Everything I wrote in innovation, lab excellence is valid for AI deployment today, you asked a fundamental question. And it was one of the big points in my book, which is buy versus bill. And this is the same advice that I would give for an AI team today, you have to buy this technology, all but the very largest global banks like the JP Morgan's of the world, only a few of them are able to actually build their own technology.

So if you're looking at an innovation programme, or an AI programme, their job is to prove that this stuff can work. All right, their job is not to deliver ready to use code ready to use a AI that is ready for production. Because you really expect them to build their own large language model and know they prove it works.

And then you need to hire, particularly for the likes of AI, one of these larger firms is going to come in and hopefully have enough liability and insurance so that when your chat GPT style chatbot comes off the rails and give somebody the wrong answer. You can you can blame them with the losses.

So that's why we work with Airtel to say, let's bridge this gap. Let's try to be the boundary between the lenders and the borrowers. And the only way you can do that is having standardised credit score.

So Airtel provides all the transactions when you buy airtime on your phone, when you buy data, pay a bill, you know, that information tells us probably you have some responsibility in your house, you're moving around, probably you have some kind of mobility, are you having so many people different one sending to you, tells you you're making sales.

But if you're just receiving from one person, we can can infer probably your law student getting some update from the parents.

If you have to remember what we talked machine learning AI today is not something that's come around today, right? What has changed today is our ability to store enormous amount of data economically. Number two is the processing speeds we have today. You know, you want a search bar before you type, your third word is already telling you what it should be. So think about it. And there are millions of people doing it at the same time, any second. And the third thing is the speed of transmission of information.

I think those three in combination literally are the most fertile ground to bring AI to life.

And that's what we've essentially done. But be mindful that when you're doing things at that speed, there are things that could happen which go out of your control.

Because again, credit scores really don't resonate well for that customer, but the fact that I can access more loans and more services from the bank makes a lot of sense, right?

So, the messaging itself can we change, make it more people friendly, make it more empathetic, again, is is an important factor that we keep working on. So understand the behaviour the formats, the messaging, and then devise and develop a complete strategy around customer engagement, what is the life cycle that we can build? So segment and perhaps micro segment the customer and build optimised life cycles for these micro segments of customers that we are onboarding and all that is to ensure that these people understand their loans make their payments on time?

There shouldn't be any instance where you know we have to limit the access to the device. So the idea is never to reach that hence, how do we use AI and ML to create very effective and efficient life cycles, messaging journeys for these people. So that, you know, the delinquencies are in check.

So I think in the beginning of this business, we realised that we have to put in some real time in order to prove that with potential customers. So we've been at it for just slightly over two years. And we're able to not only back tests from source data, but actually test or run our models on real data.

And the results have been quite impressive, if I say so myself, we're happy with that confidence of 86%. We do provide then a boolean results where you know, it's a Yay or Nay, We realise we don't want to reinvent the wheel with another score that one has to contend with.

But from our side of things, I think the main thing that we're seeing is however clever you want to be about it, however many propensity scores you want to use, however you want to segment your own collections portfolio and look to engage with people, if you're not able to contact that customer, it's going to be very difficult to get a good outcome for either you or for the customer themselves.

So because of those front-end online application journeys that we're capturing data from, and because of the recency - we're updating our full database - we already have a significant coverage of UK contact channel information, which we're able to help financial services firms get access to and to engage with their consumer, so they actually can get that conversation started. And they can try and get that resolution for them. But then if they if they can't have that first conversation, the rest of the clever stuff that they can do kind of goes out the window a little bit.