A recent New York Times investigation into DraftKings should make anyone who cares about consumer choice uncomfortable.
The Times reports that DraftKings has used machine learning to help determine which customers should receive promotional offers. According to former employees and internal documents reviewed by the newspaper, one model attempted to predict which gamblers would respond to incentives by betting—and ultimately losing—more money. The more responsive a customer appeared to be, the more attractive that customer became as a target for said promotions.
These claims raise legitimate questions of both consumer choice and protection.
DraftKings employees also worked on machine-learning tools intended to identify customers at risk of developing gambling problems. Those systems could analyze behaviors such as deposits, withdrawals, and attempts to chase losses, potentially allowing the company to intervene before a customer’s gambling spiraled out of control. DraftKings decided not to deploy that predictive system. The company told the Times that it did not believe there was sufficient evidence that the technology would be effective and that its existing responsible-gaming system was a better approach.
DraftKings says it already monitors more than two dozen indicators of potentially risky gambling behavior and can intervene with warnings, educational materials, questionnaires and, in some circumstances, account closures.
That distinction deserves more attention than the scary prospect of AI-fueled gambling.
As in other industries, the problem isn’t artificial intelligence. Rather, it’s what a company asks the technology to optimize. If the report is found to be true by state regulatory bodies and lawmakers, there are enforcement mechanisms and appropriate fines & penalties.
But, but, but: Most gamblers (still) aren’t problem gamblers
This demands reiteration because gambling is a legal recreational activity enjoyed by millions of American adults.
The National Council on Problem Gambling estimates that roughly 85% of U.S. adults have gambled at least once, and about 60% have gambled during the past year. For most people, buying a lottery ticket, visiting a casino or betting on a football game does not develop into compulsive gambling. It is more often an occasional use of discretionary spending and consumers have a good time playing the odds.
NCPG estimates that roughly 1% of U.S. adults have a severe gambling problem. Problem gambling is therefore both serious and relatively concentrated. Regulatory action needs to remain focused on fair market behaviors concerning this small segment of users.
Two facts can be true at the same time.
- Treating tens of millions of recreational gamblers as though they cannot make their own decisions would be a poor response to the genuine harms suffered by the smaller population that cannot.
- Basic consumer sovereignty means adults should generally be free to decide how to spend their entertainment dollars, including on activities that involve risk. A night out at the movies for a family of three pushes close to $100, double the median sports bet placed by the average bettor.
Consumer choice also depends on something meaningful: a consumer capable of exercising agency. This is where the Times investigation raises a harder question.
Can personalization veer into exploitation?
There is nothing inherently nefarious about a company using customer data to personalize its marketing.
Netflix recommends movies. Amazon recommends products. Airlines offer frequent travelers upgrades. Your grocery store knows which cereal you’re likely to buy.
A sportsbook knowing that I enjoy betting on NFL games and offering me a promotion on Sunday morning isn’t fundamentally different.
But imagine the algorithm discovers something else.
Suppose a customer’s betting history tells a sportsbook that whenever he loses several bets in succession, a $50 promotional offer reliably causes him to deposit another $500 and continue gambling. Suppose his deposits are increasing, his sessions are getting longer and he’s repeatedly attempting to win back losses.
The same behavioral data can suddenly tell the company two things.
This customer is extremely profitable.
And:
This customer may be losing control.
That’s where ordinary personalized marketing and gambling diverge.
According to the Times, DraftKings developed an “elasticity” model that incorporated information including betting frequency, account balances and how much customers typically lost relative to how much they wagered. The model attempted to determine whether additional promotional spending would generate additional gambling revenue. Former employees told the Times that they worried the most commercially responsive customers could overlap with those most vulnerable to problem gambling. DraftKings disputes the characterization that it targets customers based on their losses, saying promotions are directed toward customers demonstrating sustained engagement.
The Times does not establish that highly profitable gamblers and problem gamblers are the same population. That’s an important limitation of the investigation.
A person can gamble frequently, respond enthusiastically to promotions and lose a considerable amount of money while remaining perfectly capable of affording those losses and choosing the entertainment.
Nor should an opaque algorithm automatically diagnose someone as a gambling addict and lock him out of a legal activity because a computer says he fits a pattern. That presents its own consumer-rights problem.
But there is an uncomfortable asymmetry here.
If behavioral data is considered sophisticated enough to determine that a particular customer is unusually responsive to an inducement to gamble, companies should face scrutiny over how the same information interacts with their responsible-gaming safeguards.
Don’t confuse consumer protection with prohibition
This is also where much of the broader debate over sports betting goes wrong.
The relevant choice isn’t necessarily between legalized gambling and nobody gambling.
Americans gambled on sports long before the Supreme Court cleared the way for states to legalize sports betting in 2018. Illegal bookmakers and offshore gambling sites remain readily accessible today.
Legal markets can be regulated in ways illegal ones cannot.
Licensed sportsbooks can be required to verify customers’ ages and identities. They can offer self-exclusion and cooling-off mechanisms. Regulators can enforce rules governing advertising and promotions. Consumers have legal recourse when winnings aren’t paid. Sportsbooks can share information with leagues and integrity-monitoring organizations to detect suspicious betting patterns and potential match-fixing.
A bookie or offshore gambling operation based outside the United States offers few comparable guarantees.
That’s why the debate prompted by the Times investigation should focus on making the legal market work better rather than relitigating prohibition.
A regulated market gives policymakers, consumers and companies somewhere to apply pressure when market participants behave badly. Companies can change their practices. Regulators can investigate them. Rules can evolve alongside technology.
The black market provides none of those advantages.
Give consumers the algorithm, too
There is another possibility that receives much less attention: use this technology to give consumers more information about themselves.
If a sportsbook can analyze thousands of data points to understand my gambling behavior, why shouldn’t I benefit from that intelligence?
Imagine opening a sportsbook and seeing not merely your account balance but a personalized dashboard showing how much you’ve deposited, wagered and lost over the past month; whether your gambling has increased significantly; how much of your play followed promotional offers; how your current activity compares with your historical behavior; and the limits you’ve chosen for yourself.
The purpose wouldn’t be for an algorithm to declare, “You are an addict.”
It would be to say, “Here is what you are actually doing.”
That is much more compatible with consumer choice.
People routinely underestimate subscriptions, screen time, food consumption and other repetitive behaviors. Gambling apps have an unusually detailed picture of their customers’ behavior. Giving customers access to that information could make their decisions more informed without transferring those decisions to regulators or algorithms.
DraftKings has already taken steps in this direction with responsible-gaming tools and individualized player statistics. The logical question raised by increasingly sophisticated behavioral modeling is how much further that consumer-facing technology can go.
The same principle should apply to controls. Consumers should be able to easily establish deposit limits, take cooling-off periods, opt out of promotional messages and exclude themselves from gambling altogether. Those choices should be simple to activate and difficult for a moment of temptation to undo.
That’s not paternalism. It’s giving consumers better tools to govern themselves.
AI isn’t the villain here
There is a temptation to turn every uncomfortable use of machine learning into another argument about the dangers of artificial intelligence.
That would miss the point.
AI didn’t invent gambling addiction. It didn’t invent targeted advertising. It didn’t invent casinos offering complimentary rooms, meals or chips to customers they wanted to keep gambling.
What AI changes is the precision and scale at which those decisions can be made.
That can make marketing more effective. It can also make consumer protection more effective.
The important question is therefore not whether DraftKings uses artificial intelligence. Of course a modern technology company analyzes its customers.
The question is what happens when that analysis identifies a customer who is both extremely profitable and exhibiting warning signs of problem gambling.
The Times has not demonstrated that every profitable gambler is vulnerable, nor that DraftKings’ abandoned predictive system would have accurately identified people suffering from gambling disorders. Those are significant caveats.
But its investigation does expose a legitimate tension worth addressing.
A company shouldn’t get to celebrate the extraordinary predictive power of its technology when determining whom to market to, only to suddenly discover the limitations of data science when the same behavioral information raises questions about consumer welfare.
Legal gambling isn’t going away. Neither is artificial intelligence.
The opportunity is to make both work better for consumers.
Let adults gamble. Keep betting inside transparent, competitive and regulated markets. Punish companies when they violate the rules. Give consumers meaningful control over promotions, deposits and their own data. And use the extraordinary analytical power now available to gambling companies to inform consumers, not merely to understand how much more money can be extracted from them.
Consumer choice doesn’t mean pretending addiction doesn’t exist.
It means protecting the conditions that make genuine choice possible.