kirill_makarov - stock.adobe.com
What's behind the Domo machine learning tool Did You Know
Domo's Did You Know feature uses machine learning algorithms to automatically pull analytics information out of data sets for business users. Here's how it works.
Working daily in the world of data and analytics, I find that Domo is one of the better known but least understood BI vendors. Plenty of people have strong opinions about the company's CEO, initial public offering and financial performance, but it's rather harder to find people who have looked at the Domo platform objectively or perceptively.
In my last "behind the platform" article, on Yellowfin's Signals software, I noted that you often can sense a vendor's organizational structure surfacing through its software UX. Looking at a company like Domo, with celebrity CEO Josh James, you might expect that to be particularly true. After all, Domo was founded in 2010 specifically to answer business questions that James, in his previous role as an executive rather than a technologist, couldn't effectively address with other BI and analytics tools. And Domo today still focuses much of its marketing efforts on bringing insights to the C-suite.
However, one reason I find Domo so interesting is that the technology behind the executive-focused features in its cloud-based platform is often sophisticated and challenging. For example, the company offers a suite of Domo machine learning, AI and predictive analytics tools cutely named Mr. Roboto in a nod to James's fascination with Japanese pop culture and the old Styx song. However, Mr. Roboto is no toy -- it integrates AI and advanced analytics technologies into high-level features of the Domo platform.
One of the latest features added to Mr. Roboto's repertoire is Did You Know, which generates automated insights for business users with a simple feedback loop built in so they can mark answers as useful or not useful. The vision for the technology "came directly from Josh," Ben Ainscough, Domo's head of AI and data science research, told me at the company's Domopalooza 2019 user conference, where Did You Know was .
Living in the machine learning matrix
We see these automated insight features in many BI tools today, but Ainscough claimed that there are some unique ideas beneath the surface in Domo's approach to machine learning. In particular, the company uses a machine learning method known as matrix factorization to discover latent features in an organization's data that can be used to make analytics recommendations to users on their phones.
Online bookstores and other retail sites offer a well-known example of recommendation engines. Users buy and, in many cases, rate items. From this information, we can create a matrix of all the users on a site and their ratings. To generate recommendations for additional purchases, we need to be able to predict what rating individual users may give to an item that's related to one they've bought. If the predicted rating is high enough, we could make a promotional offer to entice them to buy that item.
Matrix factorization is based on the idea that ratings are influenced by some latent, or hidden, features: the author, the genre and so on in the case of books. Can we discover those features? One way to do this is to factorize the matrix by identifying at least two matrices that together can create the original matrix.
In a sense, it's just like factorizing a number. For example, a factorization of 16 results in the multiplication of 2 and 8, 4 and 4, and 1 and 16. To greatly simplify, we may find that we can factor a matrix of book ratings into author x genre or publisher x price, but we don't need to know what author, genre, publisher or price means -- we just need to do the mathematics.
Domo's Ben Ainscough and other company execs demo
the Did You Know tool.
For Ainscough, this is why matrix factorization is such an excellent tool in a business analytics system like Domo. "We have tons and tons of activity as a latent measure of what people are interested in, even though we don't directly know the semantics of their activity," he said. In other words, Did You Know doesn't have to understand what you do in your business in order to make useful recommendations based on your activity in the Domo platform.
It's worth noting that, in practice, Domo uses a hybrid approach in Did You Know, combining matrix factorization with other recommendation techniques that do understand more of the semantics of a user's activity. This means that the Domo machine learning technology doesn't have to start from scratch for every new user, but can instead serve up useful recommendations from the attempt.
Are you skeptical? Good, Domo says
The Did You Know feature includes four types of data insights: trends over time, correlations, outliers and market basket analysis. Each has different statistics that explain the insight. A significant challenge, according to Ainscough, is helping the users to understand what they're seeing without requiring technical knowledge of, say, r-values, p-values or other statistical ideas.
"This is a hypothesis generation tool," Ainscough said. "If I were to lose sleep over anything with the feature, it is that I know I can serve an insight saying, 'Here's an r-value where the correlation is strong,' but then the executive concludes, 'Oh, this is driving my sales.'"
So, for Ainscough and his machine learning team, the challenge is this: "How can we convey these ideas with the appropriate degree of skepticism? I don't need to teach executives what a p-value is, but I do need to teach them that this is how we generate a hypothesis, and you need to go and explore it."
The Did You Know feature has already proven to be successful with some early adopters, who said they like the daily prompts alerting them that there's something worth exploring in their data. The increased user engagement that the Domo machine learning tool enables, especially at the executive level, promises to make businesses genuinely more data-driven.
The Domo software can't provide all the answers, of course. And it certainly helps if users can meet the system with a degree of understanding. That said, Ainscough's team is working to bridge the gap between usability and expertise.
"I am of the opinion that business users should learn statistics," Ainscough said. "But how do we, as a well-designed product, hold their hand and take them on that journey with us instead of just throwing them under the bus?" For Domo, the answer to that question is Did You Know.
