Answers to questions like how do companies envision leaning into AI and how can it disrupt business will determine what initiatives the travel industry gets behind.

How can travel companies get started in AI – and make it core to how they operate? And how can they implement it and not lose sight of the need for an actual return on their investment?

We shared answers to these questions and more at Skift Data + AI Summit last month in New York, and go more in depth below.

First, the highlights:

Bootstrap Your AI Culture: Clarity Over Everything Else. Encourage a culture of risk-taking and experimentation so that the teams engage early with AI. Build upon momentum with targeted and meaningful initiatives. Celebrate innovative ideas regardless of where they come from. Encourage a culture of risk-taking and experimentation so that the teams engage early with AI.

Plan on AI Implementation to require investment and get complex over time: Utilize the AI Implementation building blocks as a practical list of areas of focus and investment. Understand the building blocks needed for a practical ecosystem that enables AI solutions. Pick the parts you own and choose long-term partners for the rest. Know your risk profile and understand what it will take to make AI initiatives successful and generally available.

AI is the leveler of where innovation comes from, so build your AI enablement function: Commit the core team to enable a larger team instead of putting pressure on one team to innovate. Emphasize that success is shared across the organization. Organize the core team in a manner that best works for your organization: central, federated, or hybrid.

Do not over-index on traditional measures of ROI: AI unlocks many categories for positive ROI but requires investments to level up. Start with what you know – improve that, but in parallel, focus on leveling up your tech ecosystem AND your teams. Investing in your AI ecosystem and early initiatives is going to set the tone for generational transformation. Any short-term gains or setbacks are not indicative of long-term transformation.

1. How to Get Your Team From AI-Curious to AI Native

Build a culture of experimentation: Recognize that truly embracing AI requires a ground-up transformation of your existing infrastructure. The level of transformation is determined by your data readiness, risk/error tolerance level, and ROI potential, among other things. This is an opportunity to completely reimagine your stack, with investment in AI literacy for the whole team. It’s a job for everyone to take on – from personal productivity to business process transformation to customer experiences that delight and empower.

Moreover, people may need permission to engage and experiment. They need guardrails to be able to fail safely. Leadership has to open doors to ALL parts of the organization. You need both top-down encouragement and bottom-up enthusiasm to build momentum. Approaching it otherwise will limit what you can get out of it.

Match the right tools for the right jobs: It’s essential to start by empowering your teams with the right tools, but that’s merely the first step in unlocking the art of the possible. It is not just one set of tools. You have to match the right tools for the right job functions. In other words, while acquiring the right set of tools is important, the true potential of AI can only be unlocked when your teams can connect the dots and do so without requiring excessive outside support.

2. How to Plan for AI Implementation

AI Implementation ‘building blocks’: Moving from AI pilots to full production requires targeted investments, as evidenced by the relatively small proportion of AI initiatives that are generally available compared to the number of AI pilots currently underway. Data recently highlighted by Benedict Evans underscores this phenomenon.

Source: Benedict Evans | AI Eats the World

An effective implementation framework considers multiple layers. Beyond the latest models, focus on your domain-specific context and data, ensuring it’s seamlessly available across applications.

While several AI stack diagrams or frameworks exist, the following highlights the areas of investment needed for most organizations to drive sustainable and well-governed solutions. Think of them as key building blocks for AI implementation.

Skift | Mayank Gandhi & Vivek Bhogaraju

Illustrates the ‘building blocks’ to make AI Initiatives successful, and can be adapted for GenAI, AI, and Agentic AI initiatives

Models: The bar to build impactful solutions has come down, as what once took heavy R&D investments can now often be done with far fewer resources by building upon what’s already available out of the box. Models are getting cheaper and easier to use. This is where most people start and are now able to deliver POCs/MVPs cheaply.

Having said that, out-of-the-box solutions only go so far. As models get better and newer, better SOTA models are released nearly every week. It is important to note that the benchmarks available from model vendors or AI labs are relevant as an indicator of progress, but they don’t necessarily translate into metrics that indicate tangible value for your use cases.

Building a heuristic or a set of evaluations of what works for your use cases early is key. This is an internal set of metrics that will guide your decision-making on which model to use or switch to, which approach to take, and for what specific use case.

Context and Data: While the models are getting better each week, your AI solutions still require high-quality contextual inputs. These can be in the form of your codebases or the organization’s knowledge management, or proprietary data. We have found that while the right model and tools are key, that is mostly something you can buy; marrying them to your context in a manner that these models and tools can interface with is the difference between a good AI product and a great AI product.

Making this context seamlessly available to your applications and user tooling requires investment. While new standards are emerging, making this easier, it still requires enablement, guardrails, and governance across your ecosystem.

Orchestration: This is more than just traditional workflows. It is how you marry your prompts, context, and processes into a flow before you package it into an application. Orchestration is also the building block where teams have to be thoughtful about when and how ‘human-in-the-loop’ interactions happen.

Monitoring & Observability: Another thing to consider is monitoring and observability, which is often left as an afterthought. It is important to recognize that these models are non-deterministic, which means your pilot or testing results may not always behave the same way in the real world, let alone if the underlying assumptions, prompts, or context change. Understand how the solution is operating, where it is not going as per plan, and then have an iterative approach to improvements.

3. How to Enable AI Use for Everyone in Your Organization

GenAI is the leveler of where innovation comes from: Role of an enablement function: Reflecting on our own experiences, we realize that it is crucial to create a nerve center that takes ownership and drives AI initiatives. However, those team(s) should be more of an enablement function, rather than a center of innovation. GenAI’s opportunity is that it is a great leveler. You want innovation to come from all parts of the organization. The idea is that anyone with the right tools, context, and ecosystem can start building. The role of the enablement team, therefore, is to lay the groundwork, set the guardrails, and get everyone started.

It also has to be targeted at varying degrees of skills across your organization. Hence, plan to skill up everyone. Invest in innovation challenges, hackathons, whatever approach works for your organization. Get them to the point of comfort level where they start asking questions about what is possible with this technology, and assess if they can do more with it. But you have to build up to that, invest in skilling up, ecosystem build out, demystify the variety of options, and unblock and make it easier to use this tech within your organizational guardrails.

Find the experimenters and doers: These are the “doers” in your org who are already experimenting and will set the tone. Bring them in to advise or better co-create standards, identify best tools, and build targeted accelerators. These folks will also serve as evangelists across the organization. The core team they work with can take many forms: central, federated across the organization, and a co-creation model where the first and last mile work is done in partnership with a central enablement team.

4. How to Manage ROI Expectations

Establish the ROI baseline upfront, and start from the familiar: ROI is important. Establishing a clear ROI baseline upfront – before building anything – is key. It gives you something tangible to compare with when evaluating success. But while ROI is critical to establish, we need to ensure that we are not over-indexing on one metric. ROI is not just an upfront investment ask. It needs to be an ongoing conversation about how the AI initiatives are improving and evolving.

Don’t start from scratch in unfamiliar territory. For early initiatives or teams ramping up, it is better to start with areas where they are already experts in, i.e., already have KPIs, such as quality, speed, or tone, and then ask, “Can AI make this better?”. That makes it easier to measure impact and scale what works.

Don’t overlook all opportunities for ROI: Customer-facing initiatives usually get a larger share of the investment spotlight. But don’t overlook the internal operations. Look for applications that unlock and accelerate opportunities that would be hard to scale otherwise. These are your internal processes, team, or functional operations that could use improvements, automation, or complete transformation.

ROI as a portfolio of AI initiatives: Going from MVP to production means thinking through all the aspects of building a robust, well-governed, and cost-efficient product. Therefore, understand the total cost of ownership of that, including monitoring, governance, usability, and change management.

Teams need time to mature into that responsibility. That’s where the enablement team steps in to solve for last-mile capability. It helps to look at the ROI on AI investments as a portfolio of initiatives instead of standalone projects. This gives you the necessary flexibility to plan long-term and also acknowledge upfront that not all bets will materialize.

Ultimately, customers and internal teams don’t care if it’s powered by AI; they care if it’s better. So the ROI conversation has to stay focused on experience and value, not just the tech behind it.

5. Parting Thought: How Will AI Disrupt Travel?

Uber vs. Airbnb: Disruption takes many forms: Disruption is an ambiguous term. Is AI bringing some net new functionality to the market, or just changing the way we traditionally define market leadership? Is it radically changing the existing market, or is it creating an entirely new market for the existing product? Benedict Evans illustrates this difference in his work, AI Eats the World, specifically talking about the kinds of disruption here.

Borrowing from the above example, when Uber launched, it completely transformed the way we think about taxis, especially in big cities like New York. What was once a very steep entry barrier, the taxi medallion system, was disrupted in ways that we couldn’t have imagined pre-Uber. Anyone could now be a temporary taxi driver or a full-time Uber driver. That one app changed the way consumers thought about commuting.

Similarly, when Airbnb first launched, it was perceived as the death knell for traditional hotels. And then, when there was a regulatory backlash against short-term rentals in various cities, we wondered if this was the disruption to Airbnb. But as we all know, hotels continue to thrive, and Airbnb survived in parallel, finding new customers and creating a new niche segment.

Both these companies made us value the experiential over the practical. They made us move beyond the mundane expectations of utility and efficiency to seamless service delivery. So much so that they became an indelible part of the lingo. We don’t just take a taxi anymore, we Uber. We don’t just book an apartment or a vacation rental, we Airbnb it. But both these disruptions are not the same.

The questions we need to answer (for our respective businesses) in travel are: (1) How do we envision leaning into AI to disrupt the status quo? (2) How can AI disrupt our business? The answers to these questions will serve as the north star of what AI initiatives we greenlight and get behind.

Vivek Bhogaraju, an industry advisor for data & AI, and Mayank Gandhi, Head of Cloud, Data & AI Platforms at Wayfair, shared the lessons they’ve learned about shepherding AI initiatives across an organization during a discussion at last month’s Skift Data + AI Summit, including how to get started and how to think about implementation.