00:00What have been technically some of the challenges, but also the opportunities
00:03that you're seeing, especially how much is changing with AI so quickly? What are
00:07you seeing from a technical side of things? Yeah, so definitely the advent of
00:10large language models and that really accelerated our progress. Like we've been
00:14AI first since the start, we built our own AI models and our own AI algorithms
00:19to run and automate our technology, but all of a sudden you now have LLMs
00:23behind the scenes that are kind of like the engine or the platform that you're
00:26building on top of, and the rate of innovation of those is incredible. So
00:30we're layering that into our survey design, into our analysis, but we get that
00:34automatic improvement because we build what we call LLM ops, where basically we
00:39are able to, whenever a new model comes out, we test it against all the other
00:42models based on the data we're using and whichever one performs best, we hotspot
00:46in. So we get this constant automatic innovation without having to invest in
00:49ourselves. So that's been incredible and really accelerated us kind of maybe two
00:53years ago. I think that's like the way I think about it is that the problem has
00:56changed in what we're trying to solve, just the tools we are using to deliver
01:00the solution has changed quite a lot. So we were trying to do a lot of this
01:02ourselves. It was really difficult because we just, we were trying to like, okay,
01:06how do we do survey design using technology instead of people? Like that was a
01:10highly complex problem to try and figure out. And then suddenly you could like
01:13fine tune a model to try and replicate how do you design a survey? What does a
01:16good survey look like and how can you retrieve that similar on the analysis? So
01:20it's accelerated kind of our journey towards our desired outcome. And that's been
01:24just, as I said, sometimes you're better off being lucky and good than good. Timing
01:28matters a lot with companies. And when the AI kind of wave and then when things
01:33started to improve, we found ourselves with a foundational set of customers, you
01:37know, we were profitable, making quite a lot of money for our size and we're able to
01:41like invest quickly, be agile with our clients so that they could test and co-create
01:46stuff with us and try and solve actual problems with AI. Whereas there's a lot of
01:50kind of, as you say, hype and what does it actually do for me? And yeah, it's kind of a better
01:54search. You know, that's kind of what people materialize it as. Whereas for us, it was
01:58we were able to go very practically with solutions to everyday problems and they
02:02just want us to use it. Like they still want the same problem solved. However, whatever
02:06technology we use to solve it, they don't really mind. And I think that's kind of an
02:10important thing for me on the commercial side because people are a bit scared of
02:12it going, oh, well, if, you know, I open the door, are you going to run through and
02:16then suddenly I don't have a job? Whereas it's actually just a means to solving the problem.
02:20Right.
Comments