As a general concept, matchmaking is fundamentally simple.
It serves as a means to group players in online games, based on the experiences they are looking for, and the ability or skill level they bring. The motivation for matchmaking, of course, is about giving players a rewarding experience. Games are more enjoyable – and thus more engaging – if players are given just the right amount of challenge, or thrust into teams with a dynamic that suits their ability.
Robust, refined matchmaking isn’t just for the players’ sake, though. It equally matters to game studios and publishers because it directly impacts commercial and critical success. But the reality is that matchmaking models – many of which take the form of skill rating systems – can be wildly complicated and intricate. No approach is perfect, but getting it right is vital.
As such, in this article we’ll take an overview of the logic and theory of quality matchmaking, and provide a summary of much of the nuance. A thorough mathematical breakdown of every matchmaking model is far beyond the scope of any single article, but below you’ll find links and resources to set you on the path to a thorough comprehension of matchmaking, along with key takeaways to help you consider what approach will best suit your games and business.
A Brief History of Matchmaking Systems
Ranking and matchmaking systems have arguably existed at least as long as many organised sports. One could even loosely understand the likes of football leagues as perpetual matchmaking systems. But it is the world of chess that has had the most influence on video game matchmaking.
In fact, a single chess rating system has had so much influence over video game matchmaking that its name has entered the parlance of contemporary gamers; see ‘Elo hell’.
The ‘Elo rating system’ was invented by Arpad Elo, a Hungarian-American physics professor, and implemented as a matchmaking approach by the United States Chess Federation in 1960. At a fundamental level, it provides a means to suitably match players in zero sum games by considering their potential ability. Over the years, it has been tweaked to serve as a ranking foundation for numerous high-profile video games. In a relatively small number of cases the original Elo system is used. It is believed, for example, that PlayerUnknown’s Battlegrounds uses the pure Elo system.
Elo can equally be used in non-competitive contexts. Some form of Elo powers high-profile dating apps like Tinder, as well as the relevance of Google and Amazon recommendations. Elo even sat at the heart of the proto-Facebook website Facemash.
The Elo system considers that individuals’ skill and ability within a zero sum game is a variable. None of us play at a consistent ability. Elo builds a rating that tries to reflect the range that runs from a player at their best, and at their worst. But the system also endeavours to recognise that a lower rated player at their best still has some probability of beating a higher rated player on a good day. In Elo ranked chess games, a winning player would add points to their rating based on the predicted likelihood of their victory. At the same time, their losing rival would lose the equivalent number of points from their rating.
Essentially, Elo uses the likelihood of victory or loss to adjust ratings. Of course, mathematical systems aren’t best explained in prose, so here’s a relatively simple video by popular mathematician James Grime, going over the fundamentals.
A Note on Fairness and Retention
One could understand matchmaking as a pursuit of fairness, and there’s a good deal of truth in that interpretation. But ‘fairness’ can mean different things in different contexts. In the Elo system, a perfectly fair game would be one where two players with an identical chance of winning are paired.
However, matchmaking is not always purely about fairness. It is about giving players a rewarding experience, and we all like to win. Beyond quality of user experience, players that win more games are also more typically retained. As such, an ideal online matchmaking system might balance true fairness with concessions to user experience. Strategies such as pairing new players with bots (detailed below) or reducing the negative impact of a loss relative to the gain of a win offer fairly direct solutions.
Why Should You Care About Matchmaking?
– Quality matchmaking brings a better user experience
If players are appropriately challenged, they will continue to be engaged, retaining more, lifting the potential for greater lifetime value. A reliable rating system can be significantly influential over the lasting success of your game.
– Matchmaking can serve an onboarding function
Without a matchmaking system that considers the inevitably lower skill levels of new players, churn can be high. No new player is likely to enjoy high numbers of losses after being matched against more capable rivals. Players that find themselves losing consistently – or feel that they are inaccurately matched generally – will understandably feel less inclined to stick with a game. As such, developers need to build rating systems that make newcomers’ first matches satisfying and empowering.
– Matchmaking systems can make games less toxic. Matchmaking can be used to match players based on the likes of interests, age and history of online behaviour – both in competitive and non-competitive or social online games and experiences. You may primarily consider matchmaking as a means to serve the competitive elements of your game, but it can also bring the right kinds of players together.In some high-profile games such as League of Legends have created specific matchmaking pools for toxic players (players who are abusive or otherwise a negative influence on the gameplay experience), keeping them separate from the wider player community. Typically, players will not be told they have been segregated in such a way, but in matching them with other toxic players they’re unable to spoil the enjoyment of other players. In other instances cheaters are similar quietly matchmade in cheater pools.
Nine Ways to Deliver Better Matchmaking
Simple matchmaking matters most
Teams with limited resource, budget or rating system experience still need quality matchmaking. As such, focusing on a basic implementation will provide 80% of the value. The most nuanced skill-based matchmaking system is all for nothing if your matchmaking struggles to find matches, so consideration should be made on widening criteria over time such that players don’t wait hours… Long wait times for a suitable match can be harmful to engagement and retention in their own way. If you are building a matchmaking system with limited resources, focus on the fundamentals over the nuance.
Consider ‘off-the-shelf’ options
Related to the above, consider established ‘off-the-shelf’ skill rating systems; or taking inspiration from them. Most improve on Elo’s heavy bias towards two-player games. The public domain Glicko-2 is based on Elo, and has seen use in titles including Team Fortress 2, Counter-Strike: Global Offensive and Splatoon 2 (system documentation). Meanwhile, Microsoft’s TrueSkill 2 offers a Bayesian skill rating system broadly comparable to Elo (system documentation). TrueSkill 2 is based on TrueSkill, which Microsoft has thoroughly detailed. However, its use is reserved for in-house projects, or via a license.
No system is perfect. All systems need maintenance
Matchmaking systems are not consistent entities. Much like an economy, the relative value of ratings shifts and evolves, and inflation is very real phenomenon caused by the likes of influxes of new players. In the case of Elo and Elo-like systems, new players matched against other humans can feed up points to the better players, shifting the relative value of an individual’s rating. As more players join a game, the average rating may drift. If players leave for long periods and lose familiarity, they may return to a rating that no longer reflects their ability – a phenomenon known as ‘time decay’.
Therefore, constant monitoring and maintenance is required. For example, persistently tweak your matchmaking skill ranges and improvement rates such that the distribution of wins/losses between players is reduced, while keeping matchmaking times sustainable.
Player feedback is valuable and inconsistent
Related to the above, player feedback can provide a meaningful guide to maintenance. However, as seen with the Elo hell phenomenon, player understanding of the nuance of matchmaking systems does not reliably mirror reality. Therefore, when considering player insight, it is worth focusing on broad trends across their feedback, rather than qualms specific to individuals. If a large proportion of your community is waiting for long periods, struggling to find matches, or finding games too easy, look into those issues. Remember, though, that most players will not have a full grasp of the reality of matchmaking systems.
The Dunning-Kruger Effect describes the phenomenon where people tend to rate their own abilities as much higher than average, regardless of reality. League of Legends players, for example, typically rank themselves 150 points higher than their actual matchmaking rating (MMR).Listen to your players, but consider their input within the context of their understanding and perception being imprecise.
Matchmake new players with bots
When a new player debuts in a game, they will not have any reliable experience on which to base a rating. Equally, many matches need to be played for an individual to arrive at a meaningful rating.
And, of course, any new players require a welcoming onboarding system to maximise retention and engagement. Under regular matchmaking new players are unlikely to experience multiple losses as a result of being matched with more capable rivals, resulting in a negative early experience and likely churn.
Matching new players with bots for a set initial period will give them wins that lift the chances of retention. Additionally bots can be used to solve low liquidity (i.e.number of players in a matchmaking pool) in games with low concurrents, which is especially likely in new or soft launched games.
Games like Fortnite have successfully used bots in matchmaking – with some suggestion that players of all abilities may face bots in Epic’s battle royale behemoth. An alternative to using bots is using ‘ghost data’; that is, performance data captured from real players. A longstanding technique in offline – and now online – driving games, that is widely applicable to other titles.. Ghost data is used in Angry Birds 2’s Arena PvP feature, for example.
Pick what matters to your matchmaking system
It can be tempting to build extremely precise matchmaking systems. The logic is understandable; precise matches should elevate all the advantages a quality matchmaking system brings. However, precision brings complexity, which is vulnerable to cheating, manipulation and error. Resist adding too many factors to your matchmaking criteria. Additionally, the more ‘matchmaking buckets’ you have – such as ping rate, skill and ability, or reputation – the smaller the pool of ‘matchable’ players becomes, leading to longer wait times. Matching players with a close skill gap is key for enjoyment, engagement and retention in many games, but it varies from title to title; in some mobile titles ‘player device’ may be an important bucket to match comparable hardware performance. At the maintenance stage, use observational study to focus on what matters to your game. And as always monitor matchmaking time, win ratios and player sentiment.
Match you matchmaking with your gameplay
Extending the notion of targeting and focusing matchmaking systems, it is important that any system reflects the gameplay it serves. That sounds rather obvious, but in building a matchmaking system it can be easy to fall into the trap of focusing on player ability alone. Consider specific gameplay factors such as any upgrades a player may have, or the potential for ‘home advantage’ (a user entering the map they play most). Meanwhile, a multiplayer PvE game such as Tony Hawk’s Pro Skater 1 + 2 would need to match differently from a team-based PvP game. It may even be possible that in some social games skill rating is of trivial value compared to matching players based on reputation, interests or demographic profile.
Track match data to inform future matchmaking
An unfortunate shortcoming of a capable matchmaking system is that it may match players so successfully that they keep meeting the same rivals, or being paired with the same kind of team experiences and gameplay dynamics. Establish a means to track the likes of players met, maps experienced on so on. Use matchmaking to deliver a diversity of gameplay experiences in tandem with appropriate ability pairings.
Player-facing vs back-end ratings
Showing players the full detail of their individual Elo rating is rarely wise. The same is true of the nuance of a game’s matchmaking system. Revealing the full criteria of that system makes it open to manipulation, while providing individuals with all the data on the various criteria that can make up their rating can confuse, mislead or even anger many players, and reduce trust that matchmaking is precise or fair. Angry, confused players that lack trust can have a damaging effect on retention, engagement and reputation. Instead of exposing players to full rating data, represent that rating through simple systems like ranks, or even symbols that denote the likes of bronze, silver, gold or platinum medals. Giving players that simplified data can also help with enabling peacocking, where player engagement and retention can be lifted by giving them means to assert their experience and status in game matches.
As noted previously, this article only intends to provide an overview of key considerations when developing a matchmaking or rating system, including some that are commonly missed or misunderstood. It has been written to provide inspiration points that should set you on the path to building a better matchmaking system, with a view to the many gains that brings, from quality of user experience to improved engagement, retention and success.
If we’ve given you the appetite to learn more about the mathematics and theory behind some of these core concepts, we’d recommend Twitch Senior Software Engineer Mario Izquierdo’s GDC talk, which addresses the process of building your own system as an alternative or extension to the Elo system.
Matchmaking does remain fundamentally simple, but when intersecting with real humans and the complexity of games it becomes incredibly chaotic. Getting matchmaking right takes considerable thought, planning and effort. Stick with it, though, and you’ll find the gains entirely justify the effort. Like chess, mastery of matchmaking system development is an ongoing education, and there is always room for improvement.
If you’d like to explore ways to improve your matchmaking, or refine it to optimise engagement and retention, here at Department of Play there are numerous ways we can help you. If you’d like to discuss that further – or talk about the many ways we can help your game projects succeed – do get in touch.