Jaja Finance launches a GenAI chat assistant using Anthropic's Claude 3 and Amazon Bedrock
London based Jaja Finance recently announced they've launched, Airi, a virtual chat assistant using Anthropic's Claude 3 model family – running on top of Amazon Bedrock.
They published the news back in May and this one has been on my list to write about! I specifically wanted to make sure I'd documented this one for quite a few different reasons.
First, many Conversational AI vendors will no doubt be somewhat disappointed to discover Jaja has built something like this themselves.
Second, quite a few will be curious as to how they've done this technically.
Third, many will be wondering just how they're managing the propensity for the technology to get creative. And, yes, for any of the geeks reading – we know this is a feature of large language models, not a failure. Just, when you're working in regulated industries, it's rather useful to have an idea of what your virtual assistant(s) are going to be saying to customers before they say it. (Or, like Air Canada, start creating policy on the fly.)
And Fourth, I think a lot of other banks will very much want to hear about Jaja's journey, especially given their usage of Claude and Bedrock.
I've long been a fan of what the team at Jaja have been producing in terms of their digitally native credit card product so I'm keen to see how their launch has gone. I'll reach out to the team to see if they'd be keen to discuss this.
So, here's a bit of detail:
The new service, ‘Airi’, uses Amazon Bedrock, a fully managed service from Amazon Web Services (AWS) that offers a choice of high-performing foundation models (FMs), to access Claude 3 models, offering increased intelligence with near-human levels of responsiveness and improved data accuracy.
The benefit of what Amazon is offering here is being able to effectively plugin to an API, rather than having to spend a lot of time setting up and maintaining the infrastructure for all this, enabling the Jaja team to focus on the higher level aspects like integrating it into their app experience. You can read more about Amazon Bedrock here. Amazon has also done extensive work when it comes to the likes of implementing safeguards. There's more detail about their Bedrock guardrails including, for example, PII redaction and hallucination detection.
The release continues:
Jaja projects that ‘Airi’ will reduce average customer response times by over 65%, allowing customers to get answers to their queries much faster than before, freeing-up Jaja teams to spend time on more complex queries fielded by the chat assistant.
I would think a 65% reduction in response times is eminently achievable.
The release goes on to explain that Airi will be implemented into the company's in-app customer messaging feature. The Jaja team are also expecting Airi to handle around 30% of routine customer enquiries. It's then expected that this should really help Jaja's human agents, thus:
With ‘Airi’ predicted to handle up to 30% of routine customer enquiries, Jaja’s customer service teams are expected to have 30-40% more time available to focus on complex, high-value customer interactions that require human expertise. This is critical as Jaja pursues its ambitious growth plans.
Here's a quote from Jaja's CEO, Dave Chan:
“As a digital lender we’re already employing sophisticated data models and advanced technologies to ensure we’re giving our customers the tools they need to manage their credit well. Adding an intelligent generative AI assistant like ‘Airi’ will help us unlock the power of advanced AI for our customers, while ensuring their financial health and wellbeing."
Good luck Team Jaja!
Some contacts for Jaja:
Here's the CEO Dave Chan:
And here's Chief Commercial Officer, Lucas Dalglish:
