Chasing the local optimum

The game business has become increasingly data-driven in recent years, and that's been both a blessing and a curse.

The blessings are easy to see; it's an obvious boon for developers to finally have insight into what their players are actually doing in their games. After decades of finishing games and hurling them into a largely silent void, the sudden wealth of data about what players do in games is a gift to developers wishing to hone their craft and improve their focus. From unit balance to level design to the broad question of where to concentrate development resources and efforts, the data available to game creators thanks to the ubiquity of connected gaming has been a huge step forward; not all developers are taking advantage of it to the fullest extent, but those who do have seen enormous advantages.

The curse, unfortunately, is just as easy to see. We've written about it several times in recent weeks, in reference to recent games whose failings, in some regards, can all be put down to the curse of data - to the kind of blinkered decision-making that can all too easily arise from having too much data and being too narrowly focused on optimising for that data. From the ongoing tumult over loot crates in Call of Duty and Battlefield to the sharp divide between player dissatisfaction and rising revenues resulting from NBA2k18's in-game currency system, almost every controversial design and business decision of the past year has its roots in how the industry has begun to interpret and use data to inform its monetisation strategies.

"We are seeing an avalanche of enormous, expensively developed titles targeting exactly the same audience"

There are broad philosophical points here regarding what this does to the games themselves - bolting a slot machine to progression, building in even subtle pay-to-win mechanisms, or stripping out the sense of earned achievement in favour of plugging coins into a virtual vending machine - which I think other writers have made excellently and hence will not dwell upon. The counter-argument to each of these is always the same; AAA games are more expensive to make than ever, the audience for them isn't growing like it used to, and that means that they need to find new ways to make more money in order to justify their existence.

It's a frustrating argument because, while it's true, it sidesteps the question of what's happening to players' experiences and enjoyment entirely, and that's a discussion that could and should be happening as well. What I'd like to talk about a little, then, isn't the philosophical argument around what's happening or its impact on game design, per se - it's a well-trodden argument and I don't think I have much to contribute, not least since I'm not a game designer. Instead, I'd like to talk about this from the perspective of data; to think about how the industry is using its newfound access to extensive data, and the mistakes and pitfalls along the way.

There's a key concept here that many of you probably encountered if you took statistics classes at university; the local maximum, or local optimum. This is a problem that emerges in optimisation (and what is the industry doing with all its data, if not, fundamentally, optimisation?) and relates to optimising something such that it's performing better than any other adjacent case or solution. That's fine in itself, but it can also mean that you're "trapped" in your optimised position, because no matter which direction you move, the adjacent cases are less optimised and thus less preferable than where you are now.

"The real problem arises when you've got a local optimum that half an industry are all simultaneously trying to capitalise upon"

This is a big problem if, in reality, there are much more optimal solutions that are very different to the one you've found; unless you're willing to accept "backsliding" from your current position and exploring more non-optimal solutions, you'll never find them. You could visualise this like a graph with a line that features some high peaks, but also some low bumps; a bubble that floated up into one of the bumps would be stuck there, even though there are much higher peaks nearby, because to get to a peak it would need to slide back down to a lower level first.

That's all a bit abstract, but it's an important concept - in statistics, of course, but also in business, or even in other fields such as political policy-making. In a business like games, what optimising into a local maximum can mean is that you end up laser-focused on the most profitable subset of your customers and get very, very good at selling to them. But in the process, you may be ignoring far greater opportunities that exist elsewhere, because discovering them would involve leaving this local optimum you're backed into, and striking out to experiment in markets and segments that may, initially at least, be far less profitable than the one you're in.

If you're a specialised business, this can be a healthy strategy under the right circumstances; finding something you're very good at and work to get better at it. There are a handful of game companies who have done precisely that to great effect, discovering a niche and spending years working to create the best and most successful products they possibly can within that niche. For those companies data really is a gift, allowing them to become better at the kind of ultra-specialised work they were already doing to begin with.

"I struggle to think of a point when more top publishers were engaged in trying to make barely distinguishable versions of the same game"

The real problem arises when you've got a local optimum that half an industry are all simultaneously trying to capitalise upon, and that's exactly what's happening in games right now. Several of the industry's biggest companies have spent the past decade using data to learn how to be as effective at possible at precisely the same task: extracting the maximum possible lifetime revenue out of an audience of young, relatively affluent males.

That's not exactly new, but as data has taken over the decision making process, they've narrowed their focus even further, honing in on key psychological traits. People who play competitively are easier to monetise; people with poor capacity for handling delayed gratification even easier still, and it's people who fall into those categories who are the laser-focused target audience for most of the industry's massive games right now.

This is not a criticism of that audience - plenty of people, myself included, were in the young, male, relatively affluent, competitive and not great with delayed gratification group at some point in our lives - but rather a criticism of the fact that we are seeing an avalanche of enormous, expensively developed titles targeting exactly the same audience, and trying to maximise lifetime revenues by exploiting exactly the same psychological weaknesses that are endemic to certain subsets of that audience. That's the local maximum the industry has optimised itself into; in an attempt to justify rising AAA dev costs, it has found itself devoting enormous energies to figuring out how to squeeze every last cent out of a fairly small minority of high-value customers.

"The implosion of a major firm that's over-extended on this grossly over-subscribed market looks ever more likely"

This is not and cannot be a long-term strategy; not just for the companies involved in this intense competition over a tiny sliver of the market, but for the industry as a whole. The fact that so many games are settling on loot crate mechanisms and their ilk is in part a sign of economic forces coming home to roost (the same ones that pushed mobile games towards free-to-play), but it's also a symptom of a lack of vision and intelligent risk-taking among companies that have become addicted to data and incapable of seeing its limitations in the decision-making process.

For all that the industry has suffered from sequel-itis and me-too-ism in game design for years, I struggle to think of a point when more top developers and publishers were engaged in trying to make barely distinguishable versions of the same game: a team-based online competitive shooter with a real money transaction system bolted somewhat ungracefully to its progression mechanisms.

The problem is that this kind of optimisation is self-reinforcing; the kind of changes being made to maximise revenue from the high spending groups may not be entirely objectionable to the rest of the existing player base (though they often are), but they are likely to be anathema to potential players from outside that group. Having settled upon this strategy precisely because slowing market growth was making AAA budgets untenable, companies now find themselves locked into approaches that deny any hope of a return to growth.

It's possible, of course, that this is just the curse of AAA; that a whole segment of the industry's most talented and creative people are going to find themselves trapped in a cycle of building increasingly exploitative games for an ever narrower audience of players - who, incidentally, will get more and more toxic and abusive as the pool becomes more concentrated, because that's an inevitable consequence of a business strategy that involves having one hand in players' wallets and the other patting them on the head and constantly reassuring them of what important, vital consumers they are.

Perhaps the exploration of new markets and the pursuit of industry growth is simply best left to other creators and companies, who'll know better how to avoid these unhealthy over-optimisations - who'll recognise when it's time to cast aside the spreadsheets and try new approaches. I hope that's not the case, however, because unless they learn to change track, the hole that some of the industry's biggest firms are digging into at the moment is one that not all of them will climb out of.

Even if the enormous revenues coming from such a small segment of the market are sustainable, having so many companies chasing the same consumers is not - and as the budgets and risks involved in that pursuit mount, the implosion of a major firm that's over-extended on this grossly over-subscribed market looks ever more likely.