AWS and Amperity Share Practical Lessons on Preparing Enterprise Data for AI

Industry leaders highlight how trusted, event-driven customer data can support real-time decisions and prepare enterprises for an agentic AI future.

Preparing enterprise data for AI requires more than collecting large volumes of customer information. Organisations need trusted data foundations that can combine historical records with real-time signals and support faster decisions.

That was a key message from leaders at AWS and Amperity. They said organisations should start with focused use cases rather than attempting to transform everything at once.

Speaking during the Architecting for Trusted, Real-Time Decisions session at Amperity’s Amplify 2026 conference, Steven M. Elinson, Director, AWS for Travel & Hospitality, and Thomas Koep, Vice President of Customer Strategy at Amperity, outlined a practical approach.

Organisations can combine historical customer profiles with live behavioural signals. This can create a trusted customer context that supports decisions across the enterprise.

Real-time personalisation and revenue recovery offer useful starting points. However, the larger opportunity is creating governed customer context that marketing, operations, customer care and AI agents can all trust.

Why Real-Time Customer Context Matters for Enterprise Data for AI

Travel and hospitality provide a clear example of why real-time customer intelligence matters.

Elinson said the industry’s historical “look-to-book” ratio had grown from about 10,000 to one to 100,000 to one. Metasearch and online travel agencies helped drive that increase by expanding consumer choice.

In an agentic future, that ratio is predicted to reach one million to one. AI agents could search and assemble combinations on behalf of consumers.

At the same time, one customer may have different personas depending on the occasion. As a result, businesses need to know more than who the person is. They also need to understand why that person is engaging at that moment.

“It’s not just enough to know who the person is, you have to know the occasion and why they’re coming to visit you,” Elinson said.

Travel inventory adds another challenge. An unsold hotel room, airline seat or cruise cabin cannot be held for later. A lost booking can also remove opportunities to earn ancillary revenue.

Real-time personalisation and abandoned-cart recovery can provide an immediate commercial case. However, Elinson said the more strategic opportunity involves eliminating fragmented customer views.

Marketing, operations and customer care often hold different versions of the same traveller or guest. That fragmentation can create inconsistent experiences. It could also increase risk as AI agents begin making and executing decisions at scale.

“The future requires that we eliminate that and have a single version of truth. Although the revenue recovery pieces we’re talking about today are important, it’s this second piece, the unified truth, that’s going to be most important in the agentic future that’s coming,” Elinson said.

AI agents will need accurate profile information and details about a customer’s current journey. They will also need sentiment data and relevant historical context.

A shared and governed customer record gives human teams and AI agents the same trusted foundation for decisions.

Build a Trusted Enterprise Data Foundation for AI

Many organisations still use batch processes and daily refreshes to resolve customer identity, build segments and activate data.

That delay creates a problem. A customer’s context can change before the organisation responds. The customer could also choose another brand during that window.

Koep said businesses no longer need to operate real-time and batch data as separate systems.

Organisations can collect unknown signals, such as anonymous browsing behaviour, over time. They can then connect those signals to an existing customer profile when the person logs in or provides identifying information.

“The interesting thing going forward: you can now use all those unknown signals, bring them into your tenant, store them over time, and the second a guest logs in or provides a piece of PII, you unify all that historical browsing behavior to an Amperity ID,” Koep said.

Once connected, businesses can use in-session behaviour alongside transaction history, preferences and other profile attributes. They do not need to wait for an upstream micro-batch or daily refresh.

This approach supports decisions based on the customer’s current context. At the same time, historical knowledge can help prevent businesses from overreacting to a single click or view.

Elinson said event-driven architecture enables organisations to detect signals and ingest them quickly. They can then use those insights to shape customer experiences.

For travel and hospitality businesses, the capability needs to address two lasting priorities. Businesses want to improve the guest experience while also strengthening operating margins.

“So how do you do both? How do you elevate the guest experience while also improving operating margins? It’s through things like event-driven architectures that give you the data and insights to inform both,” Elinson said.

Those signals can include structured and unstructured data. They may also include clickstream activity, information supplied directly by customers and observations from transaction systems. Partner data and third-party sources can add further context.

Streaming services can ingest these signals at high volume and low latency. Organisations can then standardise and catalogue the data. From there, they can make it available through storage and database services suited to each outcome.

Elinson cautioned against assuming that one type of database can solve every problem.

Real-time personalisation, economical long-term storage and enterprise analytics can have different requirements. Relationship-based insights may also require fit-for-purpose services.

Another priority involves making data catalogues open and available across the enterprise, Elinson said. Organisations should avoid locking their data inside a single vendor environment.

Open table formats can give organisations greater flexibility. They can experiment with emerging technologies and connect customer information to new tools without slowing innovation.

This flexibility becomes more important as AI capabilities and vendors evolve. Organisations need to test new approaches while maintaining governed access to trusted customer data.

Koep also described an open, API-based architecture that allows real-time customer information to serve different applications.

Combining event streaming and batch processes can also reduce reliance on multiple point solutions.

“Audit your tech stack and ask: do I still need event processor X and event processor Z? Because you can take that money and reinvest it back into the platforms that hold all your customer information in the first place,” Koep said.

Start Small When Preparing Enterprise Data for AI

Organisations do not need to ingest every possible signal before they begin.

Elinson recommended taking a “vertical slice.” The organisation first identifies the data needed to test one business hypothesis. It then builds the supporting pipeline, validates the outcome, and repeats the process.

“We’ve seen some brands spend two years trying to get every single possible signal into their ecosystem before they can go and do personalisation or revenue recovery, and that doesn’t bring value to the business,” Elinson said.

This approach helps organisations move from architecture diagrams to measurable outcomes. They do not need to complete a multi-year infrastructure transformation before delivering value.

It also creates an iterative path for expansion. Each successful use case can show which additional signals, services and capabilities could deliver the next increment of value.

Elinson highlighted examples that demonstrate the commercial and operational impact of event-driven architectures.

South Korean hyper-retailer Lotte Mart increased its personalisation conversion rate by more than five times after moving to an event-driven architecture. The company improved its understanding of customers, their journeys and how to activate that context.

Elinson also said systems need enough elasticity to respond to sudden demand.

Limited-time offers and major events can produce significant transaction spikes. Architecture must therefore scale quickly when demand rises. It should also reduce capacity afterwards to avoid unnecessary operating costs.

Expedia’s digital model provides another example.

Elinson said the travel seller has complete coverage across its data pipeline. That gives channel, revenue, distribution, marketing, and other teams access to insights they can use to drive outcomes.

“We know the architecture works. It works at scale, and it works for some of the most loved and well-known brands in the world,” Elinson said.

Extend Real-Time Data Across the Enterprise

Digital personalisation is one of the easiest applications to understand, Koep said. However, real-time customer context can also support loyalty programs and operations.

For example, a hospitality business could combine geolocation or check-in signals with an existing guest profile. The business could then trigger a personalised welcome.

A guest may also qualify for a higher loyalty tier at check-in. The system could update that status immediately and send it to downstream systems before the guest reaches their room.

“The signals are just events and traits, you can send that information at the speed your customers actually expect, hyper-personalising those downstream systems as fast as possible,” Koep said.

Closed-loop data creates additional opportunities.

Email opens, clicks and web behaviour can flow back into the customer profile immediately. Teams can then optimise experiences faster without waiting for other systems to process and return the information.

Prepare Enterprise Data for AI Agents

Real-time inputs can give AI systems access to current context. Previously, four-hour batches or daily refreshes often delayed that information.

More timely data creates opportunities to train, update and apply models using recent customer activity. Organisations can do this while maintaining a connection to a trusted enterprise profile.

Koep summarised three capabilities that organisations should prepare for.

First, they need to connect unknown activity to a known customer profile. Second, they need to use that context in real-time journeys at scale. Finally, they should extend the architecture beyond digital experiences into loyalty, operations and other enterprise programs.

Elinson said organisations should aim to provide governed access to high-quality and secure customer information.

“You want to give your entire organisation access to that governed, single source of truth: high-quality, secure information that lets you continue to accelerate outcomes and make sure you’re ready for what’s coming,” Elinson said.

Preparing enterprise data for AI is therefore about more than making marketing faster.

The goal is to give authorised systems and teams access to trusted context. That foundation can help them make better decisions when those decisions matter most.

About the Amplify 2026 Discussion

The insights in this thought-leadership material are drawn from the publicly available Architecting for Trusted, Real-Time Decisions session at Amperity’s Amplify 2026 conference.

This piece focuses on comments from:

• Steven M. Elinson, Director, AWS for Travel & Hospitality, AWS; and
• Thomas Koep, Vice President of Customer Strategy, Amperity.

Session page and transcript here.

About Amperity

Amperity helps brands act on trusted customer context. Its AI-powered Customer Data Platform resolves identity and unifies fragmented data with real-time signals to power more relevant decisions across marketing, service, analytics, and AI systems. More than 400 leading brands worldwide, including Accent Group, Alaska Airlines, DICK’S Sporting Goods, BECU and Wyndham Hotels & Resorts, rely on Amperity to improve customer experiences and drive measurable growth. Founded in 2016, Amperity operates globally with offices in Seattle, New York City, London, and Melbourne. Learn more at amperity.com.