Revenue vs GTR

When someone starts working at a Pennsylvania casino, they learn about  a concept called “Gross Terminal Revenue,” which replaces the traditional definition of slot revenue used in most states. It’s ubiquitous because it’s the revenue figure that is reported to the state regulators, and against which revenue taxes are calculated. Casinos in Maryland and West Virginia use this measure as well. So what exactly is GTR? Simply put, GTR = Revenue minus Promotional Play. Revenue is calculated the traditional way – wagers less payouts, or (coin-in − (coin-out + hand-pays)) – but then we also subtract Promo Play (free play, e-play, bonus dollars, etc).

As an example, traditionally, if a casino collects $1,000,000 in coin-in and pays out 92% (or holds 8%), it records $80,000 in slot revenue. But in a GTR state, we’d take this a step further and also subtract the promotional play. If that is, say, $16,000, then GTR is $80,000 − $16,000 = $64,000. This is the GTR number reported to the state and more importantly, the base for revenue taxes. When you’re paying 50%, 60%, or higher, as in Pennsylvania and Maryland, every dollar counts!

But why does this even matter to anyone (except accountants of course)? What are the implications for marketing? Surprisingly, I’ve found that many colleagues haven’t realized how the impact of a promotion may differ in a GTR state, as opposed to a traditional revenue one. Plenty of marketers and executives come to Pennsylvania or Maryland from New Jersey, Nevada, Connecticut, or other traditional revenue states without reconsidering certain promotional strategies.

The most obvious, is that all things being equal, promotional slot credit becomes far less expensive to give away than gifts or free food. In the above example, distributing $16,000 of promo play would reduce revenue taxes owed by approximately $8,800 (55%) in PA. If you gave away $16,000 in food comps, or in gift cards, or as 1600 $10 waffle irons, there would be no tax relief. (edit: in 2014, the PA Supreme Court reversed rulings by lower courts that these types of gifts should not be included in the “promotional expense” that is subtracted. So going forward, it may turn out that these expenses do provide the same tax benefits as promotional play).

Internally, EBITDA should be unaffected, whether promotional expenses are deducted from the top line or subtracted below as one of many expense items. But in reality, many management teams are (at least partially) incented on revenue or market share. Traditional revenue can be increased by giving away more promotional play. If a player visits with nothing but $100 of promo dollars, and leaves with $80 in cash, this is recorded as +$20 in traditional revenue, but it’s -$80 in GTR. So over-comping is much more detrimental to GTR than to traditional revenue. Bottom line, the GTR calculation subtracts promotional play, therefore reducing promotional play will increase GTR, all other things being equal. Yep, I buried the lede.

GTR also impacts the way we rate individual players, distinguishing between those who play mostly with their own money, and those who take advantage of every dollar of “free” money available to them. More on that in a future post.

Chinese government, a model of efficiency?

The lead story in today’s WSJ Section B is about Macau’s severe decline in gambling revenue — down 49% year-over-year February 2015 vs February 2014. There’s plenty to discuss there, especially the fact that the slowdown is widely attributed primarily to a crackdown on corruption — implying that for years, a huge portion of gambling funds were generated illegally, and that everyone seemed to know about it. Fascinating.

But what I also thought was interesting was the fact that city-wide statistical data was available and released so early in the month. Today is March 4, so even accounting for the 13-hour time difference between New York and Macau, these figures had to have been released on the 3rd. Which is extremely quick compared to U.S. jurisdictions. New Jersey generally releases monthly revenue data on the 10th. Nevada’s Abbreviated Revenue Release is almost a month behind — January data was published on Feb 27, 2015. Pennsylvania, which opened in 2006 and therefore has a relatively modern infrastructure, releases statewide slot data on the 3rd, but table games data doesn’t go out until around the 15th. Meanwhile Macau casinos are table-heavy, and yet they can release jurisdiction-wide data on the 3rd.

Why is revenue reporting not a totally-automated, human-free process? All the rules — the fields to report, the exact format — are standardized by each state’s governing body. There seems no reason, other than legacy, for reporting to be such a slow, inefficient process. This seems like a low-priority issue until you find out states like New Jersey are actually reducing the reporting requirements. There would be absolutely no reason to do this if the reporting was automated. Clearly the New Jersey casinos requested relief because their current reporting processes are burdensome. Every jurisdiction should strive to release data (and more of it, not less) faster. China can do it, so why not us?

Are you counting too much? (part 2)

In the last post, we discussed the importance of counting lift, rather than total revenue, when calculating ROI of a marketing campaign. Using total revenue will wildly inflate any ROI measurement. The advantage is that it is easy to calculate — we can see exactly who redeemed a specific offer, sum up everyone’s revenue on the redemption day, de-layer that revenue total if necessary, and just like that, we have a revenue total. On the other hand, pinning down lift, which is essentially an imaginary number — what would revenue have been if we didn’t run that promotion? — can often be difficult.

One excellent way, that most marketers are very familiar with, is to run a control group within each campaign. In the last post, we used a coffee maker giveaway as an example. 2000 players redeemed and generated $120,000 in total revenue. Let’s assume 15,000 players were qualified to receive this offer, and we held out 10%, or 1,500, as our control group. So 2,000 out of 13,500 redeemed, or 14.8%. What if we also found out that 180 members of our control group also visited that day — that’s a 12% “visit” rate without this incentive. In the example, we also stated that the 2,000 redeemers lost $120,000 in total ($60 average). Now let’s also assume that our 180 visitors from the control group lost $12,600 — that’s a $70 average. So the offer group had a higher redemption rate, but the control group had a higher average loss. If we extrapolate our 1,500 player control group to match the size of the offer group (multiply by 9), they would have lost $113,400. Now the lift generated by the promotion is pretty small — just $6,600. Now if we subtract the cost of the coffee makers (plus printing & postage, labor, miscellaneous costs), there’s a good chance that we’re actually underwater on this promotion.

Having a control group is terrific, but it’s a luxury we’re not always granted. What to do without one? One option is to review the visitation pattern of the invited group over the past few weeks. We can check how many visits were made by the 15,000 invitees and how much revenue they tended to generate. If we find out that over the past 60 days (x 15,000 players = 900,000 possible visits), this group of players made 99,000 total visits and lost $66 per visit on average, then it’s easy to calculate a typical visitation rate of 11% (99,000 / 900,000). More importantly, we can see that on a random day, the invite group generates $7.26 revenue per person (99,000 * 66 / 900,000), versus $8.00 (120,000 / 15,000) on the promotion day. This would imply that the lift from the promotion is $0.74 * 15,000 or $11,100.

Another statistic we can leverage is the non-redemption rate. In the example promotion, 2,000 out of 15,000 invitees redeemed. What if we also found out that 700 of the 15,000 visited and played on the giveaway day, but did not redeem the offer? This is a very useful piece of information, because it’s proof that some of those redemption trips would have occurred without the coffee maker incentive. Knowing this, we have to acknowledge that a chunk of the redeemers picked up a free gift while they were already here, but the decision to visit was not caused by the promotion. Exactly how many, we don’t know for sure. However, knowing that lots of players who were on-property, and eligible to redeem an offer didn’t actually do so, indicates that the gift or promotion wasn’t all that enticing, and therefore it is essential to greatly discount the revenue total, to account for redeemers whose trip was not incented by the promotion.

Finally, the most generic and simplest way of all — compare the day’s total revenue with the same weekday over the last 4 to 12 weeks. If the coffee maker promotion was on a Tuesday, and that specific Tuesday had 5% higher revenue than the average over the last 8 Tuesdays, then it’s fair to infer that 5% is your lift. Of course, you have to use common sense here, if you did another impactful promotion on the same day, if you had one “whale” who was personally responsible for a large portion of that lift on the promotion day, if weather affected some of the comparison days, etc., then obviously you need to make adjustments.

Whichever way you go, however, the key is to measure or estimate lift rather than total revenue attributable to the promotion. Marketers may not like it because it will certainly lower (by a lot) your ROI calculations, but any other way is simply fooling yourself, your marketers, and your executive team about the impact of your promotions.

Are you counting too much, when measuring the impact of promotions?

Measuring the impact of casino promotions and campaigns is essential to formulating a successful long-term marketing strategy – after all, how can you confidently plan a marketing calendar if you don’t know what has worked and what hasn’t? The pervasive problem at many (if not most) casinos is the inaccurate measurement of that impact. Marketers tend to greatly overstate the revenue gains generated by the promotions they roll out. This explains the contradiction between a marketer stating every campaign had a positive ROI while Finance reports a casino-wide revenue total that is down year-over-year.

Why does this happen? At most casinos, the revenue attributed to a campaign is the sum total of all revenue generated by every player who participated in that campaign. That is, if you invited people to pick up a free coffee maker on October 14, and 2000 people actually did, then all 2000 players’ total revenue on October 14 would be counted towards the coffee maker promotion. But wait, you might be saying, that’s not true – we de-layer the players’ revenue, so it doesn’t all go towards the coffee maker (if you are not familiar with de-layering, please see the footnote below). Still, the flaw is that every dollar of revenue is attributed to some marketing campaign – implicitly assuming that every visit by every player who redeems an offer was incented by your marketing. This is demonstrably false. Many players happily accept the free slot play, free gifts, free food, and whatever else your casino offers them – yet this is not the primary reason they visit, nor would their play completely dry up if these incentives were reduced or eliminated.

So going back to the coffee maker example… let’s assume the 2000 players who picked up their gift lost $120,000 in total on this day, for a $60 average. To attribute the entire $120,000 in revenue to the coffee maker is simply wrong. Instead, you must have some benchmark that measures what the likely revenue would have been without the incentive, and then calculate the difference. Only this amount can be attributed to the promotion. If these players would have generated $90,000 on a typical day, then the lift from the coffee maker giveaway was $30,000. If each coffee maker cost the casino $12, so total cost was $24,000, then we have a “profit” of $6,000 and an ROI of 25%. Which is certainly nice, but far different than a 400% ROI, which is what we’d have if we assumed all $120,000 was generated by the promotion.  It seems obvious, yet many casino analysts don’t perform their calculations this way.

So how do we create accurate benchmarks in order to determine lift? Stay tuned, we’ll look into that in a future post.

De-layering: splitting up a player’s total revenue into smaller pieces, assigning a unique piece to multiple “layered” marketing offers. So imagine a player in the above example who lost $240 on his/her trip, during which s/he picked up a coffee maker, and also redeemed a $40 free play coupon, and also enjoyed a free buffet. An analyst would likely attribute $80 (1/3) towards each of the three offers. This prevents double- or triple-counting, where the full $240 would be counted three times, as each of the three promotions was analyzed.

Two gamblers walk into a casino…

Unfortunately I don’t have a punch line.  But I do have a scenario.  So two new players walk into your casino — Alison generates $600 in theo but loses only $200, while Betsy generates only $200 in theo but with some bad luck, loses $600 real dollars. Long-term, who is likely to be the more valuable customer?

My gut was telling me that Betsy, who actually lost $600, would be worth more. After all, you know that her wallet, or potential loss, is 3 times larger than the Alison’s. Although casino marketers think in terms of theo, players themselves think in terms of dollars — so evidence of a $600 budget would seem to be worth more than evidence of a $200 budget.

I pulled some data on a group of players whose first visit was similar to the above. The Betsys had high actual losses but low theos (minimum 3x ratio), while the Alisons generated high theos but only lost a small amount, or even won some money. I focused only on players between $250 and $1000, to prevent a handful of very high worth players from dominating the data. Over on the right, you can see what the two groups look like, graphically.

There were about twice as many Alisons, however the graph makes the ratio look even larger.  Because theo can’t go negative, the Betsys (in blue) are much more concentrated on the graph.  Alisons are more spread out since the actual loss, on the y-axis, can be anywhere from +200 to -800.

All the players I looked at had their first visit between January and June 2012, giving them a full 2 years to establish a visitation pattern and gaming history. Here is the comparison of the two groups over the next 2 years:

Surprisingly (to me, at least), the high theo / low actual group (the Alisons) turned out to be worth more long-term, primarily because they are more likely to return and to visit more. 87% of the Alisons returned vs only 69% of the Betsys. Even after eliminating the non-returners, Alisons have returned 22.6 times vs 17.7 visits from the Betsys. The additional trips give the Alisons a huge advantage.

On per-trip spend, the results are a little bit more mixed — the Alisons (high theo on first visit) continue to generate far more theo on future visits ($174 vs $142) but the Betsys (high actual loss on first trip) continue to lose more actual dollars ($138 to $120). You may have noticed that the Betsys have, on average, a very small gap between their actual and theoretical losses ($4) while the Alisons have a much larger gap of $54. This indicates that the results of the first trip weren’t entirely due to good luck and bad luck — the Alisons, who generated high theo despite not losing much, continue to generate a lot more theo on subsequent trips. This indicates a playing style — perhaps a lot of video-poker or low-hold slots, more max betting, etc. — that tends to generate more theo than actual losses.

There’s another important factor driving these differences, however, and that is the marketing to each of the two groups. Whether your casino utilizes Avg Daily Worth or some other method that attempts to include actual wins/losses into the equation, theo is still the dominant marketing measure. I would guess that every casino in America would have sent more generous offers to Alison, based on her $600 theo, than to Betsy with her $200 theo, even though Betsy had the much larger actual loss. If the Alisons get better offers initially, it shouldn’t be all that surprising that they are more likely to return. Further, the effect tends to snowball — the results seem to indicate that Alisons may be more experienced players, and they know how to lose less of their own money while spending more of the casinos (via free slot play — this helps explain the large theo vs actual gap). So instead of this being a clear example of one group being worth more than the other group, we must consider how much the different marketing to the two groups contributes to the difference in value. A chicken and egg problem indeed.

A wise strategy here, it appears, would be to spend a little more investing in and cultivating the Betsys. When a new player loses a lot more than the theo generated, the casino should base its initial offers entirely off of the player’s actual loss, regardless of theo. We don’t want to allow one class of player to underperform because we are under-marketing. Higher offers are more likely to get the player to return (an abandoned player is worth zero, so it is imperative to get them back), and over time and with more visits to look back upon, offers can be adjusted to reflect their long-term value.

Allowing your players to win occasionally can be a good thing! Who knew?

How much of an impact does good luck play on your customers?  Analytically, we choose to largely eliminate the impact of luck (because we can’t control it) by focusing on theoretical.  But players don’t think about theo, they measure wins and losses by dollars.  So does their wallet.  Common sense would suggest that winning players are more likely to return to your property, first, because they feel lucky, and secondly, they have some extra money with which to gamble.  So how do we check this hypothesis?  How large is the impact?

This assumption was simple enough to test out — I looked at a sample of regular players (minimum of once per month) who had won or lost $250 in a single day, and then calculated how many days passed until their next visit. Not too surprisingly, players who won did in fact come back more quickly. Slot players who lost $250+ came back, on average, 6.5 days later. But those who won $250+ came back after just 4.7 days — nearly 2 days earlier.

I broke down the groups further — by the size of the loss, by the frequency of the player, and by slot vs tables. The entire table is below. The magnitude changes but the theme remains constant — winners return to your casino much faster than losers do. This result is probably not a huge surprise, but the overall impact is surprisingly large. Remember all the players in the sample are “regulars,” so we might assume they visit rain or shine, win or lose. Instead, these figures show their return can’t just be taken for granted.

Consider that 6.5 days between trips is 37% higher than 4.7 days.  That’s a lot of extra trips, especially since these are already relatively frequent guests.  Now, while we can’t magically create winners, we do have some control over the players’ experiences.  Especially during lengthy periods without organic growth, executives start proposing raising hold percentages or lowering free play reinvestment. Both concepts are attempts to extract more profit per player-trip, and as such, they make it tougher for a player to have a winning trip. When these ideas get rolled out, it is assumed that squeezing a few extra dollars per customer per trip will improve profitability — but this only works if trip counts don’t decline.  Usually that impact is ignored or assumed to be very low.  Now we know it’s not so small.  Certainly, raising holds by a tenth or two won’t affect all trips, so the trip decline won’t be anywhere near 37% — but it’s likely to be higher than anticipated.

* This analysis only measured trips where the player eventually returned to the property — any final trips were not included since “days until next visit” can’t be calculated. Therefore, trips that led a player to completely abandon aren’t included. One would imagine that’s more common after a losing trip, rather than a winning one. Therefore, the gap is likely to be slightly larger than what is reported.

Why are so many of my players playing below their averages? (part 2)

In Part 1, we introduced the skewed distribution of daily theo results, showing that players played well below their average on most days, offset by a handful of days significantly above their average — perhaps 2x, 5x or more. Here’s what that histogram looks like (note: the right-most bar at 400% actually combines all results above 4x average):

And not surprisingly, a raw coin-in graph looks almost identical:

But how about actual win? After all, that can go below zero (“negative win” is from the casino’s point-of-view, it’s when a player has a winning day). Well, I’d love to post that histogram here as well, but doing so would reveal certain sensitive statistics — like the exact percentage of players who actually have a winning day. Suffice it to say that from zero up, that graph looks remarkably similar to the previous two, with the addition of pretty normally-distributed probabilities to the left of the zero (representing player wins).

As nice as these histograms are, it’s still somwhat difficult to specifically quantify the likelihood of specific outcomes. Hopefully the following summary table will answer the most common questions:

So, for example, a random player’s daily theo total will be under his/her average 62% of the time. The median percentage (the percentage at which half of trips will be above and half below) is only 67% of the player’s average. That is a huge departure from the 100% expected from a normal distribution, and a major impact if you’re designing a marketing program with the assumption that players generally play to their average theo — in fact, they play a full 1/3 less than that. Sure, a small number of customers should make up for that by having the outlier days and playing 3x, 4x of their average or more — but if your marketing program favors, say, a free room to a $250 player over a $200 cash customer (let’s forget about revenue taxes and other miscellaneous costs for the purposes of this comparison), you might find that the majority of those free rooms don’t actually generate the revenue that would have been locked in by the cash customer.

Why are so many of my players playing below their averages?

It can be a sobering realization — that your promotion or campaign wasn’t as successful as initially projected, in large part because most individual players did not play up to their daily averages (whether the measure is Theo, Actual Win, or Handle). My $250 ADT guys only played to $180, my $500 guys only played to $350, etc.

First of all, the more valuable the promotion, the more likely people will visit just to pick up the giveaway item or the free slot play or whatever it is you’re giving out. Which means a large number of respondents aren’t really there to gamble, or they are just squeezing in a little gambling on a day they normally would be doing something else. So, paradoxically, the better the offer, the more likely it is that most players will not achieve their typical averages.

However, a more influential factor is simply that on most trips, all players fail to reach their personal averages. The reason? The distribution of outcomes is heavily skewed to the low side. What do I mean? Well, let’s think about a player with 5 visits and a $100 ADT. Knowing he is a $100 player, would you expect that an $80 theo day is equally as likely as a $120 day? If you said yes, your mind is probably modeling a normal distribution of daily outcomes — i.e. being 20% above the average is just as likely as being 20% below the average (you can substitute any number for the 20%).

But Theo cannot be distriubted normally. Theo can’t be below zero, but there’s no upper limit. Right there, we’ve violated the normal distribution. Further, think of how slot machines pay off, or how players generally act. Most slot machines are set up to have a lot of small outcomes (more losses than wins, obviously), offset by a small number of large slot jackpots. Players will come to the casino with a bankroll they have to lose, which they often do, again offset by that rare “hot streak” where they triple or 5x or 10x their money.

Let’s return to our $100 ADT player. Considering the distribution scenario just laid out, it’s likely he reached a $100 ADT via days like these: $75, $46, $235, $61, $83. Here, 4 of his 5 trips are below-average. Other than the one big day he had (we’ll assume he was having some really good luck on this day), his average is only $66. Seeing a distribution like this, would you be surprised if he played below $100 on his next visit? Hopefully not.

It’s one thing to invent a hypothetical player that fits nicely into a certain scenario. But this blog likes to focus on facts. So I calculated the averages for regular players (at least once per month) over the past 18 months, then compared individual daily results with the players’ averages. Here’s the distribution of the daily-to-average Theo ratio:

Not too surprisingly, this is anything but a normal distribution. The average value is definitely not the most likely outcome. A day with almost no play is twice as likely as any other outcome, often influenced by the reaoning listed above, that a lot of trips are motivated more by the gift pickup or free slot play than a specific desire to gamble that day (for the record, I eliminated all trips with zero coin-in or theo — if we counted “no play” trips, the left-most bar would be even higher). But moreso, the long tail to the right demonstrates that player averages are influenced by a small number of very large outliers. It also shows that for most players, their “average” daily Theo and “most likely” daily Theo are not the same, actually often pretty far apart. Going back to the original scenario, if you’re expecting your $100 ADT player to reach $100 on half of his visits, you’ll be sorely disappointed.

More in Part 2

Jackie Gaughan, 1920 – 2014

Although I knew of Mr. Gaughan, I certainly did not know him, the man, personally. Truthfully, I don’t know a whole lot about him, either, and in doing a little research, it appears I’m not alone. It’s rather difficult to find any old articles (say, from before 2009) about Gaughan or his casinos.

But even with the limited amount of hard news about the man, what becomes very clear very quickly from anecdotes, personal stories, and testimonials, was that he loved the business of gambling, loved Downtown Las Vegas, and most of all, loved people — both employees and customers. And that, in and of itself, is worth celebrating.

Without knowing the man, I can’t say why he didn’t expand much beyond Downtown like Sam Boyd, why he wasn’t interested in spectacle like Jay Sarno, or why he didn’t have the outsized reputation of Benny Binion. I imagine he saw the potential beyond Fremont Street, and recognized the opportuntity — after all, his son Michael has been wildly successful building locals casinos all over Las Vegas. My guess is that he was comfortable Downtown, these were his employees, his customers, his people, and he simply didn’t have much desire to go anywhere else.

Gaughan’s portfolio of casinos didn’t include any headline-grabbers — Elvis never played there, or the Rat Pack, or Seigfried & Roy. He catered to the blue-collar crowd, another tradition that rubbed off on Michael. As casinos continue to proliferate around the country, most of them are less like the Bellagio or Caesars Palace, but more like Gaughan’s places — comfortable, if not luxurious. It’s too bad he didn’t write a management book — then again, guys like him probably thought the secrets to success were pretty obvious: work hard, treat employees like family members, always give customers a fair deal, understand every aspect of the business and every corner of your property.

The El Cortez, as part of its brilliant marketing campaign embracing the property’s old-school roots, has repeatedly dropped reminders that Jackie Gaughan still lived in the hotel’s penthouse and still played cards in the casino almost every day (details that nearly every obituary has picked up). While today’s corporate entities and private equity firms and foreign banks utilize Las Vegas casinos simply as vehicles to enhance their portfolios, Jackie Gaughan was a man who saw Downtown Las Vegas, and the casinos he owned, as the destination itself. Not just to make his fortune, but for his life. That’s an attitude that deserves respect. Rest in peace, Jackie, I wish I’d known you better.

Theo and Actual eventually converge, don’t they? (part 2)

In the previous post, we saw that Theo and Actual are unlikely to converge upon a single “true” value for individual players, during a practical evaluation timeframe used to make offers (i.e. 3 to 12 months). About half of your high-frequency players will have a gap between Theo and Actual of greater than 40%, making it very difficult to ensure your players are being segmented into the appropriate bucket.

A few readers may be thinking, “my casino uses ADW (Average Daily Worth), we combine both Theo and Actual into a single calculation, that should solve this issue!” Except it really doesn’t. Yes, ADW is closer to Actual than pure Average Daily Theo, but only by a little. In the table below, you can see that ADW is more likely than ADT to be within range of Actual, but the improvement is not that much.

Look at the numbers in another way. Using pure ADT, 57.9% of my players will have a Theo-Actual gap exceeding 40% (1 – 42.1%). Using ADW, 50.4% will have a 40%+ gap. That’s a reduction of just 13% of outliers. ADW helps, but not a whole lot.

So if half of your players have ratings that can’t be “trusted,” and by that I mean that the two primary statistics that you pay attention to are telling different stories, how can you effectively evaluate your players and send them the optimal offer?  Which do you prioritize?

It’s often stated that Theo is the more “pure” number, that it eliminates luck (which neither the casino nor the player can control) and measures how much revenue the player should generate given his/her volume of wagers and the house advantage.  There are two problems with that assumption: a) Theo is still highly dependent on luck — when a player is winning, s/he can play much longer and therefore generate more wagers, and thus more Theo, and b) the player is primarily interested on his/her Actual win or loss, so that the casino is emphasizing a measure that does not reflect the player’s viewpoint of the success or failure of the gaming experience.  When players have a really bad day and lose very quickly, not only have they had a bad gaming experience but in addition, the casino tends to “short-change” them as they don’t generate too much theo.  ADW might help here a little but most ADW formulas only credit a player with 40% of his/her actual daily loss.

A better solution, instead of relying on one single calculation to drive an entire marketing campaign, would be to calculate multiple statistics for each player, compare them, and utilize the most relevant one(s) for each player.  At a minimum, you can calculate ADT/ADW, plus ADA (average daily actual), cumulative actual loss and cumulative theo.  Beyond that, you can calculate those same statistics after removing specific outliers (perhaps the player’s max win day, thereby not penalizing them for having a great day), you can calculate volatility (standard deviation, to see who plays similar amounts each day, and who is all over the map), and you can replace arithmetic averages with geometric averages (which will always return a lower value, assuming the inputs are all positive, like Theo or Worth).  Further, you can make these calculations over multiple time periods and find out things like whether a player’s recent stats are significantly higher or lower than longer-term values (signifying an upwards or downwards change in play patterns) , whether a player’s actual loss is consistently much greater than his theo (indicating he may be a low-skilled video poker or blackjack player, for example), whether a player adjusts his visitation upwards after actual wins and slows down after losses (a potential measure of overall loyalty), among other things.

Without a doubt, things can get complicated quickly, and there’s no reason to add complexity without a clear improvement in the end result.  But starting out by performing some tests can definitely guide you towards which statistics are most likely to influence player behavior.  The alternative, and I’ve seen this at plenty of casinos, is to rely on a single, basic statistic (like ADW) for marketing efforts and to never attempt to adjust or improve.  These basic stats have a built-in, quantifiable error rate (see the graphs and tables above), so it’s sub-optimal to be complacent.

Theo and Actual eventually converge, don’t they?

When we analyze casino performance for a month, or a day, or even a single day’s promotion, we often look at both Theo and Actual generated. In the aggregate, the two numbers are usually reasonably close, similar enough to give an impression of whether that month or day or promotion was succesful. When they differ substantially, further digging often uncovers a handful of, or even just one, large players who either got very lucky or unlucky, and the size of their action heavily influenced the total Actual. Once these outliers are removed, the Actual and Theo usually come pretty close, in aggregate, for the remaining 98-99% of your guests.

However, when we send offers to our customers, each guest is evaluated independently, where his or her previous visits are the only data points considered. As opposed to an entire day, or promotion, where there are usually many hundreds, if not thousands of players to aggregate, reviewing a single player’s history over the past 3, 6 or 12 months leaves us with far fewer data points. And this leads to a dilemma — when Theo and Actual are wildly different, which should we prioritize?

At this point I imagine you saying, “ok, for the very infrequent guest, sure, Theo and Actual are probably pretty far off. But among my regulars, my high-frequency players, they converge. The math eventually catches up. I once sat in a seminar about this…”

I was curious how true this was. I had been investigating a group of players who had lost a lot of money, but who were bucketed in low-worth segments. The high number of these players surprised me. Evenutally, I ran some database queries to really test out this whole concept of convergence, focusing only on high-frequency players — at least one visit every 3 days:

Each dot represents one player. Placement on the x-axis is determined by their number of visits in 90 days. The y-axis represents their ratio of Theo / Actual. 1.00 or 100% indicates that player’s Theo and Actual were identical. 200% means the player actually lost 2x his or her theo, and so on. A ratio below zero means the player was a winner. What is so surprising is how wide the distribution is — if Theo and Actual really converged as expected, nearly all the dots would be in a tight region right around 100%. And keep in mind the minimum number of trips in this data set is 30 (over 90 days), so it only covers highly-frequent guests.  95%+ of your customers visit less often, and are therefore even more likely to have a huge gap between their Theo and Actuals.

Eyeballing this chart, it’s clear that the number of outliers gets smaller as the frequency increases — making it appear that convergence increases as trip count rises.  But this is partially an illusion — there are fewer outlier because there are far fewer data points in total in the very high frequency range.  In this data set, the number of 30-39 visit players is 10x the number of 70-79 visit players. So that eyeball view is misleading, because what appears to be increased convergence is partially due to simply fewer data points. To get a better picture, I calculated the probability of a specific player’s ratio to be within 100% of true convergence (i.e. ratio between 0 and 200%), and then within 40% (i.e. ratio within 60 to 140%):

This is only a single data sample, but it appears to indicate that convergence doesn’t get significantly stronger until you get to 80+ visits. Until then, 15-20% of your guests will be outside a 100% window, and nearly half outside a 40% window. 40% would translate into a “true” $200 player being evaluated at below $120 or over $280. Depending on your offer matrix, that could easily be two, three or more segments.

Given this lack of convergence and the high error or volatility rate, how can we segment our customers correctly? If the data is unreliable, what can we count on?  What can we possibly do for the majority of my players — who don’t visit anywhere near 30+ times per quarter? Which should I prioritize, Theo or Actual? Will Average Daily Worth, instead of pure Theo, help me?

We’ll try to answer some of these questions in Part 2…

Can Obama’s braintrust really help a casino?

Today’s fascinating New York Times article focused on a young company, Analytics Media Group (A.M.G.), which is essentially a spin-off of Barack Obama’s analytics team during his 2012 bid for re-election. With the election over, the company, which was founded in December, is now selling its brainpower and marketing expertise to companies. And the Times article details one of its first major clients — Caesars Entertainment.

Caesars, led by former Harvard Business School professor Gary Loveman, flaunts its expertise in technology and analytics any chance it gets. In magazine articles, books, blogs and in person, it loves to brag that it’s the leader in the gaming industry when it comes to brainpower (it should be noted that all this so-called expertise contributes to a company which has lost $3 billion total in its last 3 fiscal years, so the effectiveness of Caesars’ strategies to increase profits is far from proven). So the chance to work alongside the analytic geniuses behind the Obama re-election must have been exceptionally attractive.

My question is how well-suited A.M.G. is for this job, whether their area of expertise translates well to the needs of a company like Caesars. Off the top of my head, I can think of numerous significant distinctions between an election and a company:

1) Caesars is an ongoing concern, that needs to attract players day after day, month after month, year after year. Caesars needs to stimulate repeatable demand in order to be successful — there is no end date. An election is a single event on a single day. The campaign team only needs voters to take an action once, all on a single day, which allows for greatly coordinated action and communication, and requires no ongoing commitment from voters (i.e. customers).

2) To a campaign, every person is equal, worth exactly the same amount — one vote. If the campaign can convince ten people to vote for Obama, that’s 10 votes. For Caesars, every person has a different dollar value, based on how much they gamble (less how much they consume in offers, comped food, free hotel rooms, etc). Caesars could successfully market to 10 new players and later find out they all have very low spend, meanwhile it may have ignored one individual whose spending is greater than all 10 combined.

3) The campaign can appeal to voters in intangible, but admirable, ways — civic duty, patriotism, controlling one’s future, etc. There is nothing like that for Caesars to use to generate action.

4) As a publicly-traded company, it’s not enough for Caesars to generate revenue or show a profit (although, having lost money for the last 3 years, just showing any profit would be welcomed). It needs to constantly grow, i.e. beat last year’s numbers. Thus, the bar gets higher every year, and all the gains in one year need to be improved upon in the next. Growth didn’t matter to the Obama campaign — in 2008, he earned 52.9% of the popular vote, but his burden was not to beat that in 2012, just get over 50% (and he only got 50.95%). In electoral votes, which is what really matters, Obama earned 365 in 2008, then just 332 in 2012. This performance would not be acceptable for a public company, but works just fine for a winner-takes-all campaign.

5) Because electoral votes, and not the popular vote, is what really determines the winner, and because electoral votes are awarded by state, the campaign was able to count on winning certain states without having to do much work (and knew it would lose some states no matter how hard it fought). This enabled the campaign to ignore a huge majority of the country (California, Texas, New York, etc) and concentrate all their efforts on just a few swing states that could potentially go for either Obama or Romney. Caesars doesn’t have that luxury, it can’t just take entire states and write them off. Therefore, its resources will be spread much thinner (including internationally).

6) The Obama campaign could count on tons of free marketing assistance from neighborhood leaders and organizers. People who would knock on doors, plant signs, and make phone calls on behalf of the president without compensation. Pretty sure that Caesars won’t get the same kind of volunteer support.

7) Casinos rely on their player database for nearly all targeted marketing, which leads to a bit of a Catch-22 — players won’t visit Caesars until they receive a marketing offer, but they won’t receive a marketing offer until they visit a Caesars casino. Marketing to unknowns is possible, but it’s expensive — purchasing customer lists from credit card companies, hotel chains or airlines, sending hosts out on recruiting missions, lots of mass media. This limits the company’s ability to generate new customers. Meanwhile the campaign had access to voter rolls, census data and other public information in order to find “un-engaged” voters and recruit them to the Obama side.

8) The Obama campaign had one opponent to beat, Mitt Romney. Caesars has tons of competition to which a loyal customer can defect. In fact, it’s likely that the majority of Caesars customers already have a relationship with one or more competing casinos, so they are likely receiving marketing offers from companies besides Caesars. Further, the competitive landscape is far different across various geographic areas. Caesars is the market leader in some jurisdictions, and it trails in others. The company can’t be laser-focused on a single competitive threat like the Obama campaign could.

I’m sure there are other major differences — these are the ones that quickly come to mind. These are certainly plenty to wonder whether the A.M.G. team can translate its campaign experience effectively into a commercial marketing tool. The article doesn’t state whether A.M.G. actually landed the gig with Caesars, nor does A.M.G.’s web site provide a list of clients.

Despite the concerns listed, there are a few ideas from the campiagn that could improve the effectiveness of a casino’s marketing strategy. Maybe A.M.G. covered these in their discussions with Caesars. My primary takeaways would include:

1) Seek out the customers with the most upside potential and market to these players in a disproportionate manner. Although Caesars can’t completely ignore customers as in Obama’s “swing state” example, the company can mine its database to find players with high upside and focus a large percentage of its effort on these individuals. This doesn’t simply mean more offers and more costs — it could mean assigning hosts to players who might not currently play enough to receive personal attention, make offers more flexible, perhaps sharing information with these players like which slot machines have the lowest hold percentage, etc. Earning loyalty early can have a huge payoff in the long-term.

2) Innovate on the measures used to evaluate players. Nearly every casino segments customers by their average gaming spend, or a derivation (average daily theo, average daily worth, average trip theo, etc). The fallacy here is using a universal measure across all players. So if a company decides to look backwards for 6 months, it generally does that for everyone — whether the customer has visited once in that 6 month period, or 60 times. Not only is the relevancy of the earlier months vastly different between these players, but computing an average ignores directional trends, doesn’t differentiate on recency, hides max and min values, lacks statistical significance with too few samples, etc etc. The Obama campaign switched from traditional Nielsen television ratings to better data from a new company called Rentrak. Similarly, perhaps it’s time for casinos to move beyond simple averages to more sophisticated statistical analysis.

3) Consolidate the marketing calendar into fewer, bigger events. Many casinos feel pressure to “drive the day,” every day — therefore there aren’t any empty, non-promotional days left on their calendars. They do this despite the fact that most players also receive individually-targeted offers, so they would never be completely without incentives to visit, even on a “blank” day. The problems with this strategy are that a) expanding the marketing calendar doesn’t necessarily expand the customers’ bankroll… yes, you may be stealing share from a competitor, but you’ve also given loyal players a reason to split their spending over multiple days to take advantage of multiple offers (and higher costs to you) and b) the individual days’ promotions will be less special… splitting a marketing budget over 30 days means smaller promotions than splitting it over, say 12 or 20 days. As mentioned in #1 above, coordinating efforts to maximize the impact of a single event will likely prove more effective than spreading out resources. A single event isn’t realistic for a casino, but it can reduce the number and make them more impactful.