with Natalie Osborn, Senior Industry Consultant, Hospitality and Gaming Practice, SAS
We’ve taught analytics 101 through the last couple of blog posts, and now that you have passed that course, you are ready to take an advanced course in analytics. Ok, not really, we won’t subject you to that, but it is the appropriate time for us to introduce some more forward looking concepts – particularly because as the volume of conversation around analytics increases, so does the volume of conversation around these very complex topics. As always, our goal is to help you understand these opportunities at enough of a level to participate in a conversation about how they support your strategic initiatives.
The purpose of this blog is to cover three important “emerging” areas for the application of advanced analytics. These three areas, machine learning, digital intelligence and text analytics, have become topics of conversation mostly because big data analytics technology has evolved to the point that we are finally able to capture and store complex, non-traditional data formats and apply advanced analytic techniques to them at scale.
Machine learning started as a branch of artificial intelligence, but separated from that field in the 1990s, moving towards a discipline more focused on statistics and probability theory. Today, machine learning refers to any algorithm that is developed on a set of test data and then deployed on new data to perform the same task. The algorithm “learns” the new data and is able to automatically perform the same analysis on it, adjusting for new patterns as they evolve. This “learning” takes place with minimal intervention from analysts, meaning that you should be able to segment a large chunk of new guests without the analyst having to write a new model or reconfigure an old one. Machine learning is a term you have likely heard associated with a wide variety of problems, and probably pitched by some of the analytics vendors you have come in contact with.
Regression, clustering, decision trees, factor analysis, logistic regression and neural networks are all classifications of models that can be considered machine learning algorithms. The idea is that as the algorithms analyze new data observations over time, they become more accurate with future predictions.
Machine learning is highly related to data mining, and even uses many of the same algorithms. However, data mining is typically focused on exploration and discovery of previously unknown properties or relationships in the data. Machine learning, on the other hand, uses known properties or relationships in the data to predict properties or relationships for new data added to the data set. This distinction between exploration and prediction is crucial to understanding how and where to use machine learning algorithms. It could be said that data mining is used to develop the machine learning algorithms. Machine learning algorithms are deployed in production software environments, like responsive websites or marketing algorithms.
Machine learning is used broadly across many industries. Some common applications include search engines, recommender systems (like Amazon or Netflix), advertising, detecting credit card fraud and natural language processing. In hospitality, machine learning will most commonly be used in segmentation analysis and website analytics.
Text analytics have become more widespread since the advent of the social web. Text analytics can be used to mine the content of any unstructured text document, either created on external sites like Facebook, Twitter or TripAdvisor, or created internally like call logs or open ended questions on a guest survey. There are several methods available to quantify the contents of these unstructured text documents. These methods are based on natural language processing, a type of algorithm that understands language in context and can interpret or infer meaning from it. Natural language processing is most effective when it is applied natively, as opposed to on translated text, which is why it is important for a global industry like hospitality to work with software that has the largest available portfolio of languages in their text analytics.
- Content categorization. This identifies key topics and phrases in electronic text and sorts them into categories. It eliminates the manual work of reading and tagging documents, giving you much faster results. Text documents can be organized and tagged for search, making it easier to find, sort or process the content. This approach makes it easier to assign certain issues to specific departments that can resolve the issue. It also makes it easier for internal teams to find specific content stored in the text repositories.
- Text mining, similar to data mining. This uncovers related concepts in large volumes of conversations. It surfaces key topics that can be used in future analyses, like predicting or understanding guest behavior.
- Sentiment analysis. This helps you understand guest opinions by applying natural language processing to the text documents. It identifies how guests feel about key attributes of your product, brand or service – often in great detail.
Text data is by nature big data, so it needs to be stored differently than traditional quantitative information and will require a large amount of processing power to analyze. Once the data is quantified as we describe here, the results can be incorporated with traditional data sources into a wide variety of analyses, including revenue management algorithms or predictive analytics for retention, response likelihood or lifetime value calculations.
Given the prominence of reviews and ratings in hospitality, and the wide variety of applications, Natalie will cover this area in more detail in our next blog.
Digital intelligence is the process of deriving insights from digital interactions, so analysts must work with new types of data like click stream. Digital data is collected at every stage of the guest journey in a variety of systems and formats. The biggest challenge is to stitch together these fragmented sources of data, which come from online, offline, and even third-party sources. All of this data needs to be pulled together in a format useful for both analytics and reporting. There are challenges associated with integrating this data, cleansing it and ensuring it is analytics-ready, particularly because much of the data is in non-traditional formats.
Data-driven marketers use advanced analytics to perform sophisticated analyses, like regression, decision trees, or clustering, but they have traditionally been limited to using offline data (data collected through on-property interactions, or through reservation systems). This has been primarily due to restrictions on access rights to online data from third-party technology vendors. Even if hotels get access to online data, commonly available web and digital analytics tools mainly aggregate and report on historical information and thus are not well suited to perform predictive analysis. In this aggregated environment, obtaining an omni-channel, integrated view of a single guest across the fragmented digital universe has been extremely difficult. As a result, it has been practically impossible to get a data-centric, comprehensive view of the guest that could feed integrated marketing analytics, or more specifically, provide prescriptive recommendations for marketers. Enter digital intelligence.
Digital intelligence is defined as: “The capture, management and analysis of customer data to deliver a holistic view of the digital customer experience that drives the measurement, optimization and execution of digital customer interactions.”1 This requires that marketers focus on understanding the “who,” “what,” “where,” “when,” and the “why” of digital experiences, collecting detailed one to one data across channels, as opposed to aggregated snapshots channel by channel. As with any data project, it’s also important to consider the downstream activities and use cases you wish to support.
- Predictive analysis to identify what unique behaviors or attributes in a visitor’s digital journey are closely correlated with revenue generating events (like a conversion or up-sell). These behaviors and attributes can then be identified and fostered for future visitors.
- Analytically forecasting website visitation by traffic source, and identifying which ad-centric channels have the largest effect in increasing overall traffic (attribution modeling). The ad strategy can be adjusted based on which channels are most productive.
- Predicting online and offline behavioral drivers of digital conversions using analytically driven segmentation techniques, and improving outbound and inbound targeting rules for future marketing communication and personalization efforts.
To support the opportunities outlined here, web and mobile data, if collected and prepared appropriately, can be merged with your company’s or company-owned customer data, and then streamed into your analytics, visualization, and interaction automation systems.
Recent innovations in technology are making it possible for hospitality companies to move beyond the limitations of traditional web analytics (i.e., aggregated data and historical performance analysis as opposed to predictive modeling). Marketing departments can now integrate digital data with offline guest profiles, and use that complete picture of the guest in predictive modeling to support personalized content delivery, offers and recommendations digitally as well as when the guest is on property.
There are new and exciting use cases for predictive analytics on the horizon for hotel organizations. As the industry as a whole grapples with big data, you can better come to terms with the data from the explosion of interaction points with our guests. Using rich data and predictive analytics techniques you can drive towards improving the guest experience, optimizing operations to improve costs, maximize revenues and profits as well as increase the overall value of the relationships that you have with your guests.