Predict Your Customer’s Next Action – Propensity Modelling in Travel

Get in touch

Do you know what, when, and why your passengers are going to buy? Knowing how likely a passenger is going to act, and under which circumstances, is the cornerstone of propensity modelling. Using propensity models enables travel retailers to focus their efforts on the passengers where they generate a meaningful change in behaviour.

Propensity Definition

noun
an inclination or natural tendency to behave in a particular way.
synonyms: tendency, inclination, predisposition, proneness, proclivity, readiness, susceptibility, disposition;

A propensity is a natural tendency to behave in a certain way. We all have propensities — things we tend to do. Even animals have a propensities – dogs have a propensity to bark!

What is … a propensity model?

A propensity model is a statistical scorecard that is used to predict the behaviour of your customer base. Propensity is calculated by the application of mathematical models to data in order to try and predict whether someone will take a particular action, such as identifying those most likely to respond to an offer, or to focus retention activity on those most likely to churn.

Propensity models use sophisticated machine learning algorithms to predict what a customer is likely to do next by exploiting patterns in human behaviour. Propensity modelling dates back to 1983, but it’s only in the last few years that machine learning has unlocked its potential. In fact, travel retailers with a good data science team and access to the right tools can create comprehensive models. With enough data, you can develop highly accurate propensity scores. Armed with those scores, it’s possible to not only understand the probability that an individual customer will transact, but also estimate what you expect the value of that customer to be.

For example, propensity predictions support very sophisticated digital marketing campaigns that target a “segment of one” with highly customised offers and super-personalised travel experiences, to optimise conversion rates, online media spend and revenue per visit to your e-commerce site.

Propensity models are the key to unlocking the value of Big Data in travel.

Example: Travel Insurance

A traveller could have a high propensity to purchase travel insurance if they are naturally risk averse and purchased insurance in the past. In many cases, a customer may not even have consciously registered that they have a propensity to purchase something, but the machine learning algorithm can predict it.

Propensity is not a new idea, but it is only with the availability of large amounts of high-quality data, cloud infrastructure and machine learning that the models are starting to generate predictive signals with reliable accuracy. In particular, some of the more sophisticated retail banks and insurance companies have used this approach with great effect over the past few years. A combination of demographics and personal financial circumstances can play important roles in determining the propensity to purchase a personal loan, mortgage, investment or insurance product.

These same principles apply to travel: propensity in travel is revealed by a combination of a consumer’s demographics, their personality, their transactional history and attitudes. The best marketing and product teams in travel retailing have realised that they must embed these insights and combine these with a Big Data perspective to compete effectively in the digital world.

What do predictive propensity models look like?

Let’s start with the data sources. There are 4 main sources or dimensions to this type of propensity model, all of which relate to “offline” activity – in contrast to the online activity of a consumer that can be tracked by Google Analytics or similar platforms with tag management functionality. These dimensions are:

  • Demographic information that tells us “who” this person is based on gender, age etc.;
  • Transactional information that tells us “what” a person has purchased in the past as well as their estimated purchase capacity;
  • Psychographic information that tells us something about “why” a person purchases things in terms of attitude and opinions. Facebook “likes” are a good example of this;
  • Personality information that also tells us something about “why” a person purchases things in terms of their personality biases. This data is usually collected via surveys and can be difficult to acquire.

The data that we use across this universe varies depending on what propensity we want to predict. Generally, transactional data serves as the foundation to everything else in a propensity model. Transactional data is “hard” evidential data that tells us what buying patterns and preferences a consumer has had over time so serves to validate everything else.

Another way of looking at this or another ‘slice through this data’ is 1st party data v 3rd party data:

  • 1st Party Data: Information generated from your e-commerce website, social platform and mobile web or apps about your customers. It typically consists of your customer’s personal information (name, email, addresses, phone number), demographic information (gender, age) and limited behavioural data (site interaction, purchase history, interests). It is typically stored in a CRM or web analytics system, and you, as the owner of this data have all of this free – but it can be hard to combine it all.
  • 3rd Party Data: Information generated from internet interactions and other websites. This data is used to give you deeper insight into your audiences, such as individual demographics (income level, marital status) and household attributes (number of children). It can be used to build consumer segments for more targeted advertising. It’s collected and licensed by third-party providers that have no direct relationship with your customers, and you have to purchase this data to access it.

Until you combine all of this data, you cannot truly claim a customer 360 view in travel.

Key Sources of Data for a Propensity Model.

Demographic, psychographic, transaction and personality data are the sources to ‘train’ a propensity model for travel. This training process is an iterative process that uses machine learning to “parameterise” a model. These models vary depending on the exact type of propensity model being built, but typically they use logistic regression or k-means clustering for the simpler models, all the way up to support vector machines and neural networks for the more sophisticated models. Yes, advanced Big Data in travel uses some very high-end statistical models!

So what do we do with the propensity models?

There are in fact many ways that propensity models can be deployed and monetised by an e-commerce business. Any data that tells you which segments, which individuals even, are most likely to purchase your products creates a scenario where highly targeted, precision marketing can become a reality. And with this, conversion rates and revenue will improve and marketing costs will fall.

Creating effective propensity models

For a propensity model to be truly effective, it has to be dynamic, productionised and scaleable. Let’s take a closer look at each of these:

  • Dynamic: A good propensity model changes over time as newer data is put into the model, so that the model itself can become smarter, more accurate, and evolve with underlying trends in the data. As a result, travel retailers should have a pipeline of data to retrain your propensity model on a regular basis.
  • Productionised: A good propensity model must have a proper standard and framework for regular data ingestion, validation, and deployment. This means the model can deliver predictions that are understandable and actionable, and enables measurement and evaluation of the model performance.
  • Scalable. Many propensity models are built as a one-off and then abandoned. This is a waste of resources. Good propensity models have to be capable of producing large numbers of predictions and they need to be able to be easily adapted across other scenarios.

The best models propensity model show you the return on investment from getting your customer to take the action you want and maps this to the cost of implementing this campaign. Using propensity models will help you continually optimise your marketing funnel driving efficiency and effectiveness on an ongoing basis.

Summary

Most travel retailers create a large volume of data and have access to a lot of customer information, but they don’t really know how to leverage it to make good strategic decisions. Without this foundation, adding big data into the mix often adds little value because it lacks actionable information to make effective.

We are seeing amazing advancements in the way we can analyse data: there are now have algorithms that can understand and translate spoken words, and analyse them for content, meaning and sentiment. There are algorithms can now look at photos, identify who is in them and then search the Internet for other pictures of that person. There are even advanced cognitive algorithms that emulate the learning algorithms of the natural world.

Today, travel businesses haven’t really arrived at the day when Big Data can reliably tell them why customers behave in a certain way. Propensity models are just one example. At OpenJaw, we believe that part of the ‘Big Data’ journey is to know and understand propensity modelling, as this is the key to unlocking the value of Big Data in travel.

big-data-for-airline-growth-ebook-upgrade

Download the guide to using Big Data for airline growth.