Martha Matthias Blog: Rajan Krishnan of Oracle: AI Should Take Us on a Journey from Hindsight, to Insight, to Ultimately Foresight

Earlier this spring at CRM Magazine’s annual CRM Evolution event, I had the absolute pleasure to moderate an all star panel on how artificial intelligence (AI) is and will impact customer engagement. The panel consisted of executives at some of the leading vendors in the CRM industry, including:

Marco Casalaina, vice president of products at Salesforce.com’s Einstein division
Volker Hildebrand, Global Vice President at SAP Hybris
Rajan Krishnan, Group Vice President of Applications Product Development at Oracle
Michael Wu, Chief Scientist at Lithium Technologies
Kishan Chetan, principal PM manager of Microsoft CRM Dynamics 365

Below is an edited transcript of just a portion of this great panel discussion that really dug into some meaty topics. To hear the full session click on the embedded player below. Unfortunately due to a low volume on his microphone I wasn’t able to get good transcript of Kishan Chetan’s valuable contributions to the panel, but I’d advise you to crank up the recording and check them out on the recording.

* * * * *

Small Business Trends: Can you share your general definition of AI?

Rajen Krishnan: AI has been in hibernation for about 20 years. Now there’s a coming together of cloud access to data, large-volume datasets, unimaginable global compute capabilities. Our infrastructure is distributed across 21 data centers around the world … bringing this all together; and it opens a realm of possibilities that did not exist.

Yes, there was one-to-one marketing, but this truly takes it to the Nth degree where it’s possible to deliver on the promise. But at the same time we need to be careful not to get ahead of ourselves, dial back a little bit, look at what options truly exist from where we start, and where we can potentially go. We could go in a hundred different directions. This is all exciting stuff. Great time to be in AI across the board, and not just within CRM, I’m speaking both from a CRM/ CX standpoint and broader enterprise standpoint.

Marco Casalaina: We have just reached an inflection point. We have a critical mass of data and metadata in the cloud. At the same time we have a critical mass compute capability so we can apply this massive compute to the data we already have and we can make AI work much better than it ever has before. That’s what we’re doing with Salesforce Einstein. it’s AI for business

Small Business Trends: Compare and contrast Machine Learning vs. Deep Learning

Michael Wu: Machine learning is just the process of using data to come up with the model. Traditionally when you’re trying to make any predictions you create a model right. Additionally these models are created based on our expert knowledge, and we believe these things should be weighed this way or that way, so you create a model out of your expertise. But machine learning is simply to take more of an inductive approach. I’m not going to assume anything. I’m going to plug in a lot of data and then let the data constrain my model. So you’re using the data to help you define or create the model.

Deep learning is basically a class of neural network. Neural networks have layers of neurons and it’s trying to mimic what our brain is doing. Our brain has hierarchical processing at the lower layer of our brain basically processing very simple features. And then as they go up on top it processes more complex features. So traditional neural networks can only process maybe two or three layers with the computing capability we had and the data that we had.

But now we can compute on the order of tens to hundreds of layers. So that’s why they’re called deep networks because there are many, many layers. There are so many layers deep. Deep learning is essentially deep neural networks. That’s what it stands for.

Small Business Trends: What goals are being set for AI in the context of customer engagement?

Volker Hildebrand: There is a lot of focus on automating things. I think that’s only half the picture. Yes, AI can help fully automate a very repetitive task, whether it’s simple things like checking accuracy of travel and expense report, or accuracy of invoices, or providing answers to comparatively simple questions. To a certain degree you can automate it and also scale it.

Yes, I want to automate, I want to reduce costs. Yes, let’s face it, you can replace human beings with machines and have human beings do other stuff. I think what’s important in that context is the business outcome.

When it comes to things like customer service I think if you take the approach of, yes, let’s automate things and try to cut costs and deflect incoming customer service calls to the virtual agent; but that’s actually the wrong approach. You’ve got to think about how you can improve the customer experience, that should be the objective.

Is this going to be more convenient for my customer? Is it going to provide a faster answer? Is it going to provide a more reliable, accurate answer? Focusing too much on automation and efficiency is the wrong thing. In fact I believe the greatest potential of machine learning or artificial intelligence may not be in the automation but actually in teaching humans entirely new ways of thinking. I think that’s the real potential that goes way beyond smart automation.

Small Business Trends: What are some misconceptions of AI?

Rajen Krishnan: We are seeing customers really all across the spectrum when it comes to AI. All the way from customers who think that AI means something like a robot running operations for them to customers that have some experience and probably some more sophistication as it relates to the data.

We have five billion profiles, about three trillion dollars of transaction data, about 750 trillion data points added on a monthly basis and about 110 million U.S. households are picked up by that data.

So customers that are used to data intensive processes, whether it’s marketing, targeting and so on, moving to the next phase with data, data signs, machine learning, decision signs to further automate and predict those offers or actions is one set of customers.

The other set of customers who probably don’t have as much experience with data, and by extension data signs, the expectations with AI are all over the place. But fair to say that, as we move along this journey, we’ve come from a framework of hindsight with just the OLTP (online transactional processing) systems to a bit of an insight driven model with analytics, and where AI can go is a foresight model. So the journey across the board will be say, moving from hindsight, to insight to ultimately foresight.

Marco Casalaina: One of the main reasons that is today is that chatbots will just work out of the box. Training is the new coding. So nowadays like to get these AIs to work we need them to have the data. And to get a chatbot to work, you basically need to have successful conversations talking about the same stuff that your agents would be talking about with your customers. And those are hard to come by. A lot of times you’re not recorded. They’re not actually text format. And also who is labeling them as being successful. How do you know this agent didn’t give a totally wrong answer. So labeling and training data sets is now one of the most difficult parts of doing AI; being able to train your AI to do what you want it to do.

Volker Hildebrand: Well I think it really depends. There are a number of customers who feel like hey, we have no clue what it is but everybody’s doing it so we need to do it too. There’s definitely that kind of category. There are others with very very high expectations, and I think this is really where when we’re talking to our customers, we have conversations about how much can you actually; do you actually want to automate a certain process or other things. Because there’s definitely a lot of use cases, and it’s not necessarily the case that whatever AI comes up with as a result is 100% accurate. And so there’s a certain confidence level that the predicted outcome, if you will, is the right outcome.

If this is in the range of 70 or 80%, that might sound good. Or even if it’s 90% that might sound great. But then you need to look at the use case. If you use it, for example, to score leads or opportunities, 90% accuracy of which opportunities you should be focusing on as sales rep, it’s a pretty good outcome and you’re probably doing a better job than you did before.

Now if you have a use case where it impacts directly the customer experience, like in a service scenario, if the answer of an artificial intelligence virtual agent, as an example, if that’s a 90% accuracy it’s probably pretty bad because it means that one out of ten customers will be pretty mad because they get the wrong answer all the time.

So I think that’s really an important thing to understand. What’s the use case? What am I trying to achieve?

Small Business Trends: What will be the impact of AI on CRM?

Volker Hildebrand: I’m also glad we kind of moved a little bit away from focusing on just the service and Chatbot use case, because that’s kind of the sexy use case that everybody likes to talk about. But they’re potentially so much bigger ones, and you’ve already heard some of those. Predictive maintenance, by the way, is really already huge. So it’s across sales, service, marketing, commerce.

Part is automation, part is prediction, part is just better decision making. Making recommendations … identifying untapped opportunities.

Another increasingly important data source will be the whole IoT area. That’s also tied to predictive maintenance so I think there are opportunities in that space, because there’s going to be a ton of data and all of a sudden, you will not only have customer attributes or behavioral clicking on the web, or what were they hovering over? What were they doing there? And combining it with transactional data, and all of a sudden, you have information about how your customers are using their products, which makes a huge difference and will have a significant impact on the quality of the learning and the outcomes.

Marco Casalaina: I keep saying that a critical mass of cloud is not just about CRM itself. It’s also about the fact that companies are moving to Office 365 to G-mail and so more and more of their data is going into cloud and cloud accessible spots. And that allows us to automate a lot of the data entry that would otherwise have to have been done manually. So you’ll see more of contact auto creation from your email; or detection of buying signs and things like that from e-mail which is something that we do called opportunity insights. You’ll see more information filled in; you’ll see more predictive forecasting stuff like that. So a lot of these things that have traditionally been very manually driven by a whole bunch of edit boxes on the screen, are suddenly going to start being filled in automatically from other data sources that you also have in the cloud that you just connected together. And that’s one way I think that AI is going to change CRM subtlely. But quickly.