AWS Summit London 24: Tui, Zilch & Lonely Planet take off with GenAI

Given the remote, automated nature of our connected, cloud-assisted, AI-driven workplaces, it can be a shock to walk into a scrum of people at a physical event; especially one with over 18,000 participants, as was the case at this week’s AWS Summit event in London.

As I was expertly shepherded into the event’s keynote at London’s Excel Centre, the army of new and experienced cloud user delegates (many of whom are software developers) was impressive, although slightly less so was the gender ratio – which I’d estimate as one woman to every seven men, despite Amazon’s best efforts to improve the pipeline in this region.

It will come as no surprise to learn that artificial intelligence was the summit’s main theme, with keynote speaker Tanuja Randery who is AWS’s VP and managing director of EMEA, announcing that the UK economy is sitting on the brink of a £520bn opportunity in accelerating AI adoption by 2030, according to a recent Amazon report.

Naturally, the vendor is offering some tools to assist with the unlocking of these business opportunities. Among one of the first announcements at the event was news that the hyperscaler’s GenAI offering, Amazon Bedrock, has now launched in the UK.

First unveiled last year, AWS’s Bedrock service includes an all-you-can-eat menu of gen AI models from Amazon as well as other third-party partners, including AI21 Labs, Anthropic, Cohere, Mistral AI, Stability AI and Meta Llama – all offered through an API.

Bedrock also allows users to build apps on top of generative AI models and customise them with their own proprietary data and, as of yesterday, with their own proprietary AIs.

Another stalwart the vendor is keen to push is its SageMaker product, a managed service in AWS public cloud. This tool is not new – it was released in 2017 – but it provides the tools to build, train and deploy machine learning models for predictive analytics application. Many early GenAI innovations have seen this tool used in tandem with Bedrock.

Dentsu and Lonely Planet

Several enterprise use cases in GenAI highlighted how firms can increase their bottom line using these technologies. Randery pointed to Amazon’s partnership with global ad firm Dentsu, which is adopting SageMaker and Bedrock to scale its genAI use.

In Brazil one of Dentsu’s clients, Nissin Foods, used the tools to enable a campaign to achieve 109% increase in year-on-year sales and to boost social engagements by 21 million.

In Portugal, Randery added that Dentsu was able to launch an easy reading app to over 40 million people with learning difficulties.

Fellow keynote speaker Francessca Vasquez, AWS VP of professional services and GenAI Innovation Centre, also revealed that guidebook brand Lonely Planet was developing GenAI with tools to help people plan incredible trips and customise their itineraries.

“They are using Anthropic’s quad model, through Amazon Bedrock, and as a result they have reduced itinerary generation costs by nearly 80%,” she said.

Tui: from inspiration to destination

Sticking with the theme of travel, leisure group Tui’s CIO and guest keynote speaker Pieter Jordaan revealed how the firm was using AI to make the most of its petabytes of data.

“We needed to break the norm to go big with cloud and AI. To give you the scale of what we’re trying to achieve we’re taking the equivalent of the entire population of Australia (over 21m) away on holiday every year,” said the CIO.

Tui has partnered with AWS for seven years as part of its digital transformation journey, Jordaan added, and is using data in the cloud – and now AI – to “redefine personalised travel experiences”.

According to Jordaan, during the pandemic, rather that scale down IT operations Tui decided to capitalise on this unexpected downtime and “go big and to go all in on AWS”.

This involved building a cloud-native platform and globally migrating its call centres to Amazon Connect to improve the customer experience –a move he claims that has reduced agent handling time by 20%.

“We transformed our IT organisation into a globally distributed workforce and have now trained 90% of our IT staff on AWS skills. And now we’re in the process of training them on generative AI,” he said.

Expedia: opening the data treasure trove with gen AI and LLMs

Post-Covid, the strategy has been to lean in and focus on creating curated personalised travel experiences, which AI has enabled them to do, Jordaan maintained.

“The move to the cloud native platform allowed us to accelerate and embrace AI. We increased our model deployment in Amazon SageMaker by 1000% in one year,” he claimed.

Last year, the CIO launched Tui’s own AI Lab to drive generative AI adoption through the entire organisation – generating use cases from board level to the flight staff to the travel reps in the resort.

One of the biggest challenges in this area was to find use cases that delivered measurable value, Jordaan admitted.

His solution for this was to use AWS PartyRock – a tool that can be used within Bedrock as a kind of ‘AI playground’ to make foundational models with AIs from Amazon and other leading AI companies.

“We used PartyRock to work backwards and coach to train the teams to really produce clear and measurable problem statements so that we could quickly protype ideas,” Jordaan explained.

Site-seeing

One such use case to emerge was content generation. At any one time the travel group has 100,000s of holidays on sale and they all need inspirational content. Yet to create SEO to scale that matches the tone-of-voice of the brand and is also personalised s a huge challenge, the CIO noted.

“And yet, from a business perspective, even if we achieve just a 1% point move in online organic traffic with better content, this is the equivalent of 8.5m Euros in cost,” he calculated.

“The team was already familiar with SageMaker and Llama and we were able to use Bedrock to manage our foundation models and enhance our SEO content beyond just the travel facts, and really personalise with a tone-of-voice that matches the experience,” he said.

Of Bedrock he added: “By utilising model stacking we could solve a problem that we couldn’t solve with a single model.

“We first used Llama2 to create the content and then we would pass it to [Anthropic’s] Claude 2 to do alignment on tone of voice, and then we moved onto the fine tuning for the languages.

He added that in GenAI multiple model choice is becoming important: “that ability to quickly iterate over new models whether that’s inspirational search or cutting our call centre time in half.”

Another product in its AI armoury is AWS’s gen AI-powered chatbot for called Q, which has been trained on 17 years of AWS data, making it useful to help build web applications.

Jordaan added that Tui was now in the process of rolling out Amazon Q developer to over 2000 of its engineers to improve efficiency.

From Zilch to $2bn

Another speaker, this time from the world of fintech, was Philip Belamant – CEO and cofounder of London-based Zilch, a buy-now, pay-later unicorn, currently valued at $2bn.

Explaining the company’s unique business model Belamant said: “In the UK we are paying $150 a year on fees and interest on credit. We wanted to drive this sum to zero – or zilch – for customers by introducing brands directly to buyers at the point of payment and using ad budgets to subsidise the credit process.

“So, a customer buys a pair of shoes. We underwrite the customer in real time for an affordable line of credit.  We then package up that intent and share it with our ad partners and what they allow brands to do is programmatically bid in real time on that BNPL sale.

“The customer takes out a line of credit and pays zero interest and zero fees of any kind and Zilch is still able to operate this at a 50% + profit margin. This is what we’re calling the world’s first ad subsidise payments network. We’re currently doing 3,000 transactions of these every second and climbing.”

Cloud native

Infrastructure wise, Zilch was born on AWS Cloud says Belamant, and the firm has been scaling the company the cloud ever since. At the event the CEO announced “a deepening in its collaboration” with the vendor focussing on AI.

Bud Financial: using AI and LLMs to drive bankable insights

According to Belamant, the fintech uses AI across four lines: fraud, credit, customer serving and buyer intent prediction. Touching on two of these areas he says:

“If you look at credit underwriting this is fundamental to our business. Our team would take weeks to build a new underwriting model and then test that model and deploy it. Today with SageMaker we can do all of this within just days.

“And here’s the interesting part: we are currently ingesting more than ten times the data for that model than we used today. Initial tests are showing that the predictability of this model is now almost twice as predictive as it used to be. And we’re just getting started.

The second area the fintech entrepreneur enlarged on was intent prediction: “This is critical: A customer tells us where they like to shop. What is interesting is what if you knew where you were going before you told us? And one day what if we knew what was valuable to you before even you knew? So, we’re going to be combining GenAI and ML to serve this up.

He explains: “We understand what the customer might need or view as valuable, we feed it to gen AI through Bedrock and various models and then we serve this in natural language or a really rich content way – it could be images – this will transform content as we know it.”

Belamant claims that Zilch has already saved customers half a billion dollars in fees and interest, adding “if we get this right, we may have the most credible way of eliminating the high cost of consumer credit for good.”