Are you ready to bring more awareness to your brand? Consider becoming a sponsor for The AI Impact Tour. Learn more about the opportunities here.
DotData today announced version 2.0 of its artificial intelligence and machine learning platform for enterprises. The company automates data science so it can accelerate the adoption of AI and machine learning in corporations.
DotData CEO Ryohei Fujimaki said in a fireside chat with me at our Transform 2020 event that enterprises can implement AI and ML tools that generate better business insights and money-saving results.
“Everyone is under high pressure to deliver more results with less resources to survive in this economic downturn,” Fujimaki said. “AI automation will change this game. It significantly accelerates the turnaround from months to days.”
DotData spun out of Japan’s NEC in 2018, after more than 10 years of research into AI. Fujimaki is a data scientist, but clearly there aren’t enough people like him in the world. And that’s why DotData exists. Its role is to bring disruptive scale and speed in AI development through automation, and that automation enables people who are not data scientists to benefit from AI insights. Version 2.0 takes this a step further.
The AI Impact Tour
Connect with the enterprise AI community at VentureBeat’s AI Impact Tour coming to a city near you!
“This one is specifically designed for realizing what I explained today for all enterprises,” Fujimaki said. “AI automation is now ready for everyone to adopt. When we started the development of DotData 1.0, we had to consider that this is the tool for data scientists. However, during this past year or two, we have have seen this big change in the market. So 2.0 is really designed for enterprises who need the speed and scale, even without data scientists.”
As an example, Fujimaki pointed to client P&C Insurance, which generates $50 billion in annual revenue and has more than 30,000 insurance agencies that resell auto insurance. DotData needed to create an AI-based intelligent policy recommendation system for those agencies so that agents could propose the right product for the right customer.
The system analyzes customer profiles, past payments, claims, and more than 50 different types of behavioral logs. Then it produces three things. First, it creates a personalized product recommendation. Second, it considers what product is best for a particular customer. And third, it recommends a personalized product video. The human agent then shows the customer the personalized video and explains why the product helps the customer. The system debuted in February, and it resulted in a 250% improvement in conversion rates, or adoption of new policies.
Fujimaki said that his company’s platform created hundreds of AI models, starting in early 2018. The company had nine months to build those models, and then it started a field trial. The results were huge, he said.
A shortage of data scientists
During COVID-19, it isn’t easy to hire data scientists, who were in short supply to begin with. Walmart sponsored a competition to improve sales forecasts for 3,000 products for 10 of its flagship stores in three states. The data consisted of more than 42,000 time series, extremely multi-dimensional time-series forecasting — a very difficult task even for experienced data scientists.
More than 5,500 teams participated in this competition and worked on tuning their AI models for four months. The competition finished last month. DotData inputted the data for its models for each of the 10 stores. The AI automation took three to four hours per store, and the total time took about 43 hours, with most of that being computation time.
“The result was very exciting,” Fujimaki said. “Our AI automation was ranked at top 2.5% among more than 5,500 teams.”
The company found that, with AI automation, even business intelligence engineers or business analysts — who are not trained as data scientists — can build as good AI models as world-top data scientists, he said.
How it works
AI development is not just tuning a machine learning model but a very lengthy and complex process. It isn’t easy to get a complete set of enterprise data in the right form from different sources. The company has to cleanse, architect, profile, aggregate, and do data manipulation based on its domain knowledge to discover useful patterns and prepare data for AI processing.
DotData’s key innovation is inventing AI that automatically explores raw data and discovers hundreds of promising business insights without domain knowledge, he said. Finding business insights from raw data was possible only based on intuition or experience before, but AI automates the process on DotData, Fujimaki said. That’s why it can do in days what otherwise might take months, he said.
Fujimaki said the company is careful to make its model transparent and explainable. Otherwise, it would be hard to trust the results.
“It is our fundamental philosophy that enterprise AI must be a white box solution,” Fujimaki said. “What business insights were discovered? How will AI make predictions? What are their performance and business impacts?” DotData explains these things, he said.
“Like our insurance customer, you can easily produce hundreds of AI models to generate disruptive business outcomes,” Fujimaki said. “Also, like the Walmart competition case, BI and analytics teams can do as good as world-top data scientists and make their dashboard and reporting more predictive and prescriptive.”
Other methods require a lot of upfront effort and don’t provide the same outcomes, he said. “AI automation makes it more integrated and more agile. You run AI automation and get the first model within a day,” Fujimaki said. “You can start to use it as a minimum valuable AI model, and you continuously improve it. AI automation fundamentally changes a way of delivering AI projects, and the organization must become familiar with the new practices.”
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.