You’re familiar with self-driving cars. Well, SAP engineers self-driving enterprises.
How is the company doing that? How do its solutions become intelligent through machine learning? And which use cases are suitable?
“We’ll never run out of ideas for compelling machine learning use cases,” says Daniel Dahlmeier, the Singapore-based machine learning expert.
At the SAP Innovation Center Network location in Singapore, Dahlmeier leads a team that works on machine learning solutions for sales and service. One of these is SAP Service Ticket Intelligence, which automatically categorizes customer service tickets, routes them to the right agent, and provides the agent with recommendations for solving the issue at hand. The more tickets the solution processes and the more user feedback it receives, the more efficient it becomes. In short, it learns as it goes along.
“It’s in processing this unstructured data in customer tickets that the strengths of machine learning really come into their own,” explains Dahlmeier. “Thanks to built-in machine learning algorithms, the model understands the semantics of the tickets, is able to recognize similarities, and improves over time. That’s just not possible with conventional programming.”
SAP Service Ticket Intelligence is offered as a discrete business or web service and as part of SAP Hybris Cloud for Service. And it’s a prime example of SAP’s strategy of making all of its solutions smart ‒ a strategy that describes the machine learning team’s vision.
All of its solutions? “All those that are suitable candidates,” says Dahlmeier. SAP usually automates business processes. Its developers do that by programming (simple) rules to create a basic payroll or MRP process, for example. But where a problem is more complex because, say, the data involved is both highly unstructured and extensive – as in the service ticket intelligence solution – “that’s the sweet spot for machine learning,” explains Dahlmeier.
For his team, these two criteria – large data volumes and complex rules – are central to assessing whether a machine learning solution is suitable for solving a business problem. One of the key requirements for self-learning algorithms is the availability of large datasets. These are essential for creating a model with the ability to learn and to propose effective solutions or predict future developments with a high degree of accuracy. And complexity enters the mix when the data to be analyzed is in a foreign language or includes images or handwriting.
Diverse Use Cases
Machine learning, Dahlmeier says, has applications in many areas where SAP is active. “If you look at the SAP Digital Transformation Framework, you can see that the use cases for machine learning extend right across it,” says Dahlmeier. Which is why his team is currently focusing on various cross-topics that will appeal to a large number of customers. SAP Cash Application is a good example.
Today, most companies work with automated accounts receivable processing. But accounting personnel still need to intervene manually if customers omit key payment data, mistype a number, or make an error when paying multiple invoices in a single transaction.
“On top of that, country-specific differences in electronic banking create an added level of complexity for companies that operate globally,” explains Sebastian Schroetel, who is responsible for the machine learning solutions for SAP’s digital core. “This is where the power of machine learning comes in, because algorithms can learn about these differences automatically and, as they learn, gradually render manual country-specific adjustments obsolete.”
SAP Cash Application integrates with SAP S/4HANA and uses machine learning to speed up the tedious process of payment matching. It either ties a payment to the correct invoice automatically based on historical data. Or it tells the responsible employee which payment is the most likely match for the invoice in question. It then remembers which steps the employee took to assign the payment correctly.
In this way, it keeps on and on learning and, as the degree of automation increases, it saves the company money. To take an example, thanks to SAP Cash Application, Swiss energy company and SAP co-innovation customer Alpiq has succeeded in relieving employees in its shared service center of routine tasks and achieved an automation level of more than 92 percent. SAP Cash Application is just one component in end-to-end business processes that are gradually being automated through SAP Leonardo Machine Learning. As Schroetel says, “Our ultimate aim is to automate all business processes in S/4HANA and to create a “self-driving ERP system powered by machine learning”.
Another machine learning solution on SAP’s price list is SAP Brand Impact. It helps companies analyze the impact of their sponsorship investments. For example, it measures the frequency and duration of brand asset exposure during live transmissions of sporting events, music festivals, and so on. Algorithms in the solution take care of the previously tedious task of gathering relevant information by hand, identifying in near real time the size of the company logo, its position on the screen, and the length of time for which it is visible. By correlating these results with comparative data for other brands and with set key figures, companies can gauge how effective their sponsorship investments are.
From Retail to HR
Human resources is another area that lends itself to machine learning. The SAP Resume Matching solution, for example, helps recruiters speed up the process of matching suitable candidates to open posts. And developers are already working on an application that helps employees plan their careers by recommending which steps they should take to develop professionally and build the expertise they need to grow into their ideal role at the company as fast as possible.
SAP’s development teams are also focusing on the retail sector. With the help of machine learning, they are designing systems that can identify which colors will be fashionable in the summer, thus enabling companies to make tailored product offerings to their customers. Information like this is also helpful to fashion manufacturers, who can use it to predict which products are likely to sell well and to place their orders accordingly.
Machine learning could also help jewelry makers in certain scenarios, such as when customers submit images of damaged items like necklaces. The system would immediately show the customer service agent whether the same necklace is still in stock or suggest what repair work might cost. There are also possibilities in customer management, where software could help companies develop strategies to avoid losing dissatisfied customers (SAP Customer Retention), in the banking sector, and in the supply chain. The list is endless.
Which takes us neatly back to what Daniel Dahlmeier said at the start: “We’ll never run out of ideas.”