Video Game Matchmaking: A Data-Driven Take from Blizzard

How Blizzard blended data science with deep customer empathy to build the matchmaking system for their video games?

What are some of the common attributes of a successful video game? The usual suspects tend to be a compelling storyline, stunning visuals, and a magic formula that get players “hooked” to the virtual experiences. As Nir Eyal explained persuasively in her empirical research into habit-forming products, a deftly crafted “addiction roadmap” comprised of triggers, actions, rewards, and user-side investments seems to be one magic formula behind successful video game titles[1]. For Blizzard, an Irvine-based entertainment company that has produced legendary titles such as StarCraft, World of Warcraft, and Diablo, however, there seems to be a fourth dimension to a successful video game: matchmaking between players. This blog reviews how Blizzard has injected elements of both arts and (data) sciences into their matchmaking system.

First of all, matchmaking systems are extremely important to any video game with some elements of peer-to-peer competition. For new players, a good matchmaking algorithm can effectively prevent them from being “bullied” by seasoned players who may use matchmaking loopholes to “farm” new players to pull some quick wins and to safely rank up at the expense of new player experiences. For existing/returning players, a good matchmaking algorithm will shield them away from getting bored of a “grindy” experience marked by predictable outcomes, uneventful playing styles, and a general lack of surprise elements or X-factors. In this sense, a good matchmaking system will not only drive top-of-the-funnel user acquisition but also ensure sustained engagement from a vibrant, sticky players community.

By nature, video game matchmaking systems tend to be highly data-intensive and quantitative. In a simplified model, players with similar matchmaking ratings (MMR) ought to be paired so that players with similar capabilities can play against each other and an average player within that MMR band can achieve roughly 50% win rate. This is indeed what Blizzard did for its famous Battle.net ladder system, a ranked reputation board for many of its popular titles. Then human nature struck. Players intentionally lose several games in a row to artificially drop themselves to a lower MMR band, where they could farm lower-skilled players to achieve a much higher win rate, hence the in-game benefits that came with each victorious game. The mandate that only players with similar MMRs could be paired further led to a stale environment where the thrill of getting paired against a much more creative, skillful, and challenging opponent was virtually non-existent. To tackle the issues of farming and staled matchmaking, Blizzard leveraged their ever-accumulating data asset that encapsulated a rich spectrum of player behavior and characteristics beyond the bare-bone win rates.  For example, Blizzard used statistical rules to segment Hearthstone (Blizzard’s popular online collectible card game) users into newer vs. older cohorts, while looking at a completely different dimension to segment users into “casual” vs. “competitive” cohorts by deepdiving into the in-game card collection and behavioral data of each user[2]. Blizzard then customized the matchmaking algorithms based on the characteristics of different player cohorts. For instance, the “objective function” for pairing new players might lean towards maximizing in-game content discovery and a relative high startup win rate; the “objective function” for grindy players with jaded appetite might lean towards optimizing freshness and creativity by matchmaking across different MMR bands. These experimental cohort-specific customizations in turn generated a new stream of data to inform the evolving player preferences and cohort trends in an dynamic feedback loop of learning.

The end results seem promising. Unlike other titles which tend to create a huge splash in the gaming community and then quickly faded away, Blizzard’s titles have been time-honored leaders in monthly active usage across multiple categories such as MMORPG, RTS, and CCG. Behind the scene is the ingenious combination of data analysis and deep customer empathy that has allowed Blizzard to “shapeshift” its matchmaking algorithms to keep the Timmy’s, Johnny’s, and the Spike’s “hooked.”[3]

 

[1]. Nir Eyal: The Psychology of Building Addictive Products (https://medium.com/startup-grind/nir-eyal-why-you-are-addicted-to-facebook-slack-pinterest-468a86eb562)

[2]. https://venturebeat.com/2017/10/09/hearthstone-boss-ben-brode-gives-me-new-insight-on-what-casual-means/

[3]. https://magic.wizards.com/en/articles/archive/making-magic/timmy-johnny-and-spike-2013-12-03

 

Previous:

WHOOP: The Data-Driven Fitness Wearable Lebron James & Michael Phelps Use

Next:

MedAware – Using AI to eliminate prescription errors

1 thought on “Video Game Matchmaking: A Data-Driven Take from Blizzard

  1. Interesting read. Thanks for posting. There are seem to be network effects at play here, given the more gamers there are playing, the more valuable the game is to me as a player. Given match making is prone to being manipulated, what could Blizzard do to prevent its games from losing losers and thus devaluing the game they’ve created?

Leave a comment