Sergey Nivens - Fotolia
Looking for something else?
Advanced, next-generation technologies, like AI and machine learning, are changing the way we work. Technology like chatbot AI, background analytics, image recognition and more are affecting virtually every business process. In this Q&A, Timo Elliott, SAP vice president and global innovation evangelist, discusses three ways that these technologies will be embedded into business processes.
What are some ways that next-generation technologies, like AI, will affect business processes?
Timo Elliott: The big picture of what we're clearly seeing is that, over the next few years, every single business process will become self-optimizing. At SAP we have over 45 years of experience with business processes; some of those are already self-optimizing today. For example, we've been using predictive analytics in things like the supply chain to try and optimize what goods are stored where. Some predictive maintenance applications have been around for a long time, but now it's clear that [these technologies] are going to be applied to every single business process. It's going to collect data, and rather than people looking at it to analyze and try to improve it, the algorithms themselves will be able to do that automatically.
In practical terms, where are some areas where this will come into play next year or in the near future?
Elliott: We see three big areas of opportunity. First, AI is making computer interfaces more humanlike with chatbot AI and so on. Second, it's automating knowledge work with complex repetitive decisions being replaced by algorithms. Third, there's what we call 'doing the previously impossible,' such as using technologies like image recognition inside business processes to do things that just weren't possible before.
Are there some specific examples of technologies like chatbot AI that you can talk about?
Timo ElliottGlobal Innovation Evangelist, SAP
Elliott: SAP CoPilot is an enterprise digital assistant that's like Siri and Cortana but for the enterprise. So, anything that you do inside your businesses -- daily business processes, like 'I'd like to book a vacation next week' -- rather than going into a GUI and trying to figure out different fields, you can just say, 'I'd like to book a vacation next week,' and a chatbot will come back with 'Did you mean these dates? We can confirm if you're good to go.' That can apply to any business transaction; it could be 'Tell me about my sales this week' or 'Can I have more detail about what this order is?' It's surfacing all of that complex information that's in the organization through a human-chatbot AI interface. The difference between this and Siri or Alexa is that this is in a work context, and it has a lot more context about what's going on. If I ask Alexa one question and then another question, she's completely lost context between the two, but with the enterprise chatbot, you can actually keep on the conversation, and it knows who I am, where I am and what I'm doing, and it knows roughly what business process I'm trying to achieve. It can use that to make the process actually a lot more revealing than the average Siri conversation.
How does it do that?
Elliott: It knows who I am, and the chatbot AI runs alongside the existing experience. Let's say my job is entering new orders. If I ask a question, it automatically knows the context of the order and where I am in the process and can start doing it. Or if I have a question that I want to send to a colleague, I can just click a button to send a screenshot of what I'm looking at to another person, and the system knows the context that I took that screenshot in. Instead of just giving them the screenshot, it also knows that, if you're trying to do an action, it automatically takes them into that business process in a way that's seamless behind the scenes.
How is AI and machine learning automating knowledge work, and why is this important?
Elliott: Automating knowledge work is the lowest-hanging fruit for AI that every organization can start doing now. For example, [SAP] worked with a large chemicals company to use machine learning as part of their financial operations. When they first installed SAP a few years ago, only 40% of their invoices matched their bank payments, so if they get an invoice in from a customer and they've paid that invoice, they're trying to match that in the background to make sure that they're keeping track of their cash. In the real world, that's a fairly messy process because the reference numbers don't match or people send two invoices for one bank payment or there are two bank payments for one invoice. It requires lots of manual time and effort to fix all of those different exceptions. This company started out with 40% matching, and they managed to get up to 70% after they put in a bunch of rules, but using an algorithm-based approach, they managed to get it up to 96%. That's just one example of invoice matching, but we believe that there are hundreds or maybe thousands of those complex repetitive decisions inside organizations that can be automated using these technologies.
What are some other processes that can be automated?
Elliott: Predictive maintenance is one. People can gather a lot of information about their machines with lots of sensors built into machines and then use predictive algorithms to tell when they're going to break down so that they can do maintenance ahead of time. There are also things like scanning invoices, where rule-based systems extract things like the date, the amount, the vendor and what it's for. There are limits to how well that works, but machine learning is a big leap in how efficiently you can do it, and you can apply it to things like procurement.
SAP also has a product called SAP Customer Retention. Imagine a bank that has thousands of customer touchpoints -- like you go to your bank online to check your account or conduct a transaction -- all of those touchpoints are gathered and used to create what we call the 'customer pulse.' The idea is that it's a little bit like your pulse and can give you an indication of your health. If there's a problem with your pulse, then something bad is going to happen, so it's kind of the same thing with customer turnover. Working with some banking customers, we were able to detect the patterns of activity that show if the customer was thinking of moving to another bank long before they actually left. So, then, the algorithm takes the next step, which is to tell you when you've discovered somebody that might be leaving, and you can figure out how to try and keep them. This is something that people do today just in a very manual way, but the algorithms do it for us.
The first two areas you've talked about are how AI helps with tasks that we are doing already. What's the third area where AI can change business processes?
Elliott: The third area is the previously impossible, where you can do things like image recognition. For example, SAP has a product called SAP Brand Impact that's designed for media sponsors of things like sports events. It takes videos of the event and, in near-real time, analyzes every appearance of a sponsor's logo on the screen -- where it appeared on the screen, what size, exactly how long it was on the screen -- so people can get a very detailed view of whether they're getting their money's worth for sponsoring that event. Previously, this would have literally been students -- who would be paid -- with stopwatches trying desperately to figure out how long the logo was on screen. This was very inexact, but we've completely automated that using some very powerful machine learning technology.
Are all these technologies and changes in business processes happening now?
Elliott: This is what we can do in the next year, at least. For the longer term, we have the opportunity to rethink a lot of typical business processes from scratch if we assume that more powerful AI or machine learning is available, and it increasingly is. We can actually cut out a lot of the steps of modern business and instead have more of a proactive approach where the system starts telling us what we should be looking at rather than us having to constantly track what might be going wrong.