However, research has shown that most technology transformations fail. If you want to set yourself up for success, it’s crucial to clearly understand this new generation of AI before you decide to invest. As with any other large-scale technology implementation, it will be easy to spend a small fortune and not see value in return.

In this article, we’ll explain why the current generation of AI is such a big deal, provide a framework for determining whether AI is the right solution, give some examples of good (and not-so-good) uses, and provide some things to think about as you get started.

Why Is AI Suddenly Such a Big Deal?

A brief history lesson can help explain why everyone is so excited about AI. The artificial intelligence field was founded as an academic discipline in 1956. In the early years, AI research was largely theoretical. Researchers didn’t have the data access and computing power required to actually run the algorithms. As a result of the big data evolution of the 2000s and 2010s, data storage capacity and computing power finally reached a level where researchers could apply AI in the real world, accelerating its visibility and application.

AI 1.0, the models that came about in the big data evolution, are narrow in scope. They’re optimized for a single task, like identifying images or producing product recommendations. They require significant human intervention to train them over many iterations of “making mistakes” before they can solve narrow problems without supervision.

The recent generation, AI 2.0 is a new paradigm. Single models can do many broad tasks without much human intervention. This makes them faster to deploy and more flexible to use. They can also create net-new material in a way that previous models just couldn’t. Perhaps both most exciting (and most frightening), we’re only in the infancy of this type of AI models.

Practically, AI 2.0 is getting better at the following human behaviors:

• Sensing: Perceiving and gathering information from many sources
• Understanding: Interpreting and contextualizing information
• Creating: Developing new content, ideas, or solutions (text, imagery or code)

While this new generation of AI models is extremely powerful and extremely useful, it has limitations. The models can only learn by scanning, processing, and replicating existing data. Any incorrect information, biases, or gaps in the data will be reflected in the output. This means users must be even more confident in the quality of their data, careful with what they give the models, and skeptical of outputs.

While they seem quite human-like, these models lack a crucially important piece of human intelligence, critical reasoning – or basic common sense. This means they don’t speak business, in the sense that they don’t truly understand what you’re trying to accomplish, or why. This is why we need humans both to ask the right questions (prompts), and vet the outputs in the context of the business strategy.

AI’s Huge Potential

With the expansion of potential use cases, and potential users, the value that AI-driven solutions can add has increased exponentially. Since these models can learn on their own, requiring less hands-on human development, they’re smarter and faster than AI 1.0. And since they’re better at mimicking a wider variety of human behaviors, they’ll be able to replace humans in a wider variety of manual, repetitive tasks – even ones that require some creativity. This will allow humans to shift their focus to higher-value activities that are presumably also more interesting and engaging. This will not only make people more efficient and productive, it should also dramatically improve employee satisfaction and engagement.

When Is AI 2.0 the Solution?

Smarter, faster, and more empowering – who wouldn’t want some of that! Unfortunately, it still isn’t quite that easy. AI 2.0 is best suited for problems with the following characteristics:

High ROI:

While they are getting cheaper, AI solutions still require investment in development and ongoing maintenance. The best problems for them to solve are well-defined enough that the solution will have a high impact on the business. For example, replacing labor hours in a large manual process has an easily definable ROI.

Large scale:

AI 2.0 is best applied when there’s a relatively large volume of tasks to be accomplished – think thousands of call-center requests for information or content creation for dozens of campaigns per week.

Unstructured data:

This includes anything that doesn’t fit neatly into the rows and columns of a table, like text or images. It’s difficult for both humans and traditional modeling to identify the patterns that lead to insights in this data, but this is where AI 2.0 excels.

High novelty:

The inputs to and outcomes of the process should be new and different with every itera- tion. For example, agents should offer clients packages according to their expressed preferences, not just sell the same offer over and over again. Images fed into the model will contain similar objects (like, say, king-size beds) but look slightly different every time.

Human oversight:

These models can still get it wrong. It could be the result of faulty data or a lack of ability to reason. That’s why human involvement is critical. So look for processes where it’s easy to loop humans in. For example, it’s better to surface AI-driven recommen- dations to an agent in the call center, who can interpret them in the context of the request, as opposed to communicating them directly to the guest on a website.

How Does AI Work (Or Not)?

A good example of a use case for AI 2.0 in hospitality is responding to requests for proposals (RFPs). Hospitality companies typically get hundreds – if not thousands – of RFPs for group business per year. Generating responses requires a lot of manual effort. Let’s see how this fits for an AI 2.0 solution:

High ROI? Yes. Crafting responses requires a lot of people hours. Faster responses increase the likelihood of being selected. So, there’s both a potential full-time employees (FTE) savings, as well as a revenue up-side.

Large scale? Maybe. Larger hospitality companies have higher volumes – and potentially more challenges in keeping up. But generally speaking, most companies can manage the volume of responses, if not at the fastest possible speed.

Unstructured data? Yes. Requests can come via email or other text form with very little pre-defined format.

High novelty? Yes. While there may be some standard questions, every request is slightly different and requires bespoke responses.

Human oversight? Yes. The existing sales team could review and edit responses before sending.

Pricing, a highly analytical process in hospitality, is an example of a bad use case for AI 2.0. In fact, some analytical solutions leverage AI 1.0 algorithms. On the surface, it does have many hallmarks of a good candidate for AI 2.0 (like data, scale, and high ROI). But it fails on two critical fit criteria:

Unstructured data:

Pricing data is typically highly structured, so AI 2.0 is an unnecessary additional investment.

Novelty:

Creativity is not an asset with pricing. The inputs are relatively fixed (booking data, competitor rates), and you need one best price recommendation, not a range of interesting options to consider.

How To Get Started

Like any business transformation, effectively incorporating AI 2.0 requires active coordination across a variety of internal capabilities, including:

AI Strategy:

Before jumping into the first shiny use case, define a roadmap that clarifies overarching roles in the transformation, as well as guiding principals. For example, will you self-host or use a managed service? Then, within that roadmap, identify a few quick-win use cases to demonstrate value and identify any internal hurdles.

Data and Technology:

AI is a data intensive application, so it’s crucial to understand your organization’s data landscape and make plans to acquire and consolidate the data you’ll need, particularly for the quick-win opportunities. You’ll also need to put some focus on data security, in particularly insuring that no confidential internal data is shared with a public large language model. Finally, business and IT should form a joint committee to support AI solution vetting and acquisition.

People and Process:

As you bring AI into the organization, you’ll likely need to add a few roles that are responsible both for the solution success, and the organizational transformation that will be required to fully realize value. For example, given that the first few implementations will be quite experimental, you’ll need to think about how teams are being incentivized. You certainly don’t want to penalize people for trying new things!

While AI 2.0 represents a tremendous opportunity for innovation and value creation, bringing this capability into your organization will require the same disciplined approach as any technology implementation. Make sure you select the right business problems to address, and you have the data, technology, and organizational infrastructure in place to ensure long-term success.