Working with SAP's Leonardo division, German chemicals giant BASF has gone live with a machine learning-powered bot for categorising incoming customer request emails.
Leonardo has had an interesting journey at SAP, starting out as an IoT focused product but being repositioned at Sapphire last year as a "digital innovation system" to help customers adopt machine learning, IoT and blockchain technologies to solve specific business problems.
As Mike Flannagan, senior vice president of analytics at SAP put it: "It's about taking a business problem, which is specific to one customer, find the elements of that problem which are common across the industry, then defining the elements of a methodology, the technologies used to solve that problem and package those up and make them available as an accelerator for the next customers".
There are a couple of routes to adopting Leonardo services with SAP. For quick implementations, SAP released 23 Leonardo-based innovation kits for specific industries during Sapphire in 2018, for use cases across retail, life sciences, manufacturing and automotive.
BASF, as a long-term customer of SAP, was able to go down the more bespoke route however, bringing people on-site for an extensive workshopping and prototyping process.
Speaking during a press roundtable at SAP's Sapphire conference in Orlando this week, Pedro Miguel Ahlers, digital growth manager at BASF's performance chemicals division explained: "We went with a problem and worked through design thinking to understanding what the problem we have is, talking openly, and then designing a potential solution."
The problem BASF wanted to solve was the inefficient way incoming customer request emails were routed, meaning employees were wasting time forwarding misrouted emails to the right people in the organisation.
So BASF spent five months prototyping a machine learning-powered solution called CuRT (customer request tracking) before spending ten weeks on implementation, going live on March 1.
Now CuRT is routing 2,200 emails a week - that could be a pricing request or a complaint, for example - for BASF's performance chemicals division, and is expanding rapidly into new geographies at the company.
"Now instead of getting twenty emails, the right person is just getting one ticket," as Ahlers puts it.
CuRT works by using cognitive machine learning, not a tree model, to learn from its mistake. Ahlers says it works to a 90 percent accuracy rate today.
"It's working on SAP Hybris for service ticketing and machine learning with Leonardo, so we connect channels like email, social networks, form sheets, web apps, whatever," he said.
As he wrote in a blog post on LinkedIn in February, the aim of CuRT "is to understand the type, the industry and the application of an inquiry. Finally, you will also understand who is the right colleague to route the inquiry. You'll get feedback, whether your proposal was right or not.
"With your help we will be able to focus better on the relevant things and to measure. Of course, you will learn to select the right templates to make sure inquiries are answered as fast as possible.”