Teamfight Statistics

March 07th
written by Barroi


The newest addition to Winston’s Lab’s reservoir of statistics are teamfight stats (for advanced data aquisition: fights by hero). These numbers have the potential to showcase the prowess of teams’ ultimate management (ult economy), the importance of first blood, the value of the ultimate from each hero and more. This article gives examples of some of the most interesting teamfight stats and tries to explain how to interpret and use the presented numbers.

For clarity, the presented statistics encapsulate data for LAN and online matches. An analysis on LAN-only stats can be done at a later date, once there is more data. For our purposes online data should not vary too much to LAN data, although there are some aspects (like ultimate effectiveness for specific heroes), which would benefit from a LAN environment. Additionally, only data from professionel 6v6 matches was used.

What is a “fight”?

After analyzing multiple hours of gameplay with a focus on teamfights I came to the conclusion to use the following parameters to determine when a situation should be considered a “fight”.

A fight starts as soon as one player dies.
For two consecutive kills to be considered part of the same fight the second kill has to occur no longer than 14 seconds later than the first kill. Killchains are part of the same fight as long as no two consecutive kills are seperated by more than 14 seconds. A kill happening later than 14 seconds after the previous one is part of the next fight.
For an ultimate to be considered part of the fight it must be used maximally 12 seconds before the first kill and maximally 0 seconds after the last kill was accomplished.

A team ‘wins’ a fight if it gets more kills than their opponent.
The start of a fight is the minimum of the first kill and the first ultimate from that fight. The end of a fight is the time of the last kill. Fights can, in theory, last forever.

Example for a fight (with timestamps):
(0:13) Ult not part of the fight
(0:18) Ult part of the fight
(0:29) Ult part of fight
(0:30) Kill
(0:30) Kill
(0:40) Kill
(0:54) Kill
(1:08) Kill
(1:10) Ult
(1:12) Kill (last kill of this fight)
(1:13) Ult not part of any fight
(1:20) Ult part of next fight
(1:30) Kill (start of the next fight) …

The Importance of First Blood

Karahol previously analyzed if there is a correlation between drawing first blood and winning the map. The sobering conclusion was that if there is a correlation it is negligibly small. In almost precisely 50% of the analyzed games getting the first kill lead to winning the round.

The role of first blood is a lot more meaningful when observed in teamfights. From the 11596 fights in our database, in 77.17% of them the team that drew first blood won the entire fight. When checking only “big teamfights” (those where at least 3 players die) the percentage is lower, but still at 74.78% (of 9373 fights).

Since “big teamfights” eliminate the fights where teams lose a few players and then back off, this number is more reliable in determining how powerful the first kill really is. The astonishing result is that almost 3 out of 4 fights are decided by the very first casualty.

Average ("big") teamfight stats
Average (“big”) teamfight stats – corresponding WL page

If not mentioned otherwise the following analyses will all use “big teamfights” for data acquisition.

Whose Death Is The Most Significant

Now, how important is it for specific heroes to not be the first victim of a fight? When thinking about this one has to come to the conclusion that it definitely decreases your chances to win a fight if one of your support heroes dies first. At the same time, it should not be as bad when one of your tanks is the first to fall. But does the data back those basic thoughts up and if so to which extent?

Since the data for some heroes is extremely limited due to their low pick rates, we will ignore some characters for this analysis. Their are 7 heroes with less than 60 data points (fights where this hero died first) and thus will not be included here. Those heroes are: Junkrat, Bastion, Hanzo, Torbjörn, Symmetra, Widowmaker and Sombra. Most of the other heroes have above 250 data points which makes analyses on them extremely precise.

amount of First Deaths and First Death-Win rates
amount of First Deaths and First Death-Winrates

There are four heroes for whom the chance of winning a fight sees a significant drop (by more than 2% compared to the average) if they are the first to die. Those most important heroes for a fight are Mccree, Lucio, Zenyatta and Soldier: 76. Two characters’ deaths are significantly less impactful (more than 2% increase on FD-Winrate) on the outcome of a fight. If Tracer gets send to the graveyard her teams’ chance to win the teamfight is at 24%, if Winston is the first casualty it’s even 28%.

Winston’s difference to the average is at almost 7%, making him the only outlier at a above 5% deviation. Does this mean Winston is totally useless in teamfights? No, as we will see later there are even more flaws in regards to his ultimate, but this is no sign for him being useless. In fact, because he is most used in dive compositions one could argue that by the time he dies his Genji might be perfectly set up to get 2 dash kills in a row.

We also still have to wait and see where Sombra and Symmetra end up. Right now, they look to be even more expendable than Winston and there are some obvious reasons that would support that hypothesis. Sombra is currently often used as the initiator, using her ult at the start of the fight and has thus done her part. Meanwhile Symmetra hopefully already set up her teleporter or shield generator by the time she dies. But for now their data is too small to draw conclusions.

Looking at the bigger picture of that data our original theories of supports being less expendable than tanks in regards to teamfights seem to be proven. Although, we have to mention that Ana’s FD-Winrate is above average which is a bit weird, because she is probably the best single target healer and thus essential to winning the fight. A possible cause for this is that she dying first induces that the opponents are running a dive comp, which was in general less successful in the past. Further analysis regarding team compositions and fights has to be done to look into that.

Another slightly unexpected result to me was the high importance of almost all DPS heroes. I expected there to be an almost obvious distinction between supports (except for Sym) and DPS heroes, but looking back I can see that my mistake lied in disregarding the fact that there are many possibilities to one-shot heroes, rendering healing worthless in that situation. Flame made a great video explaining why the balance between healing and dealing damage is important and stating the imbalance at the time as the reason for the tank meta. Since I could not find that specific video I will link to his YouTube channel.

How many fights do you need to win to finish a map

Next we will look at teamfight statistics by maps. In particular how many fights are fought on maps that get finished and how many fights need to be won to finish that map. These statistics are very straight forward and don’t need much explanation, so I will only briefly go over them. The full table can be seen at the end of this section – you can click the image to increase its size.

To finish an escort map you need to win roughly 5 fights and provided you get to your checkpoints in time you have about 9 fights to do that. The number of won fights that are needed to capture points A and B will be provided in a later analysis. Winston’s Lab is working on their database structure in order to provide insight to those stats in the future.

For Assault maps, the number of finished Hanamura and Volskaya maps is not really enough to include them into this analysis. In the picture at the end of this section the maps with insufficient data are written in gray. As one can see it takes about 2.5 won fights to finish the Temple of Anubis. On the first impression this might seem weird, because Assault maps only have two points to capture and in theory only one fight per point is needed. The cause for this is easily explained, though. A fight is considered “won” if one team gets more kills than the other, meaning that in one of two finished Anubis maps one fight gets “won” by the attackers in terms of kills, but they actually have to back of off the point in the end.

An interesting result when comparing the different map types in terms of how many fights are needed to finish it, is the following. On Escort maps the attacking team has to win 57.8%, on Hybrid 58.5% of the fights in order to finish it. On Assault this number is only at 44.5%. The reason is simply the design of the maps being in favor of one side (Defense/Attack). This might not be a huge eye opener for most people, but it at least proves that it is in fact easier to Defend on Escort/Hybrid maps, while the Attacking side on Asssault has an easier time. Those claim are also confirmed by the percentage of finished rounds in correlation to the total number of played rounds. For the average Assault this number is at about 55% while only ~27% of Escort/Hybrid maps get finished.

Since you can not “finish” a Control map like you can finish a map of another map type I’ll just leave the average of fights per round here without further discussing them.

Fights needed to finish a map
Fights needed to finish a map – maps with insufficient data are gray

The Importance of Ultimates

How important is it to have more ults than your opponent? Our 9k teamfights show that the team that uses more ults wins in 68% of all fights. This completely ignores which ults are used implying how incredibly powerful most if not all ultimate abilities in Overwatch really are. At some point in the future I will create a weighting of ultimates (by using the results of the next section) to construct statistics that will make comparisons of specific teams’ ult efficiency easier and more insightful.

For now it is obvious that using less ultimates puts you at a disadvantage, you only have a 28% chance to win the teamfight then.

FMU = Fights where team used More Ults | FSU = Fights where both teams used the Same amount of Ults | FLU = Fights Less Ults
FMU = Fights where team used More Ults | FSU = Fights where both teams used the Same amount of Ults | FLU = Fights Less Ults

The Value of each Ultimate

On our new fights by hero page there is a big variety of stats on ultimates. It takes a lot of effort to get into all of it, because it mainly exists for data aquisition purposes so I don’t recommend it to anyone who isn’t willing to invest multiple hours into it. But there are some amazing conclusions that can be drawn out of those stats.

One of them focusses on how valueable each heroes’ ultimate ability is. For this purpose I introduced FSUTHU stats. FSUTHU are the “Fights where both teams use the Same amount of Ults and This Hero uses his Ult” (short: Fights Same Ults This Hero Ults). The special aspect of this statistic is that fights where the considered hero uses his ult on both teams are ignored. So it counts only fights where either only one team has that hero or where both teams play it but only one of them uses this ult. This is a significant factor, because if both teams use a heroes’ ult it would result in adding one lost and one won fight, which nullify each other and thus skew the overall outcome.

the value of each heroes' ultimate ability
the value of each heroes’ ultimate ability

A high FSUTHU-Winrate means that whenever your team uses that heroes’ ult and the enemy doesn’t your chance to win the fight goes up. We will again ignore some heroes for whom there aren’t enough data points. This time eight heroes will get ignored, the same seven as above and Mercy. For all other heroes there are at least 80 and most of the time more than 200 data points, which makes using those stats reasonable.

The statistically most valueable ultimate is Zarya’s Graviton Surge followed by Reinhardt’s Earth Shatter. At this point I want to stress once again that this data can only be evaluated in regards to teamfights. Symmetra’s ult will probably end up being the least impactful because using her ult during a teamfight is far from ideal, since you want to have your TP or SG up before someone dies.

Most of the numbers in this table are not necessarily mind blowing, but having statistical prove of what is believed to be a known fact can’t hurt. The most surprising results to me are the low percentages of Tracer and Winston. Winston’s and D.Va’s numbers are actually extremely bad throughout all stats on that page. One could almost conclude from them that using their ultimates is decremental to the teams chances of winning the fight. For D.Va this does make sense, because you don’t want her mech to be destroyed, but for Winston this might suggest that there aren’t any benefits from using his ult at all or that players still need to figure out when to use it to get the biggest benefits (Miro recently provided a great example on how to efficiently use Primal Rage).

Tracer’s low numbers are almost nullified by her insanely high charge rate giving her about 20% more ults than the average hero.

This assignment of value to each ultimate ability can be used in further analyses to deduce how good teams are in regards to using and managing their ults. It can also help in figuring out which play styles and team comps are the most successful. This data suggest that Nano-Boosting a Zarya to give her 29% more ults than the usual Zarya (like Meta Athena does with Hoon) might be the best way to play the game right now.

Ultimate Economy – Excursion

For a first indication of how good a team’s management of ultimates is there are two stats on our fight by team page. These numbers are very rough and have to be polished with further data regarding the used team compositions as well as the rate at which the teams charge ultimates. I still want to mention them here, not only because that polishment will take a lot more brainpower, but also to explain the current numbers and because it gives us slight indications on the correlation of ult economy and teamfight prowess.

picture of APEX query
Ult economy stats of APEX S2 teams

The most important new stats are MUW and LUL. MUW stands for “amount of More Ults used when Won” and is the average of the difference of this teams’ amount of used ults and the opponents’ amount of used ults. This difference gets ignored if the viewed team used less ults or did not win the teamfight. LUL is the “amount of Less Ults used when Lost”. Just like its counterpart it is the average of the difference of the amount of ults used by the opponent and the amount of ults used by the considered team. Additionally, those differences are only part of the data that makes the LUL if the considered team used less ults and lost the fight.

A team wants to minimize its MUW while not lowering its FMU-Winrate (Fights with More Ults) to waste as few ultimates as possible. It should also try to maximize its LUL while not lowering its FLU-Winrate (Fights with Less Ults) to increase the amount of ults the enemy wastes. Because it is hard to directly influence the enemies ult management, one should at least try to have a MUW that is lower than their LUL, so that you “waste” less ults than the opponent does when winning a fight.

Now lets take a look at the numbers: Here is a query of all APEX Season 2 matches (sort it by LUL-MUW). What immediately becomes apparent is that all MUW stats are above 1, the reason for that is simple: To have used more ults than your opponent you must have used at least one ultimate more. As is observable KongDoo Uncia has the second lowest MUW suggesting that they are very smart in regards to knowing how many ultimates they have to use to win a fight. Lunatic-Hai does not have a great MUW, but they make their enemies use a lot of their ults when LH loses a fight, which rockets their LUL and thus the LUL-MUW.

This small introduction is everything to ult economy for now. Again those numbers are yet to be further analyzed and (if needed) polished to become really meaningful and as of now they should only be used in combination with a lot of knowledge to prevent overvaluation.


Barroi is the founder of Winston's Lab. He is coder, journalist and statistician at once.