Using Data Science for Digital Transformation at Carnival Cruise Line

By Mark Fridson, Principal Data Scientist at Carnival Cruise Line

At most companies, a data scientist’s main objective is to determine how machine learning technologies can revolutionize business procedures. In his insightful talk in February 2018 at Data Science Salon Miami, Mark Fridson Principal Data Scientist at Carnival Cruise Line, explores how the right machine learning techniques can boost company revenues and improve customer experiences at the same time.

Hello Everyone! I am Mark Fridson and I am going to talk about how to leverage digital transformation with data science to add value to your business.  

One of the big things right now is in addition to the insights we can get from the data, there's more channels to get that data from. We have email in the form of digital, we have social media through Twitter, Facebook and Instagram and  overall, there's tons of this digital data in channels where we can engage with our customers. We in the marketing department must figure out what our budget is and use it to attract the most business.  This requires identifying the audience that we're trying to communicate with and finding where they interact. The older generations are more engaged with traditional mail and email, in other words the legacy technologies that have been around a while. Generation X and Millennials are most reachable through email and Facebook.  Through Facebook, we can understand more information about their demographics. Based on what they like on Facebook, we can see what their favorite movies, destinations and hobbies are. Then, using this information, we can target individuals who have a very similar profile to our ideal customer.  

The hard part with these processes is that companies like Facebook and Google have walls set up to make it very difficult to get all that  information.  You may actually have to get people's consent to share certain information to give you that greater detail. Then, some of the details you may just not be able to get.  

Moving onto the most recent generation, Generation Z, we see that these people are using primarily digital channels through social media. We can do machine learning with this data to understand things about what activities people are doing. We can even leverage relationships on social media, such as who you follow, to know what your interests are. If you use Twitter, for example, we can analyze the text in your tweets to build a profile of what we know about you.

All social media channels have different nuances that make them different so engaging through them in terms of a marketing campaign is different if you go through mail or if you put up a target advertisement on Twitter. Propensity modeling figures out the probability of someone buying and then determines the lifetime value of people who will buy. Based on certain evidence that the potential customer has in his or her profile, you can use Bayesian inference to figure out the likelihood that they will go to the Caribbean, for example. We factor in age, family size and income level to determine the likelihood of purchase. Then we work with our marketing budget to target the 10,000 people, for example, who are most likely to respond to an exclusive offer. These would be our most valuable new customers, but we have to ensure that we're not targeting our existing customers that would book a cruise anyway and therefore not be responsible for a spike in additional revenue. We need to understand our new customers within the context of our existing customer base. Who tends to generate the most revenue currently? The people you want to attract for future sales will have similar profiles.


Get insights like this in person at our next Data Science Salon: Applying AI & Machine Learning To Finance, Healthcare & Hospitality, September 10-11 in Miami.

Register here


Existing consumers might not be responsible for the biggest spikes in revenue, but they can still increase revenue in significant ways. You already have your existing consumers’ previous booking data, which means you may be able to get them to purchase additional items in the future. There may some items, such as onboard drinks or entertainment packages, that these customers have no record of buying in the past but could bring in additional revenue if they buy in the future.  Understanding how existing customers can spend more money on their experience in the future can maximize the effectiveness of a marketing campaign. 

In terms of content personalization, we strategize about this a lot internally. Any business selling experiences must be careful of overwhelming customers. If you get inundated with an email every day about a promotion, you're going to stop looking at it. The subject matter and content has to push customers further down the pipeline. This may involve getting deeper into personal details about customer preferences. For example, do they do a lot of big group trips where they're booking six cabins? Do they book an interior room every time? Maybe we could recommend the exact room they always booked on a certain sailing and that would get them to consider it more heavily. I can just picture, similar to Amazon’s one-click buying, one-click cruising through which you could click on a proposed cruise deal with the exact room that you want. That would be an impactful result of understanding our customers’ behavior in this much detail. 

To further expand on this idea, Netflix does something where they're not just recommending things exactly like what you want. They want to find people who are similar to you who might watch something you don't watch. It may be something that has nothing to do with what you watch, but it expands your consumption of the product and then you're more entrenched in the service. If we can get someone to buy a drink package who has never bought a drink package before, maybe in the future every time they cruise they'll buy a drink package. What additional items can we get them to purchase that they wouldn’t have without the help of the algorithm?  

Another big thing right now is social relationship mapping. All technology companies are trying to figure out if there’s a pattern between what you do and what your acquaintances do. Maybe there is a missing piece that can be filled. For example, you know people who follow sports and you follow sports as well, but there may be a channel you haven’t yet seen that we can float to you on social media. The logic is that if Bill visited the Eiffel Tower in Paris, he is close to the Louvre, a museum in Paris. Maybe we should recommend the Louvre to Bill because  Bob, who is a friend of Alice, who is connected to Bill on social media, went to the Louvre and loved seeing the Mona Lisa. So if Bill goes to the Louvre after our recommendation, we should also recommend he sees the Mona Lisa while there. That’s a good example of how to extrapolate a relationship and make a recommendation based on patterns that we're seeing in a network. 

So the next big challenge is text analysis. Machines are trained in a very specific domain and as soon as you change one variable, the model can get completely thrown off. If you think about words like “bad”, for example,  they can mean so many different things across industries. Trying to figure out the context is difficult, but necessary when you're trying to figure out sentiment and the domain that you're looking at.  Through people’s sentiments, you’re trying to put all this together to figure out what are people excited about. What excites people the most about cruising to the Bahamas and how can text analysis expose it? And moreover, what are the trends across the social media posts that are mentioning Carnival? Even this generic information could be helpful in determining what promotions to offer.  

The hard part of analyzing people’s content on social media is that you often don't know what they're talking about all the time.  They could be talking about Game of Thrones or they could be  talking about a cruise they went on. You have to train each of those domains in machine learning to be independent of one another because they have different topics that mean different things.  You see world-class companies struggling with this all the time so the average company is going to have even more trouble even adopting this. It takes a lot of nuance with your industry and with your use cases to ensure that the information you're getting is valuable and accurate. On one hand, you can increase engagement with added information. On the other hand, you could ruin the relationship with your customers. You have to make sure that you're not risking your business on a technology you may not understand.

A lot of these technologies get people to believe that they are way ahead of where they actually are. At the same time, people have very high expectations of they want a machine to replicate in terms of what a human can do. Most of the time, a machine cannot do it. Can you get to a place where you have an acceptable way of using it? Then maybe it’s worth exploring. 

No one has a magical solution that's ahead of the times, so I think you really have to just understand where the technology is, where people's knowledge of it is and what is realistic with what you can work with today. 


Curious for more?

Don’t miss the next Data Science Salon in Miami, September 10-11, 2019.

Register here


Get the latest data science news and resources every Friday right to your inbox!

Sign up for our newsletter