Q: When did SAP begin to put the concept of a digital assistant for the enterprise into action?
A: Late in 2014, I was working with a small development team on an early version of an app for the German National Football Team (DFB). The idea was to facilitate communication among coaches, players, trainers, scouts, managers, travel coordinators, and anyone else associated with the team. The app can be used to send and share information such as text messages, video, and photos on a secure platform, as well as engage in private chat room discussions. While working with the DFB, we started thinking about how businesses might benefit from these concepts.
We then began a proof-of-concept project called “Engage” that looked at how people communicate with coworkers, for example, concerning a sales order. The first ideas that led to SAP CoPilot began there. Beyond being able to share information, we asked ourselves, “What if the system could also recognize related business objects from the context of that interaction?” You could then collect and share those “live” business objects, not just a static image, and engage in a whole new way by collaboratively solving issues in real time with colleagues and with the system. Thus live sharing and intelligent recognition of the business context was incorporated into SAP Fiori 2.0. This demonstrated the need for a digital assistant and its enormous potential.
The thought was that SAP CoPilot could be more than a productivity tool and become a real digital assistant, or “copilot,” if you will, that notifies you when something needs your attention and helps you complete your business tasks. We started with a very simple use case of speaking with a digital assistant to create a leave request. The development and management teams at SAP as well as partners were very excited by the potential, and we received the go-ahead for continued investment.
Q: What is the vision for the use of SAP CoPilot in the enterprise? Do you foresee users embracing this technology?
A: One of the major advances of SAP Fiori was to decompose a monolithic ERP solution into task-based, or activity-based, apps. Individualizing apps according to a user’s role is more in line with how people consume and use apps in their personal lives today — just think about how you use your smartphone. However, doing that resulted in thousands of apps; in just the latest release of SAP S/4HANA, for example, which is our core digital platform, there are about 7,000 apps. Then factor in all the applications coming from SAP Hybris, SAP SuccessFactors, Concur, and so forth. Of course, no one user has to use thousands of apps — only the ones he or she needs. Nonetheless, it can be a burden to find the right app for what you want to do.
Conversational UI is a natural progression that helps to unify the user experience across systems with different UIs. Users don’t need to know where to go to get the information they need or find the right app to perform the task they want to get done. A digital assistant for the enterprise simplifies that while at the same time supporting users with smart insights into an enormous treasure trove of data stored in the system. It’s not the first time we’re talking about this, but the technology that is now available makes it more achievable.
Q: What are some of the challenges in bringing a digital assistant to the enterprise?
A: One of the biggest challenges is the issue of trust: How do you design a conversational UI to establish trust between a human and a machine? We’re looking at many ways to do this, one of which is designing a personality for SAP CoPilot.
Another challenge is that any digital assistant must learn or increase its knowledge over time, which is where machine learning factors into the equation. In human-to-human conversation, each participant learns to adjust and adapt to the other’s conversational style. It is natural, therefore, for humans to learn the best way to communicate with a digital assistant. For the digital assistant to learn from humans over time, you have to teach the machine to understand human language — and that’s not easy to do. Just think how easy it is for people to misunderstand each other. It’s the challenge of natural language processing and meeting the expectations that people have. If a digital assistant can adapt to the way a person works, it should also be able to adapt to that person’s specific language patterns.
User research is extremely important in defining any user experience, but especially in this case because it’s so new. Digital assistants are just starting to enter the market, and there’s still a lot of development going on. The way humans and machines communicate opens up a wide area of research that is vital in understanding these challenges because there is very little precedent. To that end, part of our research has been “Wizard of Oz” experiments, where a test participant thinks he or she is communicating with a machine, but is in fact interacting with a person who is pretending to be a machine. This method helps the team to understand how people would expect and like a digital assistant to respond. For example, we are looking at the sentence structure of the participants: Do they use command-like language or are they more polite? At which points in the dialogue do they start to sound annoyed? Answers to these questions help us to design a better experience and build trust.
Q: It sounds like there is a lot of potential for a digital assistant for the enterprise, but there is still a lot to do. What did you decide to focus on first?
A: SAP CoPilot is being designed first and foremost to provide facts, and those facts are the data that is coming from SAP systems. It won’t be SAP CoPilot’s focus to tell you the weather in Palo Alto, for example. We know, of course, that customers trust SAP with their data and their processes, so it’s a natural extension that they would trust a digital assistant from SAP to be an interface to their business. A lot of customers currently involved in beta testing have said that they’d have the same level of confidence in a digital or virtual assistant from SAP as they do in SAP running their business processes.
Q: What use cases do you anticipate for SAP CoPilot?
A: There are many use cases. What we’re doing now is opening that up to all the different product areas in SAP, the application experts, to develop their own use cases. SAP S/4HANA is working on use cases for procurement, using a digital assistant to create a purchase order for a specific item, for example. Another use case being developed by a beta customer is to communicate with SAP CoPilot around days sales outstanding (DSO) metrics. As use cases emerge, they will be built by SAP, customers, partners, and the ecosystem to help make SAP CoPilot more intelligent.
When we talk about SAP CoPilot becoming more intelligent, checking daily DSO figures is a good example. Over time, if that’s a daily communication that you have with SAP CoPilot, it would learn that that information is something you ask for regularly and eventually provide the information proactively without being asked. Or, for example, if it knows that you are managing all the purchasing contracts and one of the contracts is about to expire, then SAP CoPilot could proactively remind you that the contract is about to expire. It could also provide details around all the sales orders that might not be fulfilled if the contract isn’t renewed. So you can see some of the things SAP CoPilot would be able to do based on your usage combined with the information in your SAP systems. We are also looking into incorporating data extracted from non-SAP sources, such as from social media.
Q: Will users be able to customize SAP CoPilot, not only according to a business’s unique processes, but also the look and feel?
A: Our intention is to allow customers to give SAP CoPilot the characteristics that best fit their company — whether that is giving it a name, a gender, a personality, or their branding. The vision was that this would always be done at the company level. And as SAP CoPilot learns from you, your roles, and what you have access to in SAP systems, that learning and context will in effect be another layer of personalization. This is because SAP CoPilot will become more and more suited to how you go about your work day. It adapts to you as an individual and the way you work.
Q: What are the common misconceptions people have about what it means to have a digital assistant for the enterprise?
A: One of the biggest misconceptions we see right now is that people think the development work is done once the automated speech recognition is settled. The reality is, however, that automated speech recognition is just the tip of the iceberg. Designing conversation itself and incorporating nuances in language is a huge task, as is overcoming the perception that if a machine can talk it must be intelligent. For users, SAP CoPilot is a work in progress, meaning that out of the box it isn’t going to be as intelligent as it will be after using it for a month or a year, as it learns from your work patterns.
Along the same lines, there’s a misconception that once SAP CoPilot is rolled out in English, it’s a simple matter of translation to roll it out in German and other languages. This again speaks to the nuances unique to each language that define what it means to have a natural conversation; it’s not a one-to-one translation as it would be for translating text on a screen, for example. That’s one of the biggest challenges. Because SAP supports many different languages, the expectation is for SAP CoPilot to handle multiple languages as well. The goal is to provide multiple languages for SAP CoPilot, but it’s not as easy as just translating it from English.
Finally, while SAP CoPilot is at its core a conversational UI, it’s not purely conversational. Experience tells us that sometimes it’s faster to deliver information with images, graphs, and other UI elements, and to this point SAP CoPilot is not purely natural language but rather a multi-modal interaction that will support graphical user interface elements, gestures, and so forth. We believe this will ease adoption for users as they move into interactions driven by artificial intelligence. Business is moving in that direction, and that context helps crystallize the vision we have for SAP CoPilot in the enterprise as the personification of machine learning across SAP products.