A question that enterprises are staring to ask themselves as the agentic revolution rumbles on is whether the Customer Experience (CX) is going to be the same if users are talking to a human or an AI. The answer may be somewhat nuanced.
Clay is a US AI firm set up in 2017 with a mission today to, in its own words, “power GTM (go-to-market) engineers:
Our goal is to give every company the tools to build their own revenue system, guided by the belief that we started with: give creative people powerful building blocks and let them shape those blocks around their ideas”. Customers include the likes of OpenAI, Stripe, HubSpot, Uber, Box, and Mistral.
The firm also has a claim to fame as the first firm to use Fin’s AI agent technology to scale customer support inside its Slack community. According to George Dilthey, Clay’s Head of Support:
Clay’s support has always been centered around Slack. Our Slack community was was really one of the earlier big growth moments that we had. Even as we’ve rolled out some more traditional channels, like the in-app messenger or email or whatever it might be, Slack has always still been a centerpiece for it.
This Slack centricity is different due to the public domain nature of the community:
It’s out there in the public, and so the risk is a little bit higher. The way that you think about it is, ‘OK, can we do this? My CEO is looking at these Slacks, and our co-founders, everybody’s there. Other customers are looking at it, so if there are policies that apply differently to different customers, you have to think through those things. So, just being out in public is just a little bit changes the game a little bit.
The other thing is that the expectation of privacy is just very different. We have a lot of things that we’re starting to build out with Fin, where it has access to your billing information or your table data or whatever it might be in the app, and we can’t just throw that stuff out into a public forum. So we actually, in a lot of ways, have to have a de-graded experience for Fin on Slack, not because the capability is not there, but because we have to think about that sort of that that rollout a little bit differently.
CX to the fore
Keeping Customer Experience (CX) front-and-center here is a strategic priority for everything to do with AI rollout, adds Dilthey:
Our Head of CX always talks about no matter what you’re doing with these AI things, you always want to think about the Customer Experience first. Being in Slack means that we can meet our customers where they’re at, and and having that experience, not only in our community, but also with our enterprise customers and their Slack channels, just makes it really nice, meeting them where they’re working and be able to keep that seamless.
In terms of its overall use of Fin’s agents, Clay has done a lot to build out its capabilities, with a focus on two things in particular – the context that Fin has and the data that Fin has. In the context portion, Dilthey elaborates:
We’re thinking about this knowledge base, which used to be this place that would get updated once every three months – maybe. It was always out of date and didn’t update with the same cadence as the product was shipping. We’ve had to re-think the way that we think about that context. Can we make it self-healing? Can we use cloud-managed agents that run over our knowledge base on a regular basis to make sure that we’re we’re keeping up with releases and changes in the product. So, we’re thinking a lot about context.
Also front of mind is the role of data connectors and what access Fin has, Dilthey adds:
What we sort of started to realize is that the kind of work that we’re doing with Fin is actually the same work that we’d want to do for any agent. I was starting to see other parts of the company doing very similar things with our Sales Enablement docs internally. How do we think about keeping that content up-to-date, and how do we think about keeping our engineering on-boarding up-to-date or whatever it might be? It really applies anywhere, even more on a macro-level. How do you know that what is in Slack is up to date? All of that becomes so much more important in this agentic era because you have to you have to really be specific about the things that it is pulling from.
That’s led to what Dilthey alludes to as “some really interesting moments”, explaining:
I’m getting pulled into conversations to think about these problems in other parts of business, even in the way that we’re building our product, we’re building agents in our product. I’m learning from that, and talking to the engineers that are building that out, because a lot of it is just a lot of the same process. It’s been really fun to see where that can apply differently. It’s sort of like, you know, ‘The tide rises all boats’ kind of thing.
Great expectations
It’s not only the tide rising though; so too is the floor of what is expected from support. Dilthey says:
Where we used to have regular product support specialists and then enterprise product support specialists, we’ve flattened that a little bit, because all of the questions that are making it to humans are the harder ones and the enterprise ones, and we have to be a little bit more strategic and think about those a little bit more, a little bit differently.
But now that our humans are focused on that, we want to start thinking about the ‘human in the loop’ opportunity. I used to think of AI support as very much like either AI handles it or a human handles it, and now I think there’s a really interesting middle ground where I can speed up my team, when they’re checking logs or they’re checking whatever it might be, whether it’s through Copilot or a ‘human in the loop’, just make those processes that they’re doing faster at that kind of stuff.
Getting more human?
And he’s conscious that agents capabilities are evolving all the time:
The more exposure they can get to these things, with the judgment they have, the knowledge they have, they can pro-actively identify things that, in theory, a human could if they had all the time in the world, but in practice, they can’t.
That’s good though, yes? Up to a point, it seems:
It’s scary too because you want to give [agents] access to to everything to be more effective, but you also have to be careful about the things that you’re throwing out there.
And there is still much to be said for human experience coming into play, he notes:
I find that the folks that are still the best in our team are the ones that were good five years ago, before there was AI. We’re starting to shift the types of questions that they’re answering, that human aspect and that strategic thinking and that that pattern matching, across different things that you’re seeing in the queue. How they collaborate internally with Product and Engineering, all of that becomes so much more important and is so much more frequent, because you’re not just answering the password reset questions anymore. There’s a level of curiosity and adaptability that [humans] have, that they’ve had since before the AI world. Being able to carry that through, even as they make themselves faster using AI, has been really kind of fun to see.
