Conversational AI In B2B & Enterprise

Watch The Event Again

Event Q&A

Do you think the two factors, discoverability and invocation name, are holding back the enterprise adoption for voice and conversation? And what are your strategies to mitigate this?

Khartik: Are they holding back enterprises? Yes, absolutely. And in fact, a lot of enterprises have various types of internal projects where they try to do something and they face a lot of challenges in mining the historical data and then they give up. They give up because of the challenge of technology and to know how to mine historical data to create some sort of data corpus, and then essentially train your AI on the data, and then iterate.

Khartik: I think it's a bit of a technology gap, of a know-how gap, of metadata logic gap. In terms of mitigation, realistically, you need to focus on being able to mine data. And as you started mining, what you also do is you liberate a lot of your data that you've just been sitting on for years.

You mentioned automating the design process with a human in the loop. Can you unpack that means?

Khartik: There are several technologies out there that don't even have WYSIWYG (what you see is what you get) UI, so the ability to design a conversation. A lot of the conversational AI technologies out there, unfortunately, don't even have that. So you need to be a developer to stand up apart right now.

Khartik: For example, Microsoft Azure Bot previously lacked a UI, and you would go write code to design a conversational flow, which is incredibly cumbersome. They recently launched something called Microsoft Power VA, which gives an editor that now lets us chart you know, drag & drop boxes, you can train from within each box, and then connect the boxes to basically put a conversational flow together.

Khartik: You can take that a step further and allow training in line so you don't go off to a separate page, you're essentially able to train in line. As a designer, all the elements you really need for design is right there.

Khartik: Now, what makes that even better are recommendations. So if you're adding certain types of training data, an algorithm can actually predict the doubting that will skew the model or it will corrupt the training data, or it will cause the understanding of something else in the system.

Who are the typical buyers of conversational AI in large organizations

Khartik: Yes, I've talked of three personas and each with decreasing interest, but you know, all three are buying personas. So the first one is Chief Digital Officers, and digital teams are the number one buying personas. So typically, a lot of our interactions are with the Chief Digital Officer in any enterprise and the teams who are essentially making the purchasing decisions investing in the technology.

Khartik: A second one is customer experience. Some companies may not have a digital officer, but they will have a VP customer experience. So they automatically are second in line to want to buy such technologies because they see a lot of value that the automation brings them. Not from a cost-saving standpoint, but rather from a customer experience standpoint, it's something as simple as this: The bot, you have 24/7 customer service. If you're paying a call center, typically 9 to 6, 9 to 8 in some cases, but not all companies have the luxury of running 24/7 support. So a lot of them see the ability to have a board or something good for the customer experience because now customers can engage and ask questions anytime. Every time people ask questions. Even if you can't get the right answer, you leave a clue as to what they need next, and that informs product strategy.

Khartik: And the third persona is VP for call centers, because they see it as a cost saving play. And then they're able to put a business case together and say: "You know what, it's going to let me relieve some of the agents or healthy agents". What we found is agents are not adverse to seeing bots out there, they love it. Because what a lot of these chat bots do is they collect data up front.

What are some top use cases for sentiment analysis for enterprise clients? And what are some potential ethical issues with using this as kind of getting insights from customers?

Khartik: The top use case for the sentiment analysis I mentioned earlier is the emergence of a new customer experience metric called Voice of the Customer analysis or "VOC analysis". Voice of the Customer tells you in direct terms the topics customers were talking about. So there are topic mining algorithms that can do that. Topics are indicative of a certain pattern. And you know, those patterns are essentially indicative of what's happening. So you do Voice of the Customer analysis, get the top topics by the hour, or by the minute, or by whatever frequency you want. And then you get a gauge for the sentiment. So you very clearly know if something good is happening with the business, something bad is happening with the business, something terrible is happening with the business.

Khartik: Now second is the ethics right? And I would treat the two questions a little exclusively because ethics is a broader topic. So the ethics of using sentiment analysis goes well beyond just sentiment analysis to see ethics of doing any kind of data mining. And there are ways in which the industry considers safe to do data mining on any kind of data. The first one is making sure that all information is redacted. The second one is to ensure deidentified data anytime you do any type of mining. the third one is the way you essentially put this in a pool to analyze, there's something called differential privacy. Differential privacy is making sure that you as an individual never get identified. You're always part of some larger class. And there are attributes about you that I mind but I never mind you. So you're part of some cluster that has certain shared attributes and at no time can I take you out of that cluster and know who you are right? So I can't ever do a reverse mapping.

Slides

Karthik's slides

cai-b2b%26enterprise-wysdom-ai