AI That Works for Hotel Leaders Is Finally Here: The Story Behind How Actabl Built Altitude to Turn Data Into Answers You Can Trust - Stephen German, Actabl [Sponsor Bonus]

Josiah: I feel like heading into HITEC, there's going to be a lot of conversation about a lot of uses of AI, and I want to get some more of your thoughts around what it takes to trust what AI is telling you. And I know you think a lot about this. I know data is a piece of it, but it does feel like right now everybody is seeing a lot of AI outputs, and if you're a hotel leader, you need to have confidence in those outputs. And so what does that take?

Stephen: It takes a lot of training and it takes a lot of fine-tuning when it comes down to being able to ask the questions. A lot of it is probabilistic, meaning when you are just pointing at a data set, unless you are very specific about exactly what you are trying to get, there's going to be some hallucination in trying to get the actual data that you want returned from it.

Stephen: And you can be excruciatingly detailed with it, you can build skills and reusable prompts on it, but it takes a lot of trial and error and a deep understanding of the data model and how it works to be able to source it and actually get the data that you need. When folks ask me, "Well, can we just tie an LLM into the data, and can we get the same answers from it?"

Stephen: It's like, well, you may be able to get the right answer today. The next time you ask the question, it may be different. So unless there's some logic behind it that is routing you to this is how the data works, this is how it interacts, these are the differences, and these are the discrepancies, things like if you're just asking about forecast, does ChatGPT or Anthropic know that you mean your primary forecast?

Stephen: Does it mean your locked forecast? Which one? Which timeframe? It's probabilistic. And what we're working on is we want to be able to give our leaders and our associates the ability to ask a question and confidently get the right answer regardless of how they phrase it. So that logic and intelligence that we put in that routes them to the right data, to the right interactions, that gets you an output that you can cite.

Stephen: One of the tests that I like to do is, would you trust this thing or this output without reviewing it? Is it something that you could just hand to your CFO and confidently say, "I know all of this is right"? And when it comes to using external providers, that isn't always the case.

Josiah: Well, it's funny you say that, Stephen, because I just came from a couple industry conferences, and it's interesting now versus even a couple quarters ago, you have a lot of hotel leaders saying, "Hey, we have Copilot. We have one of these ChatGPT or Anthropic processing our data." And it's interesting to hear them talk about this, and then I have follow-up conversations about how that's going.

Josiah: And so there's kind of the first pass it does, and then I'm hearing from a lot of them a lot of what you've described. It's a lot of that, "But now we're going back and checking it," and it's kind of unclear, are you even saving any time? Is the work product any better? And so that's where what you're saying is really important, right?

Josiah: Because it requires a smart approach that's not just run all this through your LLM. You need to have the right underlying data, but also the right guardrails in place so you can trust the data.

Stephen: And especially when it comes to numbers, when it comes to those actual facts, you need a deterministic result.

Stephen: You need the facts that are brought together because when you're trying to move at speed, if you're constantly questioning what's behind it, then you're not making a lot of that progress.

Josiah: Interesting. I want to talk about the data for a moment. Underlying data, we talked before, I'll link in the show notes, about the normalization process of that.

Josiah: Why is it important to be running this on data that is normalized and that is your own data? Because it feels like there's more and more connectors, which is interesting. Is there an advantage to having AI run on your direct data sets as opposed to integrations with other providers?

Stephen: 100%. It comes down again to are you comparing apples to apples?

Stephen: Just to use an example, if we take GSS, the types of questions you're asking of the guests and the actual responses, how you're scoring those responses, if you're using different providers for that, you need to be able to compare apples to apples. Otherwise, if you're asking for your portfolio, how am I trending on my intent to recommends, you may get some hallucination if you're trying to combine things that haven't been normalized.

Stephen: Say one's out of a one to 100 scale, one's out of a one to five scale, how does that actually translate into getting you the answer that you're really looking for? And that's a simplistic example. When you get into POS, when you get into accounting systems, when you get into all your different revenue and marketing systems that all have different sources, different hotels, different regions, they're all combined slightly differently, so you need to have a system that is normalizing it, that is saying, "This is how everything translates," so that when you're engaging with it, you're confident that that is giving you the true answer behind it.

Josiah: Amazing. Now, some people listening to us might be leaders, and they have other folks that are doing the technical build, and I wonder if we can give them a way of thinking about what questions to ask their teams to ensure that they have AI powered reporting that they can rely on. So I think the underlying data seems like that's kind of something to dig into.

Josiah: Is there anything around verifying the sources or the traceability of a report? I wonder for an executive that's listening, how could they talk with their technical teams about competence levels in this?

Stephen: Yeah. Timing is the next really big one outside of normalization, or as a part of normalization, because if you're getting some data daily, some data weekly, some data monthly, and you're trying to compare it, then the timing of that matters in the answers that you're getting back.

Stephen: Because you may get incomplete or more complete from some areas, and then you're not actually getting a real picture of it, and that's again where the normalization is so important, and then the sourcing of that as well. You can be a lot more confident in data that is provided through integrations that's coming directly from the systems.

Stephen: You know that there's no doctoring of it or that there's no, even just from mistakes, things that have been changed or adjusted that now aren't making their way into the data so that you're acting on incomplete information as the next really big one. So that's part of why we spent so much time building up our integrations and our normalization so that it can be compared apples to apples.

Stephen: It can run on similar frequencies and can flag where data is incomplete, where data is missing, to confidently know the state of the world.

Josiah: It's interesting. It's a really big point, and I think it's not talked about as much as it should be because I think there's a lot of versioning where it's like I do see a lot of people downloading some report or some spreadsheet somewhere, and then it lives somewhere, and maybe they're running some analysis.

Josiah: Then they send the analysis to someone, and it's like, where did this come from? And then there's some fresh data, and they're making an important decision, and the world has changed. And so it's really important that people think about this. I think you have to think about it from what business decisions need to be made and then what are all the different component parts, right, to help me make that decision.

Josiah: I want to change tracks a little bit and talk about Altitude. This is something that we're excited to announce to the market. I wonder if you could explain a little bit about what the problem was. I've enjoyed talking with you over the years around how you think about product development, starting with a problem or an observation, and I wonder if we could begin there.

Josiah: What were you seeing and hearing? And then I want to talk through a little bit about what this is and what it provides hoteliers.

Stephen: Yeah. It comes down to signals and kind of what's your journey as you're making decisions. There's some things that are milestone based, like you're approaching your budget, so you need to have certain things ready by a certain time.

Stephen: There's some things that are reactive, and so you need to know what is the current state and be able to dig into it. And then there's the proactive as well, where you're being presented with a flag, but you then need to dig into it because you need to know what is the story behind those signals.

Stephen: And so what we're building is really the ability to hit it from any angle. The next step is drilling in, is understanding it more deeply and understanding what is behind it and being able to chase that down in real time.

Stephen: I was talking to a management company recently on how they investigate problems, how they dig into the data, and they said, "Well, the main problem is we see a hazard, we see something that's trending the wrong way. But to get into it, we have to pull data from a couple different teams, from a couple different systems.

Stephen: We have to go ask them to provide the reports. If they don't have the report ready, if they have to make a report, it takes a few days, and by that time you're kind of losing the thread." People, when they have the inspiration and when they have that detective mindset, when they're in that flow state, you don't want to interrupt it having to trace down the manual because that's where you lose the plot.

Stephen: So what we built within Altitude is the ability to have that conversation with the data, to trace it down, to understand the metric, to ask it why, and to get deeper and deeper with it without that overhead, and most importantly, without needing to go to multiple systems for it. Because when you need to do that, you're again pulling yourself out of that flow state.

Stephen: You're potentially finding different rabbit trails and things that you're needing to suss out, and we want to give associates and leaders the ability to follow that thread at the speed of thought, at the speed of their experience, and remove that friction, remove those rocks from the shoe of actually finding the answer and being able to take action accordingly.

Josiah: That's interesting. It's almost like having this trusted, always-on assistant that you can rely on, which is maybe shocking. But I was talking to a CEO a couple weeks ago and they were talking about having some of these questions as they're going around traveling. And obviously there's different working hours for different people and even he had a whole team of analysts and it was still sort of like, it takes forever to get answers.

Josiah: And so this seems like a very different way of leading and being able to understand what's going on in a hotel business. It's sort of like this always on analyst that you can interact with. Is that sort of a way to think about this?

Stephen: Yeah. No, it's a great way to think about it and a great way to think about some AI in general.

Stephen: And I think what the advantage of this is, is the specificity of the prompts and the context and what needs to be provided, and the reusability of it, because the best organizations are the ones that have that more templatized and are then getting similar results and similar insights operating off of the same framework.

Josiah: Interesting. I'm curious if you feel this way, but it's critically important to empower the people providing hospitality on the front lines, right, on property. But I also, I've spent most of my career in hotel tech, and I do feel that an above property leadership view is almost a role that's underserved by technology and reporting, like a lot of tools will have different reports, but it doesn't really feel aligned to how an above property leader, let's say like a regional VP of ops or a chief operating officer does business, in my experience.

Josiah: I don't know if that has come up in some of your conversations, but it feels like this capability is more aligned with how one of those leaders tends to work and what they need.

Stephen: Yeah, absolutely. And AI is collapsing a lot of roles and a lot of responsibilities, because it has the capability of doing that.

Stephen: And so it lets you operate within the cracks of the meetings and everything else that is monopolizing your time and your calendar to be able to move through things quickly without that overhead of needing to wait for different teams to pull those pieces together. So it's very much an enablement of above property because it serves their calendar, and it serves how they operate rather than how an analyst is able to operate, where they have multiple hours of deep thinking time to be able to make these happen, which we're trying to allow above property to make the most of their time because they are some of the most busy people in the organization between meetings and travel and everything else that they're juggling.

Josiah: It seems to me there might be something sort of counterintuitive here around providing these capabilities to an above property leader. You might start to see the benefits of technology throughout the organization, not just because of the role and influence, but I think the promise of tech is always that it makes your business better, makes your life better.

Josiah: But that isn't always the case. It's really about adoption, kind of how it gets used by people. And it kind of feels that as you have leaders using this, then it's going to kind of trickle down into every element of the organization, right? Because then you have to have people digitizing their work streams, that you're collecting the data, and then everybody is operating off this sort of AI tech-powered operating system.

Josiah: And so I feel like this way of operating and this new capability might finally start to deliver some of the results that I think people wanted from hotel tech for a while.

Stephen: I think so, and I think we're going to be able to spend a lot more of our time in the actual action and the actual impactful work.

Stephen: There's the concept of the data pyramid, where at the foundation is data, and then data put together becomes information. Information with experience becomes knowledge. Knowledge with insight becomes wisdom, and that's where action is taken, between that knowledge and wisdom step. And a lot of us, both in the hotel tech space and in the actual properties and within leadership, we're spending a lot of our time in the data and in the information.

Stephen: And so this is going to enable teams to actually spend a lot more time in the knowledge and in the wisdom components that will actually move their organizations forward. Because synthesizing data, it's important, it's critical, it's foundational, but it's ultimately low value compared to the action, and we want to enable teams to actually spend more of that time in the action and in the iteration, and less time in the synthesis and in the finding out.