
ELO Matchmaking
ELO matchmaking pairs players using an ELO rating: a number that predicts each side’s expected result from the rating gap, and moves after every match by the difference between that prediction and the actual result. It comes from chess, and games often adapt it for teams and new players.
Also called
Elo rating, Elo rating system, ELO system, Elo-based matchmaking
,
How Does ELO Matchmaking Work?
The Elo rating system was initially designed by Arpad Elo for chess, and games borrowed it, usually written “ELO”. It does two things with one number per player: it predicts the result of a match from the gap between two ratings, then corrects both ratings by how far the real result landed from that prediction.
Expected result: E = 1 / (1 + 10^((opponent’s rating − your rating) / 400)). Every 400 points multiplies the odds by 10, so equal ratings give 0.5.
Update: new rating = old rating + K × (S − E), where S is 1 for a win, 0.5 for a draw and 0 for a loss, and K caps how far one match can move a rating.
A worked example, with K = 20: a 1500 player meets a 1600 player, so their expected result is about 0.36. If the 1500 player wins, they gain 20 × (1 − 0.36) ≈ 13 points and their opponent loses the same. If the favourite wins instead, it gains only about 7. An upset moves ratings more than an expected result, which is how the system corrects itself. The diagram below applies the same rule to six players, each in a one-against-one match; teams and groups need a more complex approach, covered further down.

The starting number is arbitrary, often 1000 or 1500: only the gaps carry meaning. The matchmaker then uses the rating as one rule among several, alongside latency, party size and time in the queue.
Article's key insights
Short answer: Skill-based matchmaking (SBMM) puts players in matches with others of similar skill, rated on how they perform against everyone else in each game. Call of Duty still ranks connection quality (ping) first. When Activision loosened skill in a 2024 test, 90% of lower-skill players came back less often.
What Does an ELO Difference Mean in a Match?
Because the formula turns a gap into a prediction, every gap reads as an expected share of results for the higher-rated side:
ELO gap | Higher-rated side expected to score | Lower-rated side |
|---|---|---|
50 | About 57% | About 43% |
100 | About 64% | About 36% |
200 | About 76% | About 24% |
400 | About 91% | About 9% |
So at a 100 ELO gap, the favourite is expected to score about two points in three. Draws count as half a win, and in team games the gap is usually measured between team averages.
This is how to read a skill rule in a matchmaker. One sample competitive profile starts at a 50 ELO difference and widens to 150 after thirty seconds in the queue (a sample Unity matchmaker profile): roughly 57/43 at first, about 70/30 once it widens. Whether 70/30 is acceptable depends on the mode, and how tight to set the band is the skill-based matchmaking question.
How Do You Choose the K-Factor?
K sets how much one match can move a rating, and it trades speed for stability:
High K: a rating finds its level quickly, but one bad evening swings it, and matches built on it get noisier.
Low K: a rating holds steady, but it follows an improving player slowly, so they play a long stretch of matches that are too easy.
Activision describes the same tension for Call of Duty’s skill measure: it should be “capable of adapting to a player’s ever-changing performance quickly”, while “the overall skill distribution should remain accurate in all situations” (Activision, July 2024).
One published approach changes K with experience. Chess’s governing body uses K = 40 for a new player’s first 30 games (and for juniors rated under 2300), 20 below 2400, and 10 once a player reaches 2400, even if they drop back, with K × games capped at 700 per rating period (FIDE Rating Regulations, March 2024). Games rarely publish their values. Placement matches, which move a new player’s rating faster at first, apply the same idea, and rating systems that track their own confidence, covered under matchmaking rating, adjust this speed per player instead of in fixed steps.
Where Does ELO Fall Short in Multiplayer Games?
ELO was built for one player against another, with one result per match. Multiplayer games stretch it in three places:
Teams: the formula compares two sides, so a team’s rating is usually the average of its players. Activision lists that as a requirement for its own skill measure: “the average skill of multiple players is predictive of their combined skill” (Activision, July 2024). It holds less well when one player is far above or below their teammates.
One result for everyone: a win moves every teammate’s rating the same way, whatever each one did. Some games add performance on top: in Call of Duty Ranked Play, the visible points a player earns depend on “match performance” and “the margin of victory” as well as the win (Call of Duty blog, December 2024).
No measure of certainty: a new account and a veteran with the same rating look identical to ELO, though one is a guess. Systems such as Glicko-2 and TrueSkill track that uncertainty, and the matchmaking rating page covers how, along with parties and seasons.
None of this makes ELO the wrong start. It is light to compute in any backend, easy to explain, and a sound fit for one-against-one modes. Most extensions keep its core idea: predict the result, then correct by the difference from what happened.
A word from our sponsor (ourselves!)
Most lag comes from distance, not from code. Edgegap's Edge Cloud is a distributed network with up to 615+ locations available on demand, so each match runs at the best available location for its players. Independent replays of live studio traffic measured a 58% drop in average round-trip time compared with public cloud.
What Is the Difference Between ELO and a Visible Rank?
When players say “my ELO”, they often mean the rank or points they can see, not the number the matchmaker uses. Many games keep the two apart on purpose.
Call of Duty Ranked Play is a documented example. Its Skill Rating (SR) is “a visible representation of the player’s skill journey”, while Raw Skill is “a single value representing a player’s performance relative to the rest of the player population”. Matchmaking looks for similar Rank and SR but also weighs Raw Skill, and how much SR a player gains or loses “is closely linked to the player’s Raw Skill” (Activision, December 2024; a breakdown).
The split solves a design problem. A rating that measures skill accurately tends to move little once it settles, which feels unrewarding to climb. A visible rank can reset each season and reward progress, while the hidden rating keeps matches balanced. The same paper notes the visible number “is not always predictive of outcome”, which is why it isn’t the one used for matching.
Edgegap's Take (just our opinion, take it with a grain of salt!)
ELO Helps Balance a Match, but Latency Impacts Its Perceived Quality More
ELO is a useful tool: it is simple to run and turns a rating gap into an expected result. But a balanced match on a poor connection still feels unfair to the player who lags. Activision puts it bluntly: “Ping is King”, and connection “is the most critical and heavily weighted factor in the matchmaking process” (Activision, July 2024).
One check for your matchmaker: set the latency limit first, then let time in the queue widen the ELO gap before it raises that limit. A long wait should loosen skill-based matchmaking, not ping.
Edgegap’s matchmaker supports both as rules: a number-difference rule for ELO, and a latency rule built on ping beacons that drops locations above a maximum ping and groups players with similar latency, with expansions that relax each rule on a timer (Edgegap docs). The match’s server can then deploy near that group, from 615+ Edge Cloud locations (platform data, as of September 2026).
,










