Hi Ruplu, it's Chris. So a couple of things in that because it is a good question. The first thing just for clarity, when we talk about close to 1,000 customers, that's close to half our client base is using generative AI that we have installed. And some of that is our partner's technology that we've installed and configure and manage, some of that is obviously our intellectual property that we've installed and managed. And that is using at scale day-to-day using our operations. This is not a proof-of-concept stuff. This is real, industrial enterprise use of the AI technology. And for clarity, when you look at what clients are using that for, it is very different. There are some clients who are using a full stack where we're doing automated bot where we're using our generative QA system, where we are using our generative AI coaching system, where we are using our generative [AI TAS] (ph) system, all sorts of things like that. When it's not customer-facing, clients tend to be very receptive to using it. And those proof-of-concepts, frankly go very quickly into production. When it is anything to do with an external brand engagement, we continue to see despite all the press releases, everyone talking about, we continue to see a reluctance to interface with high-value interactions with customers where the AI is doing everything. What we see and still this day still conversations is that it's normally a human the last connect but that human is AI-powered to get better answers, better help service in a more intimate personalized way and deliver what they need to do. And so we still see that playing out to this case. Obviously, with the model is getting more advanced, that will somewhat change. But as we've seen and shown as we've kind of automated and put the technology in, we've won net new business that historically hasn't been there. In terms of proof-of-concepts, we still have hundreds of proof-of-concepts, but very transparently, some we roll out and the client goes, okay, we've done the economic model on this. It is going to cost too much to do the queries to an LLM versus how we can do it other ways. We are going to kill it. Others are like, yes, this is perfect. Let's get it into production as quickly as possible. And so this quarter, we added, I think, 25 new -- I think it's 25 new at scale implementations of our Gen AI technology into our client base from POCs. And so we are converting them, but we are always having these new POCS coming into the ecosystem.