Arming Analytics

The study of data is wide-ranging, but a focus is necessary to achieve goals

“In business, as in baseball, the question isn’t whether or not you’ll jump into analytics. The question is when. Do you want to ride the analytics horse to profitability or follow it with a shovel?” – Rob Neyer, ESPN

We’ve all heard the buzzwords: data science, big data, machine learning, analytics. But in with operators both inside and outside of gaming, there’s a lot of confusion about what these terms mean.

This is understandable, as simple web searches turn up a lot of different definitions and examples. But to be certain, while we can agree that as an industry we do an excellent job of collecting data—marketing, gaming, point-of-sale, hotel, labor, financial—there is still a large opportunity to leverage that data to improve operations.

In this article, I’ll share my perspective on how analytics as a discipline is evolving in the industry, and some not-so-obvious areas where a good analytics program can provide value. And how, as author Rob Neyer so eloquently puts it, operators can “ride the analytics horse to profitability,” and not “follow it with a shovel.”

One of the most common questions I’m asked is, “How do we get started? I don’t have a huge budget.” I’ll address that, but first I think it’s helpful to provide a framework for the different ways we use analytics, and how that relates to your business.

 

Four types of analytics, and how we’re evolving as an industry

“Analytics” is a broad term, really referring to any program to analyze and utilize data, so it’s helpful to break analytics down into subcategories. While there are many ways to do this, I prefer the following quadrant-oriented picture of analytics:

The left-to-right axis ranges from analysis of past events to analysis of future events, with the latter being prediction-based. Bottom-to-top we are moving from the who-what-when-where to the why and the how, adding intelligence to our information.

These axes cut analytics into four major categories:

Descriptive analytics

This category includes summary data and gives a lens into “What happened?” Examples include daily operating reports and marketing campaign summaries. Descriptive analytics answers questions such as, “How much coin in/win did we do?” “Who visited the property?”“What time of day did we peak?” “How many redeemers did this campaign have?” “What was our hotel occupancy versus the same day last year?” Seasoned operators have well-defined processes for descriptive analytics, and most key leaders within the company receive top-line information that they use to manage their areas of the business.

Diagnostic analytics

This category takes descriptive analytics and layers in the “why?” and the “how?” This typically blends descriptive analytics from different sources, and it requires a few things to really achieve success. These include a strong data analysis team and excellent communication between analyst teams and operations, as this data scientist reminds us:

Diagnostic analytics often stem from questions such as “Why was the coin-in so low this weekend?” or “Why was the slot hold so low last month?” or “Why was EBITDA so low last month?” (Yes, these questions are thematic, as analytics teams know all too well!)

Answers start with descriptive analytics—first verifying that indeed the observation matches the data and preparing a summary versus prior month or year. The diagnostics then can lead analytics teams into the usual: weather, calendar, marketing, competitor programming, events in town. But it also can lead down paths such as looking at year-over-year game mix, hotel bookings and yield curves, restaurant covers and COGS, and labor.

Collaboration with operators here is key. If the analysts aren’t creating models that operators understand, the operators can’t build confidence, and analysis goes unused. All analysts understand that their recommendations won’t always be followed, but helping analysts understand the process—how and why decisions are being made—allows analysts to begin anticipating questions and thinking like an operator. Learning how operators approach problems helps analysts generate qualitative improvements to their models and analysis and provide insights and confidence to decision-making.

Predictive Analytics

In predictive analytics, analysts predict future scenarios. We already do this during budget season: What will our revenue be next year, and how will it spread? Take what we spent last year and add 2 percent. That is a prediction. These can be very useful predictions—if unsophisticated. What will a patron spend the next time she plays? Let’s use her 12-month ADT—this is another prediction. It’s important to realize that we only care about past play for its use in predicting future play and behavior. If we somehow knew that next month, the patron would play at half her ADT, we’d adjust reinvestment accordingly. Maybe we could make better forecasts of future play using seasonality, around tax season or the holidays. Maybe the patron has been declining for some time, yet redeeming all of her free play, so we should cut the patron’s offers before she becomes unprofitable.

Enter machine learning.

Machine learning, a subfield of artificial intelligence, is the process by which computers process data to “learn” without being explicitly programmed. For example, computers can “learn” spending habits by season and day of week, or the attributes of new card signups that lead to no return visitation. Both can help you market more effectively. These techniques are ubiquitous in the tech world, but there are very few analysts in gaming doing this sort of work.

Prescriptive analytics

Think of prescriptive analytics as scenario-based predictive analytics. Predictive analytics answers the question, “What is the likely outcome?” Prescriptive analytics takes this a step further and answers the question, “How can we intervene to get the best possible outcome?” From here, we can start to make great business decisions. For example, we can predict what our table revenue will be on Friday based on historical data. This assumes no changes to operations.

A prescriptive approach might look at all possible yield strategies and tell you that you’d increase profitability by raising a few of the $10 minimums to $15 and staggering table openings instead of opening six new $10 blackjack tables at 8 p.m. Friday like clockwork. Similar techniques aim to optimize the bundles of offers in a monthly mailer, product mix on the gaming floor, and room rates.

If you’re at the “descriptive” phase, you’re on par with most of the industry. If you’ve got a good “diagnostic” analyst, hold on to her for dear life. And if you’re pushing the envelope into machine learning and advanced predictive or prescriptive models, well done—you’re at the forefront of the industry right now.

 

Predictive Analytics in Practice

Most of the examples we see about advanced analytics seem grandiose—value every patron at the individual level, predict every trip’s worth, provide the right incentives to optimize profit. And there is a lot of value to be extracted from the marketing data we have. But analytics can help in all areas of the business, and here are a few examples to illustrate the point.

Slot Maintenance

In exploring the exception codes that machines send to the slot system, we saw results similar to the following:

This plot shows the number of times per day that a card reader failed to read a swiped card. Everyone sees the large spike, of course, but the real feature here is seeing hiccups before the catastrophic failure. These hiccups allow us to predict the card reader failure, and we can send a technician to service the device before it causes several days of poor guest experience. Oh, and imagine this is the bill acceptor.

Turnover

We all know how costly turnover is. In some recent work, we were able to observe strong correlations between drive time to work and call-outs, no-shows and turnover. If you’re saying “well, duh!” then I’d ask: Does your hiring team favor a candidate that lives a shorter distance over a similar candidate that lives farther away? How would your hiring be different if HR said “your first-choice candidate is 40 percent likely to turn within a year, and your second-choice candidate is only 10 percent?”

Consult with your HR and legal teams before building this model to ensure that you aren’t at risk of violating any labor laws, but these models can mitigate staffing issues from hire-time. “But what about all the employees I have that live far away today?” Well, if drive time is an issue, make sure you’re creating an environment your team members love—maybe car pool incentives as a “green” initiative or weekly gas card giveaways for high performers. Involve your HR team and your marketers—they get guests to drive in every day. And for your property, it may not be drive time. Maybe it’s closeness to the competitor, or something else, but you have the data and just need to view it through the right lens.

Scheduling

Use prediction to optimize scheduling. Most of your hotel check-ins are the afternoon? Then you probably don’t have as many front-desk staff members on grave as you do on swing. But are you more efficient on grave or on swing? How many check-ins and check-outs did you process per hour of labor? With maximum efficiency, you’d process the same per hour of labor on all shifts. Even with labor as the most expensive line item in many casino budgets, most properties cannot answer this type of question. With the increased pressure on margins, it’s important for operations to seek efficiency, and this is a great place to start. Develop hourly “key volume indicators” for each position, and see how efficient your labor is.

Guest and Employee Satisfaction Surveys

Predictive modeling can assist with analysis of survey data as well, answering questions such as, “How much difference in overall guest satisfaction does a hotel satisfaction 5 vs. a 3 rating make?” Or, “What responses can we use to predict turnover from employee surveys?” Or, “What indicators in free responses are precursors to a guest defecting?” These are all quantifiable with the right modeling techniques—again, you already have the data.

An Analytics Roadmap

“So where do I start? I can’t afford a data scientist!” Fair question. Here are some simple steps that organizations can take to begin to incorporate analytics into existing processes:

  • Develop a list of questions: Maybe you’re inspired by some of the examples above, or maybe you have burning questions about your own organization. Develop a list, polish it up, and ask management to support the effort with a resource. We’ve seen a lot of success with “10 percent time”—allowing an existing analyst to work on a project outside the ordinary scope of her duties for 10 percent of her time, say, Friday afternoons. Many analysts will even work an extra four hours if given this opportunity.
  • Get alignment from management: None of this works without a commitment from the executive team. They need to buy into the process, and this also creates urgency and importance for the analysts.
  • Build reports: These can be in a business intelligence tool like Tableau, or simply executed in Excel. The goal is to develop reports that help operators do their jobs better. If the reports are hard to read or poorly designed—perhaps just a table of numbers—you won’t get adoption. And adoption is key. I’m a reports expert, and I still work with a designer every time I build a report. It’s that important to develop something people want to use.
  • Start with the P&L: Don’t think that “analytics” must be fancy. We’ve had major successes just working through operators’ P&L statements. Compare properties if you have two or more, or versus prior years if you have only one. Find out how your spend mix has changed and why. It’s very easy for unnecessary expenses to remain hidden for years.
  • Don’t start with software: With commitment from management, targeted questions, resources to work on these problems, and metrics to measure success, you’re ready to evaluate software to make your team’s life easier. Without these things, you’ll have a beautiful Ferrari in the driveway and everyone will be looking at one another wondering who can drive a stick.
  • Focus on the methodology: “How much did we generate because of the giveaway?” is a common question, and a common answer is, “Well, we had 1,000 redeemers, and they averaged $150 ADT. The gifts cost us $20 each, so we netted $130,000.” But this ignores how many of those 1,000 patrons were going to play without the gift. Or maybe some of them played less because of the long line for the gift. Or maybe some people came in on Thursday for the gift and as a result we lost their Saturday trip. It’s not easy to develop a framework for how to answer these questions, but having analysts who think through the questions to challenge current thinking and processes is a great start. And this must be collaborative—remember, we’re all on the same team: Team Profit.

Analytics can be used to enhance every area of the business. Arthur Neilsen, of Nielsen Media Research fame, once said, “The price of light is less than the cost of darkness.” Expect the reliance on analytics to grow in the coming years as the industry continues to turn on the lights.

Brian Wyman
Brian Wyman, data scientist and senior vice president of operations and data analytics at The Innovation Group, holds a doctorate in mathematics from the University of Michigan. Wyman works with clients to transform their data into actionable insights and bottom-line financial results. Along with The Innovation Group’s analytics efforts, Wyman oversees the group’s sports betting advisory services and its Las Vegas office.