predictix, there’s no need to have to try to manage the process outside of the smart database.
“With Machine Learning, where you say my demand depends on the day of the week or location of the store, you give all your data to the system and it will automatically train regression models to be able to answer rest of queries,” says Nikolaos Vasiloglou, who leads Data Science special missions at LogicBlox, including developing the forecasting model with Machine Learning. “We can train the models using the database.”
Thanks to the Cloud’s computational power, Machine Learning models can be more complicated and drive toward better accuracy with modern algorithms that can finely factor in interactions between values, he explains.
Each percentage in forecasting prediction accuracy improvement can “translate to millions of dollars of savings for a client,” he says, because they won’t need to spend as much capital on inventory and there won’t be as much leftover product following a promotion. When there is a lot of extra supply because of less demand in the wake of a promotion, that not only means that retailers have to do more and bigger markdowns to move the supply, but they also wind up losing real estate that could go to more profitable items.
“That means there is an opportunity cost, too,” Vasiloglou says. “So if you are able to do forecasting right and improve accuracy, then it’s a huge savings.”
The Home Depot is one of Predictix’ clients, the company uses its technology for localized assortment planning. It’s provided a way to add science to assortment design, complementing merchandising teams’ knowledge and experience, Aaron Surasky, Senior Director of Assortment Planning and Analysis at The Home Depot, said at NRF 2016. “It’s a core bedrock of how we do things at Home Depot,” he commented during his presentation. Indeed, Gonzalez Caloni says, putting Predictix to work is not about supplanting merchant teams’ judgment. “We’re not taking away from the role of the merchant in the assortment process but giving them insight they simply wouldn’t have time or computing power to get,” he says.
Self-Service and New Sectors on Tap
Continuing the iPhone metaphor, Gonzalez Caloni points out that Apple’s revolution moved from collapsing multiple functions into one environment to releasing its SDK that let millions of developers drive further innovation by creating their own apps. To date, the focus for LogicBlox has been on companies leveraging the platform for corporate-defined applications, but the real difference maker is opening up the technology to domain experts themselves. “It’s the last big piece of the puzzle,” he says. Now that the company has collapsed together components to make it possible for people to build highly sophisticated apps in a more efficient way, the goal over time is to add a layer to support self-serving users, he says.
Retail doesn’t provide the only use cases for gaining benefits by taking advantage of the LogicBlox platform, the company notes. Planixs uses the smart database to quickly build flexible apps running multiple simulations for project portfolio management and optimization and intraday liquidity management for global banks, for example.
That said, retail is going strong: Infor also recently invested $25 million in Predictix and will infuse its technology into its development of Infor CloudSuite Retail, a set of enterprise applications delivered in the Cloud for today’s retailing landscape that was announced in October in collaboration with Whole Foods Market. As Gonzalez Caloni notes, retail has been the ground for battle testing. But LogicBlox is starting to look at other partners that might take its underlying smart database technology to build apps for other industries. “There are domain experts in other industries who can take the powerful underlying technology and build applications to serve those industries,” he says.