Embracing AI in Revenue Management: Overcoming human bias for better results

Every inch of my German heart urged me to keep control, trusting in my superior driving skills and lifelong experience. But another voice reminded me of the long, monotonous drive through the Netherlands at a frustratingly low speed limit ahead of me.

So I did it. And it worked. And though I cannot believe I am putting this in writing, it was probably better than me. While cruising through the Dutch countryside I could not help but wonder: Why aren’t more Revenue Managers taking this leap? What is holding them back from running on autopilot?

Let’s face it. Letting go of a habit is hard. It gets even harder when it involves deeply anchored skills you have carefully developed to serve this very purpose. But there is more to it.

Eventually I arrived at my destination and had the opportunity for a slightly nerdy read that I warmly recommend: "Noise: A Flaw in Human Judgement" by Daniel Kahneman, Oliver Sibony, and Cass R. Sunstein (2021). In it, the authors lay out how random circumstances can influence human judgement. One specific section caught my attention. It investigates the foundational question of whether models are good enough to outdo human predictions.

The answer is a resounding YES – clearer than I ever expected. In fact, studies that date back to the 60s and 70s already revealed that “any linear model, when applied consistently, was likely to outdo human judgement in predicting an outcome from the same information.”

The authors even note that “it proved almost impossible in that study to generate a simple model that did worse than the experts” (Kahneman et al. p.121).

How is this even possible?

Human judgement is noisy

As a seasoned Revenue Manager: Think about how you make a pricing decision for any given day. Most likely you will apply a well-vetted mental formula based on factors like pickup, competitor pricing, or forecasted occupancy. Each component holds a certain weight in your decision-making process.

Now ask yourself: How often do you break your own formula? Would a very high pickup render the other facts irrelevant? Would a specific competitor’s behavior tempt you to follow suit, despite your own OTB situation?

I have yet to meet the RM who did not, at least once in their career, panic in the face of slow pick-up. The fact is, we all fail to follow our own formula and don't even notice as we successfully justify to ourselves the reason behind each of those decisions.

The authors note that the "common intuition that the same [factor] can be inconsequential in one context and critical in another" (Kahneman et al., 2021, p. 121) is a classic example of noise in judgement.

This phenomenon causes us to prioritize a strong pickup factor over our compset factor and vice versa based on circumstances or, believe it or not, our current mood.

Yes, even factors like the weather, the daily news, the office temperature, or when we had our last meal affect our decision-making in the moment. These seemingly unrelated factors might make the difference between feeling brave and holding course with our rate, or taking the safe route and staying close to our competitive set.

So why don’t we fully trust systems in Revenue Management?

The reasons behind the mistrust, as outlined in Kahneman's book book, are as intriguing as they are humbling. Hands down, I realized I had subconsciously encountered many of the common pitfalls myself. Interestingly, overall scepticism of algorithms, fear of unemployment, or general dislike for computers proved to be only the tip of the iceberg. Here are the three reasons I found most relevant to the world of revenue management:

Deeply anchored belief in the value of complexity

“The combination of two or more predictors is barely more predictive than the rest of them on its own” (Kahneman et al., 2021, p. 127). Let that sink in for a minute.

In Revenue Management we love to believe that the more you know, the more the system can see the better and trustworthy the prediction. But as the system can never know everything the RM knows, the system will never be good enough. That is a misconception.

The key is for systems to use the right and fresh data to run stable models suitable for the use case. Anything else might be a distraction or a redundancy (Kahneman et al., 2021, p. 122).

Expecting perfection

I think we can all agree that we're not perfect. Human beings make mistakes and we are quick to analyse, justify and accept them as a natural byproduct of decision making. But, we do not cut systems the same slack.

We expect algorithms to run with close to 100% accuracy at all times. In the world of pricing tools, it is not uncommon for a customer to completely give up on a product after disagreeing with the suggested price for a single sellout night. The system might have done an excellent job for all other 364 days, but one breakdown in logic can trigger mistrust and question the entire algorithm and overall use.

What’s intriguing is that the judgement of whether or not the system is right is mostly made against what we would have done ourselves - our own intuition. This brings me to the next, possibly most profound, reason for lack of wider adoption.

The sheer power of gut feeling

Don’t we love to be the ones having made the right decision? It proves our value, our expertise and much more physically, it triggers an emotional reward in our brains. Yes, making decisions based on gut feeling can be highly addictive.

Don’t we all take some joy in micro-managing our RMS and especially “catching it” when it is about to make a mistake (in our opinion)? This exercise is not necessarily a reflection of inherent distrust in the system but a way to amplify our internal reward signal

When assessing whether we should fully trust a system, the underlying question is not really how much better it is, but whether it is good enough to justify giving up on practising our intuition. If a system doesn’t prove to be flawless and 100% in line with our own gut feeling, odds are we won’t fully adopt it. Ever. (Kahneman et al., 2021, p. 145).

I can already hear the readers thinking “Yes, but…” and coming up with the reasons why their use case is vastly different. And maybe it is. But odds are it is not. This raises the question:

What can we do to overcome these obstacles to system adoption?

What can we do to overcome these obstacles to system adoption?

Recognize the noise in our own decision making

Step one is awareness. We, as human beings, are not immune to noise and our confidence in decision making is nothing but an illusion anchored in years of successfully justifying past decisions.

Systems are not swayed by external circumstances. They don’t change the rules based on their mood of the day, and they surely don’t get influenced by others. Next time you catch yourself monitoring the day-by-day output of your system, ask yourself”

- “Why am I doing this? Am I adding enough value or am I just chasing an intrinsic reward?”

-"Am I adding enough value, or am I just chasing an intrinsic reward?"

Select the right system

As mentioned, studies have shown that pretty much any algorithm will outperform you (sorry to have to rub it in again). However, I do not necessarily suggest installing the next best system in your hotel, just like I would not suggest delegating absolutely every decision to AI.

A system needs to earn your trust by proving it bases itself on accurate and relevant data, running at the desired frequency and optimising the different offers you find strategically relevant for your hotel. A strong vendor can supply you with ROI studies, explain the reasoning of the algorithms so you are not left to, and disclose which factors the system takes into consideration.

In a best case scenario, these proof points are even inbuilt in the system to offer a next level of transparency.

Rationalise perfection

My RMS made a mistake. We have all felt that way or worse, feel that way on a regular basis. The truth is that your RMS will be wrong, just not as often as you- but that is hard to measure. As we are a very biased judge of our own decisions we tend to hold the RMS to a higher standard than we hold ourselves.

To avoid these cases, it is recommended to establish an overall measure for success, consider the following:

1. Does the system increase RevPAR at my hotel?

2. Does it drive occupancy or rate, depending on the strategy I have entrusted it with?

3. Does it update rates more often than I ever could?

If the answer to all these questions is yes, potentially, I won’t need to slap it on the wrist every time it is $2 off from what I consider optimal.

Apply your mental bandwidth to more impactful areas of your job

In my role at Lighthouse, I happen to have interviewed a vast variety of revenue managers about their daily, weekly, and monthly tasks. Interestingly enough, when looking at the big picture, day-by-day pricing is a rather small piece of the puzzle.

Daily price tweaking is a tactical and repetitive task with rather small margins for optimization, especially compared to other areas of the business where the full brain power of an experienced revenue manager can really make a difference.

Consider these strategic questions instead:

1. Are all outlets in your hotel driving desired revenue?

2. How can you monetize your dead lobby space? Is the room type structure still making sense?

3. Is the room type structure still making sense?

4. What upsell opportunities can drive additional spending?

In the end, your time is money. Where you choose to allocate it is your investment portfolio, and your goal is to generate the greatest return on each and every investment. Most likely, daily price tweaking is not that return you are looking for.

To be clear, the purpose of this article isn’t for you to close your laptops and blindly trust the next best system. I believe we owe it to ourselves and our productivity to be aware of our own biases, the inevitable noise in our judgments, and some hard facts, to perhaps slowly and occasionally ease off the gas and observe the magic that we did not deem ever possible.