As founders plan for an increasingly AI-centric future, Gusto co-founder and head of technology Edward Kim is cutting back on his existing team and hiring a slew of specially trained AI engineers. He said it was the “wrong way” to do so.
Instead, non-technical team members “actually have much more insight into what situations customers might be in and what they're confused about than the average engineer.” He claimed that this gives them a “deeper understanding of the system” and puts them in a better position to guide functionality. It should be built into AI tools.
In an interview with TechCrunch, Kim, whose payroll startup generated more than $500 million in annual revenue in the fiscal year ending in April 2023, outlined Gusto's approach to AI and explained how the customer experience team's A technical member explained that he was writing a “recipe” like the one below. Gus, an AI assistant (launched last month), guides you on how to interact with your customers.
Kim also said the company recognizes that “someone who is not a software engineer but has a little bit of technical awareness can build very powerful and innovative AI applications.” For example, CoPilot is a customer experience tool deployed in the United States. The Gusto CX team joined in June and is already seeing 2,000 to 3,000 interactions per day.
“Gusto is actually able to upskill many of our employees and help them build AI applications,” Kim said.
This interview has been edited for length and clarity.
Is Gus the first big AI product you've released to customers?
Gus is a big AI capability that we've delivered for our customers, and it ties together a lot of the point capabilities that we've built in a lot of different ways. That's because the app is littered with AI buttons that say, “Press this button to do something with AI.” In our case, it was, “Press this button and we'll create a job description for you.”
But with gas you can get rid of all of that. Just when we feel like Gus could do something of value to you, Gus shows up out of nowhere in a discreet way and says, “Hey, can you help me write the job description?” You can. This is a cleaner way to interface with AI.
Some companies have been working on AI for a million years but haven't been paying attention until now, while others have only realized the opportunity in the last few years. Which camp does Gusto belong to?
The big change for me is that when we talk about software programming, it's inaccessible to most people. You have to learn how to code and go to school for years. Machine learning was even less accessible. Because it's a very specific type of software engineer, and you have to have a data science skillset and know how to create things like artificial neural networks.
The main thing that has changed recently is the interface for creating ML and AI applications. [has become] It will be more accessible to everyone. Previously, we had to learn the language of computers and go to school for it, but now computers are better able to understand humans. That doesn't seem like a big deal, but when you think about it, it makes building software applications a lot easier.
That's exactly what we've seen with Gusto. Even people who are not software engineers but have a little technical awareness can build extremely powerful and innovative AI applications. In fact, we use a lot of support teams to extend the functionality of Gus, but they don't know how to program at all. The interfaces they use today simply mean they can do the same things software engineers have always done without having to learn how to code. If you'd like, I'd be happy to walk you through one example of each.
That's great.
There is a person who has been working at this company for about 5 years. His name is Eric Rodriguez and he actually joined our customer support team. [and then] I have been transferred to our IT team. While he was on that team, he started to get really interested in AI, and his boss came to me and said, “Hey, we're going to get into AI. I want you to see it. ”It was the first time I met him in person and he showed me what he had built. It was essentially our copilot tool. [customer experience] Ask your team a question and get answers in natural language. Similar to ChatGPT, except you have access to an internal knowledge base on how to navigate within the app.
At this point I showed this to my support team and they loved it. This has completely changed my workflow and efficiency. Basically, whenever you get a support ticket, instead of referring to the knowledge base that we've built, you actually ask this CoPilot tool a question, and the CoPilot tool actually answers your question. There is still a human between CoPilot and the customer, but in many cases you can get a response from the CoPilot tool and copy and paste it to the customer. They verify that it's accurate, and most of the time it is.
transferred immediately [Eric] To the software engineering team. Believe it or not, he actually reports to me and is currently one of our best engineers. Because he was one of the early adopters of experimenting with AI and is now at the forefront of building AI applications at Gusto.
Not everyone is as technically minded as Eric, but at Gusto, we're learning how to leverage non-technical expertise within our company, especially our customer support team, to build more powerful AI applications. I found it. , allowing gas to do more.
Every time our customer support team receives a support ticket, one of our customers contacts us because they need our support team's help with something, and if it happens repeatedly, Ask our customer support team to write you a recipe. Gas, that is, you can actually teach gas without any technical ability. They teach Gus how to solve that customer's problem and sometimes even take action.
We built an internal interface, an internal tool, that allows us to write instructions to Gus in natural language about how to handle such cases. And there's actually a no-code way that our support team can tell Gus to call specific APIs to accomplish a task.
There's a lot of conversation right now that goes along the lines of, “Let's eliminate all jobs in this field and pay millions of dollars to hire AI specialists with unique skill sets.” And I think that's the wrong way to go about it. Because the people who can develop AI applications are the ones who actually have domain expertise, even if they don't have technical expertise. In fact, Gusto can upskill many of its employees and help them build AI applications.
The scary scenario for AI is a top-down scenario where executives say “we need to use AI” and are disconnected from the reality of how people work. This seems more bottom-up, with teams building tools to tell the AI what it can do.
that's right. In fact, non-technical people close to customers talk to customers every day and have a real idea of what kind of situations customers might find themselves in and what they're confused about on average. They have a much deeper understanding than engineers. . So they're actually in a better position than engineers or AI scientists to write instructions on the gas to solve that problem.
I think others I spoke to noticed the same thing. Great AI engineers are actually domain experts who have learned how to write great prompts.
When you think about how this will play out over the next few years, do you think the head count for the different teams at the company will be about the same, or do you think it will change over time as AI is implemented across the company? mosquito?
I think my role has evolved a little. I think you'll see a lot of CX people actually creating recipes and doing things like making quick adjustments to improve the AI, rather than answering questions directly. Everyone will move up the abstraction layer. This obviously improves your company's efficiency and improves the customer experience by getting answers to your questions quickly.
This will allow Gusto to do more for its customers. We have a huge roadmap for what we want to do, but resource constraints prevent us from doing it.