We know how imperative it is to start thinking and operating with an AI-first mindset in customer service. But don’t just take our word for it. We brought together some of the most innovative and ambitious C-Suite executives and CX leaders to talk about the shift from human-first to AI-first customer service.
The webinar, led by Ada’s CEO, Mike Murchison examines what’s next in the customer service landscape and the evolving horizons of AI in customer service, specifically:
Joined by Jeff Epstein, Ex-CFO of Oracle, Operating Partner at Bessemer Venture Partners and board member of Twillio and Okta, Beka Swegman, Senior Vice President of Customer Success at Bark, and Casey Woo, Founder of Operators Guild, Ex-CFO of Landing, the panel discussed the impact AI agents have on reducing costs and boosting revenue and unpacks the roadmap to integrating AI into your customer service.
Missed it? Check out the recording or the recap below.
Mike Murchison: It’s been about a year and a half since ChatGPT became a household name. I'm curious to hear your take on how the landscape of AI expectations and its real-world use has shifted since then.
Let’s start from the management perspective and then we’ll look at it from the board level. Casey, can you start by sharing how the companies you work with are putting AI to use most effectively?
Casey Woo: I think there's an expectation and active effort by every company to see how AI can benefit the company, whether it's the product side or the productivity and cost side.
I’ll represent the operators. There's a very active frenzy of, “What tools are you using? What are you doing? What’s working? What's not?” And of course there's a very big risk equation. As well as a sometimes incremental cost equation that you have to present to the board of “We want to try this. Here are the pros and cons.” But yes, I think everyone's trying to figure it out and try to see how they can benefit the company.
Jeff Epstein: I went back and looked it up. Artificial intelligence was first coined as a term in 1956. So it's actually very old. What happened — and I think we all know — is in November 2022, ChatGPT launched the chatbot. It's been around for a long time, but it just opened up everybody's eyes and it created all these new, very easy to use cases.
I think of 3 categories of use for different kinds of companies: The first is new AI companies that have been created since November of 2022, where they are AI large language model companies from the beginning. And an example of that is a company that I work with called Luna.ai. It replaces SDRs. It's basically what an SDR does in terms of writing emails and reaching out to people and then responding to emails. It could be done all automatically. It's like magic.
Then there's companies like Ada (https://www.ada.cx/), which have been around since 2016. And you were AI from the beginning, but not a large language model AI. The launch of large language models just supercharged your product and gave you incredible capabilities to do things you couldn't do before. So you've got a category of AI companies that are now just much better at what they do, like Ada.
Then there's companies that didn't have AI at all, like Adobe. They were just a software company, and now Adobe’s adding AI features. Essentially every software company is going to be adding AI components. Twilio and Okta are companies that I work adding them as well.
So I think those are 3 categories: the brand new AI companies, the AI companies that have been in business for a long time and have been great, but are now supercharged, and then the third category of legacy software companies that are going to add AI features.
Mike: Jeff, from a board level perspective, how have your expectations of the companies you sit on the board of changed as a result of this next wave of AI?
Jeff: Before November 22, it was all science fiction. We read about it. We read about artificial intelligence, but no companies that I was working with actually were using it practically, internally. Of course, you at Ada were one of the pioneers, so you were using the prelim large language model versions of AI. Within the engineering department, I would say people were using it as tools. But it didn't really reach the board level. It didn't reach the product level or commercial level.
I remember when Netscape went public at the beginning of the internet. The Netscape IPO was the beginning of the race for the internet boom. And I think the ChatGPT chatbot was the beginning of the race for AI. So it's now on every board agenda, in every board meeting. We're asking questions about: Are we falling behind? How do we catch up? What are we doing internally? How much are we investing? It’s one of the top topics at every board meeting.
Mike: Beka, I know you've got experience speaking to your board and engaging with them about said topic. Maybe you could walk us through how you've been applying AI into your organization.
Beka Swegman: Sure! The use cases are pretty endless. We've used it anywhere from support to our training materials and how we're training. You can set up AI prompts so that it can battle test your agents as you're training them. So there's all kinds of things that you can be doing.
I really tried to get my entire team comfortable with the thought that they should be using it in some capacity in their roles, all the way from tier one agents up through our operators, our operations team and really infusing it into everything they do.
At the board level, it's talking about efficiency. So how can we continue to scale companies to remain efficient and cost-effective? When we're talking about it with the board — or for me, most often with our CEO — it's always about what elements of the job or the jobs could be done by AI so that we can better utilize the people that we have in other capacities in our organization. That's essentially how we've been looking at it comprehensively.
Mike: Now, given that the interest from the board seems to be more efficiency-oriented or cost-oriented, have you experienced a friction or a tension between the efficiency impact of your AI investment and your experience goals as a customer experience leader?
Beka: Absolutely. I think there's a huge gap between standing this up, how you implement it, and what your customer experience is going to be. Of course, we could talk about the normal KPIs: CSAT, customer effort scores. There's always gonna be friction when you're doing something that's automated because there’s a lot that goes into making sure that it’s done well so that you're not diminishing the customer experience. There's a lot of safeguards that you have to put in place.
I think of it as training an agent. If they're brand new to your organization, you're not giving them the most difficult ticket to work on on day one. You really have to grow and scale into those capacities and those responsibilities. And so I approached AI in a similar fashion.
I hit all the low hanging fruit first — the things that are easy to automate, and then you can kind of grow from there. You also have to build some competency and understanding of what all goes into a large language model and making sure that it's answering efficiently and giving us a good customer experience.
Mike: Casey, does that relate to your point about managing risk? I think you were touching upon that. It sounds like Beka's got a crawl, walk, run, approach to rolling out her AI agent. Is that consistent with your perspective on risk management? Or is that different?
Casey: The short answer is, yes. I think there's 2 categories. There's critical functions and non-critical functions. Like any critical function, you test, walk, run, parallel process before you ever switch over. Customer support is definitely one of those — anything touching a customer.
And then of course, internally, there's FP&A tools being used where it's controlled in the FP&A team, and you can kind of fuss around with it. Which is more fun, because you can quickly test, etc. But yeah, those are the two different ones.
Mike: I think if it was only about efficiency you could just get rid of your website. Turn off your phone number, block all contact. So clearly, there's a balance to strike between the efficiency savings and the experience that is being driven.
Jeff, what do you think? I I think that this tension between efficiency and some of the more long term strategic impact of AI is interesting.
Jeff: At the board level, the risks that we look at are quality — another variation of that is accuracy — and security. So for quality and accuracy, an example is a company called Abridge, which is AI for doctors to take notes when they see a patient.
When I went to my doctor last year, I would talk to the doctor, they'd be talking to me, and while I'm there, he's sitting in front of his computer and taking notes and just talking about my condition and what medications he's prescribing. And then when I leave, he probably spends another 5 or 10 min writing things down. Over the course of the day, that's a couple of hours a day that doctors spend not really seeing patients, but instead interacting with a computer. And it's needed to record what you're doing and also to get paid because that's how they get reimbursed by insurance companies.
With this company, Abridge, what you do is the doctor just puts a phone down. It says, “Would you mind if I listen to the call?” Most people say yes, and then the doctor doesn't type anything. At the end of the meeting the AI says, “Here's a transcript of everything they said.” Here's all the conditions of the patient. Here's the bills that we have to send out to the insurance company,” and the doctor reads it, either says yes or no, and you know, might make a couple of changes. But it seems like it's saving an hour a day for a doctor.
Incredible AI use case. It has to be accurate. If it's not accurate, people are going to die. So the testing that they do is obviously very rigorous.
The other issue about AI is, it seems to have this strange characteristic that it could be accurate today, and then the same question tomorrow could be inaccurate.
You can't just set [AI] and forget it. You have to continually be testing for accuracy.
Mike: Beka, it seems like that AI management approach you're taking is core to what Jeff's talking about, which is ensuring you're getting the value over time. Could you share a little bit more about what you're doing?
Beka: I think the thing that I would say for any support leader who has not gone into the world of AI yet to reduce your support costs, the things that are most important are to not only have the organizational structure that supports it, but to hire the key roles to manage the AI. I think it's incredibly important.
You can call it whatever you want, but we call it a bot manager. Depending on the size of your company, you may need two or three, because your bot could be handling hundreds of thousands of interactions every month. You need to be able to QA that to some degree. Now, there are plenty of wonderful tools that basically will surface to you any major issues that are going on within the conversations. Of course Ada is one of those. It makes your QA more efficient.
I think that if you have not considered hiring or moving someone into this role whose entire job is to make sure that your AI experience is a good customer experience, and depending on what industry you're in is compliant, then that is definitely the first thing to consider.
I know support leaders are like, “They're not gonna let me hire.” The way that I talk to a board or our CEO about this is just a shifting of roles. In the same way you have someone who writes your support articles. You have someone who does your workforce management. This is just another operator role that manages, I would argue, probably one of the most important customer facing tools that you have. When you start to talk about it in those terms, and then you can layer in the cost savings from the amount of interactions the AI is able to handle, that becomes a very easy conversation.
Also, do the math if you haven't. I would add that. If you're a support leader, and you say, “Oh, our CFO does the math.” No. Do the math. It’s pretty easy math, calculating salaries and potential amount of tickets that could be handled by AI, and that definitely helps to have that conversation.
Mike: We've got an opportunity here to have this covered. We have customer experience leaders. We have CFOs right here, right now.
Casey, does that resonate with you, Beka’s approach? It clearly worked at Bark. What advice would you offer other customer experience leaders around how to package their ask around hiring someone to manage their AI.
Casey: I think it's like any other business proposal. What are the resources? What are the goals? What are the dependencies?
I agree with Beka. I think there’s people who are like, “I can’t forecast it.” I was like, let’s walk through it. To your point. So do you think it's 10% savings? I encourage people when they go to a finance officer or a business leader to try to quantify.
Roughly right, precisely wrong is okay. You’ve got to have some sort of basic understanding and not be afraid of perfection, because 100% of forecasts are wrong.
It's really the act of it, the alignment of it, the exercise of it, the thought process of it, the communication of it. That is the most important thing when you're talking about any type of proposal. But yeah, I agree with her process, and I think it's iterative.
I think the other really cool thing that's happening in communities is people are teaching each other what they've learned and actual success. So that's happening very rapidly because it's very similar problem sets for support teams, or, you know, billing teams, etc.
Mike: If we think about AI's impact over the next two years — given how quickly things are moving — let's go around the room and hear any bold predictions or core predictions that we think are going to change the way businesses are going to relate to AI.
I think Jeff would be valuable to hear from you from a board management perspective. AI is top of mind on a quarterly basis right now, is that still going to be the case two years from now or will it be different?
Jeff: It absolutely will. It'll be more important two years from now. The financial valuations and stock market hype may go down, and some of these companies that are valued billions of dollars may go out of business. It's very similar, I think, to the internet.
When it first became popular, there was an explosion of thousands of internet companies. Some of them turned into Google and Amazon, and some of them went out of business. But the internet is pretty pervasive today. It's in everyone's lives. This large language model multimodal — it's partly language, but multimodals are also images and videos — so anything having to do with language, images or videos or sound will affect everything that we do. It's going to be incredibly important for all of us.
Beka: I was actually thinking about this from the last time we talked. I don't know why, but it kept coming to my mind the image of when the car was invented. What I kept thinking about is, right before the car was invented, if you ask someone, “How fast can you go?” They're thinking, how fast can a horse go or a buggy with the horse.
The day that the car came onto the market, probably they were still thinking in terms of the horse and the buggy, and then within years it was a totally different conversation. When you say, “How fast can you go?” It's how fast a car can go, and you completely shift over into this new model.
I think the same is actually going to be true with AI.
Right now, we're thinking about efficiency with jobs and different roles. I don't even think we have scratched the surface of what we're going to be able to do with the advent of large language models.
Of course they've been around, but with the ability to take them to the general public in a way that is easy to use and is just going to grow, I don't know that any of us can predict how quickly this will go here.
I think that for customer experience leaders, it's going to lead to hyper-focused contextualized conversations with your customers that are incredibly detailed. It's a marketing dream. You're going to have plugins that know everything about how your customer has interacted with your product.
They're going to be able not only to upsell them, but they're going to be able to anticipate situations and problems that they're having with your product. They're going to be able to do a lot of the learning capacity, whether that's B2B or B2C. And they're really going to be an infrastructure within every product would be my prediction. So hyper, hyper-focused and contextualized customer experiences.
Mike: I’m hearing you say that it also brings you closer to revenue generation, or could?
Beka: It's going to touch customer success. It's going to touch pre-sales post sales. It's really going to be every touchpoint of the customer journey, probably all the way up the funnel. It'll be really interesting to see that web of connectivity that it creates for businesses as we move forward.
The three types of companies shifting to an AI-first approach:
How companies are doing this in practice:
The future of AI-first customer service
Delve into the evolving horizons of AI for customer service, and how you can transform your business with AI labor and management.
Get the recording
