What Does it Take to Win in War Robots?

What are the essential elements for success in War Robots? It’s not as simple a question as it sounds.

One answer is surely in the gear. If you have better bots and weapons than your opposition, then you are likelier to prevail in a conflict. No great insight there.

Another is clearly player skill. War Robots is a simple game- but not an easy one. You field five robots with equipped weapons, and knowing which ones to take and which to leave in the hangar involves making decisions where skill and experience will tell. Not only that, but being able to move your bots around the battlefield effectively takes skill. Knowing how (and when) to discharge your weapons, that takes skill too. (And lest anyone take umbrage at the term ‘simple,’ it’s worth pointing out that chess is a simple game too.)

But beyond that, what if we turned to experience as our guide? I don’t mean the anecdotal experience of our own impressions, which can be wildly subject to emotion and perception. Rather, what if we looked on a larger scale, where we could get a sense of what factors appeared to contribute to victory by studying a large sample of post-match data.

Like, just over two hundred large.


Let’s go ahead and get this out of the way first.


Looking at the post-match screen, there are three main factors to victory: Kills, Damage, and Beacons. Without the ability to review each match in detail, we must be content with making inferences based on correlation and distribution. But let’s set the table so that the data to follow makes sense to everyone. Otherwise, what’s the point?

TechTarget has a definition of correlation that’s delightfully simple: “Correlation is a statistical measure that indicates the extent to which two or more variables fluctuate together.” Causation, meanwhile, is the notion that one of these variables is the cause of the correlative fluctuation in the other one.

Or we can skip the word salad and say it this way.

Causal correlation: more people use swimming pools the hotter it is (one thing influences the other)

Non-causal correlation: global warming and piracy (one thing has jack diddle to do with causing the other, but they happen to correlate)

Why is it so important to bear this in mind? That’s because despite the data we have at our disposal, we cannot conclusively point to a single factor as having caused victory. Let’s look at an end-of-game summary, from any ol’ game on a Monday evening.

If you picked up on a correlation between “standings in the table” and “Jay’s likelihood to take a screenshot,” you just got extra credit

How did we win? The damage tally looks fairly even. They killed more bots, but we got more beacons. You might conclude this was a beacon win, but that doesn’t necessarily tell the whole story. Did our damage have an impact? Our seventeen kills?

In truth, victory on the battlefield is always a result of the confluence of multiple variables, and without being able to watch replays or get perfect data from a match, we can’t conclusively say why the match was won. What we can do, however, is see how different factors correlate with victory. If every time a team won they had more beacons than the opponent, we could say that beacons have a high degree of correlation with victory. Although this isn’t true (you can win with a lower beacon count), it doesn’t mean we can’t sift for relationships.


I asked the Wiki Forum community for screenshots of matches, and boy did they deliver! Special thanks to: Berzerka, Thunderkiss, GuitarGuy, drake1588, dangit, tricky48, mijapi300, BLACKPAYNE, Paps, SoCalGrindr, maxboxer, pillowofdoom, Russel, P Waggin’, oakwhelie, zer00eyz, Firebeard, zephyr1138, piginapoke, DarkVagabond, nocluevok, ogpondskum, inspirace, adrenachrome, mech, acdcfan, bugdoc, wrongtarget, indricotherium, procrastinatron, Heavy Panda, seanh, justsomeguy, and snk. Apologies if I missed anyone!

Also, extra special thanks to two of my mates from the Aurora Nova family of clans, Ubiond and mmarius. Without their selfless and generous contribution of time spent logging the data into spreadsheets, this would have had a much, much shallower sample pool.

Now, let’s get to the data!


In a way, kills are to War Robots statistics what Gold is to ingame advancement. Both are often latched on as being the most important thing early in the game by newer players, and both are deceptive. Eventually, you come to realize that individual kills don’t matter as much as damage, and that Silver, not Gold, is the game’s most valuable currency.

Nevertheless, damaged bots shoot back, and dead ones don’t. Not only that, but killing (up to) thirty of your opponent’s bots is a certain win condition. kills have value, even if that may be occasionally overstated.

So how do kills look? For this, average is somewhat useless because the number of kills is contingent upon the number of slots players in a game have. The kills in a Recruit-level game will be lower than Expert, simply because there are less bots available to kill. But we can compare kills by the winners and losers to determine a rate.

The average kills for a winning team in our data was 19.11 per game, while losers only managed 15.78. That means that winners killed 1.21 bots for every one the losers killed, or six for every five. That makes sense- if you kill more than the other side, you’re likely to win. But how likely? After all, you can kill more and still lose the game. Let’s go deeper.

Fig. 1: Kill ratios

On the left, we see the kill ratios. These represent the answer to the question, “for every bot of ours they killed, how many of theirs did we kill?” Only just under 20% of winning sides managed to secure victory by killing less than their opponents, so killing your opponent’s forces has a high degree of relationship with victory. Not exactly “stop the presses” stuff, but it’s nice to have some numbers behind it.

Put another way, if you can’t at least match your opponents’ kill rate, you will likely win only one game in five. It’s clobberin’ time!


Fig. 2: Damage ratios

Now let’s take a look at damage. A common misconception from novice players is that every point of damage is equal. This is sort of true, but only in the Orwellian sense of, “all damage is equal, but some damage is more equal than others.” There is quite a difference between points of damage inflicted by a Thunder in the trenches of Power Plant versus a Trebuchet up on the hill, in that the latter is damage in its purest sense. Ranged weapons and artillery can put up good numbers, but they do that to the exclusion of most any other tactical consideration or objective.

The data tells an interesting story here. Teams that do at least as much damage as their opponent manage to win 74.13% of the time. This is a worse rate of return than we saw for kills, so we can infer that damage is a bit less of a predictor of outcomes than actually finishing bots off. This validates the conventional wisdom of the community regarding the diminished value of raw damage.

Note also that the single biggest category in both tables is parity, the 1.00 value of approximate equality. You often hear talk of “blowout games” from an “unfair matchmaker,” but the data we collected doesn’t appear to bear that out. Blowout games happen, but they’re comparatively rare. Of course, these games tend to punch above their weight in terms of their emotional impact which is why we seem to think they loom larger than they do. This is something we also saw with our research on the prevalence of clubbing.


Fig. 3: Beacon ratios

Finally, let’s take a look at beacons. Veteran players will tell you, “it’s all about the beacons, dummy,” but are they right?

Again we see a high correlation of beacon capture with victory, but this data is a little different than what we’ve seen thus far. With both kills and damage, the most populous group was parity (an even 1:1 ratio), and by a healthy margin. Here, curiously, parity isn’t quite enough. Rather, the single most successful outcome was to get 25% more beacons than the opponent. Teams that were unable to accomplish this failed to win even half of the time. This is also true of kills, but there it was a hair away from being a true coin flip (49.75%).

This seems to indicate that you can get away with less kills and less damage to a larger degree than you can beacons. Of course, it must be noted that none of these three things exist in a vacuum. Damage results in kills, things that are killed cease doing damage, and beacons are the focal points of interaction between enemies (and thus where most damage tends to get done).

In terms of parity-or-more, 77.11% of teams who can at least attain parity see a win at the end. This slots in nicely between the same values for kills (80.10%) and damage (74.13%), which could offer some insight on impact priority. Given how casually kill tallies are dismissed, there’s a touch of irony present. But these are fundamentally two different things.

The kills stat that gets dismissed is the individual one, largely because it is a poor indicator of battlefield performance. If you manage to pick off a few enemies that are on their last legs, you are surely doing your team a service, but they (arguably) did the bulk of the heavy lifting getting you to that point. Imagine a center-half in soccer taking the ball all the way upfield, dishing off at the last moment and a forward slots home a sitter. Goals are crucial, so a team won’t really care where it came from, but if you want to get down to it who made the play happen? Ultimately, we want to look at kills through the lens of the team, not an individual.

Put another way, when you realize each kill immediately reduces your opponent’s presence on the battlefield by at least 16.67%, and overall threat depth by at least 3.33%, kills take on a much greater significance than simply a false proxy for battlefield prowess.

Finally, to scratch the itch of curiosity, how did average beacons stack up? Of the three metrics looked at today, this one may be the cleanest. Average damage is useless, given the vast disparity of damage tallies when comparing lower-league games to higher-league ones. Average kills is more useful, but again subject to some variation given the likelihood for more slots having been unlocked the higher up in leagues you go.

But beacons? They stay the same game to game, every game. Whether Recruit or Legend, players are fighting over the same beacons, and their capacity for capturing them is about equal. Sure your leveled-up bots will have higher speeds, but lower-level players are also more likely to be playing the faster Light bots, so it’s probably something of a push.

The answer is, in our sample the winning team averaged 10.65 beacons, the losing side only 8.83.


If you’re in the Chevy Chase, “Uh, I Was told there’d be no math” camp, you might want to give this section a miss. However, for those genuinely intrigued by math, this last segment will look at the statistical correlation between each of our three sides of the triangle.

Here’s the TLDR: Statistical correlation assigns a value to the degree of correlation between two sets of data. Some data has a “loose correlation,” meaning they’re only somewhat in step with another another. Others have a much higher correlation, like the heat index and swimming pool users example I used above.

Here’s a superb explainer from Dummies.com that shows what the numbers mean:

Img 1: Correlational coefficients explained

By the way, go ahead and ignore negative relationships, they don’t apply here. If you were wondering what they are, it’s when one set of values going up relates to another set of values going down. The example of heat index and swimming pool attendance are both numbers going up. For a negative relationship, think instead about temperature and articles of clothing. The more the temperature decreases, the more the number of articles of clothing people wear outside increase. That’s all.

Let’s start with the obvious one, Kills and Damage. Turns out these two march pretty well in unison, with a coefficient of 89.47. You do more damage, you get more kills. New hawt tech revealed! Definitely an example of the kind of hard-hitting analysis you come here for.

What about Kills and Beacons, any connection there? I mean, we all do a lot of fighting trying to flip red things to blue, right? Turns out, not so much, a coefficient of 38.17 or a “weak” relationship. People die all over the board, and folks bemoaning “damage whores” who ignore beacons are common.

Any different with Beacons and Damage? Nope, it’s even worse- but not by much: 34.66. This is certainly impacted by all you Cossack and Stalker pilots out there who make a career out of a “love ’em and leave ’em” approach to the battlefield.

These values are all from the winning side of the occasion. For the sake of thoroughness, I ran the same process through for the losers. They saw a slightly higher correlation between Kills and Beacons (possibly reflective of what hills they choose to die on on their way to inglorious defeat), but otherwise were right in line.


In one sense, this didn’t tell us anything groundbreakingly new. We all know that more Kills, more Damage, and more Beacons are how you win the game. But often it’s helpful to get under the hood and understand what that means. For instance, learning that attaining equal or less Kills per game only results in victory about half the time, but teams getting equal or less Damage nevertheless get the win 62.19% of the time lends itself to some interesting observations about the nature of attrition in War Robots.

Similarly, the fact that teams that only manage to get 75% of the Damage that their opponents inflict still win almost a quarter of their matches. Kills (11.94%) and Beacons (17.91%) are much less forgiving. So here are a few possible conclusions that can be drawn from today’s data.

Damage is a selfish statistic. Since Damage is how the post-game standings are listed- and therefore, how Gold rewards are distributed- it tends to loom larger than the actual impact it has on the game. In fact, it has the least impact on the game of any of the three factors we assessed today. You can win more often by doing less Damage than you can by nabbing less Beacons or having less Kills.

Now, here’s an interesting question. Do you skip the Raijins? I mean, some see a Raijin (or Leo, or whatever) as a free “bag of Silver,” but is there a tactical advantage in prioritizing lower-health targets over higher-health ones? There’s certainly a case to be made.

Faster Pussycat! Kill! Kill! This should put to bed the notion of “kill-stealing” in the game. Kills are so important that if you’re keeping pace with the opponent, then the outcome is something of a coin flip. Obviously, Kills don’t exist in a vacuum- if your opponent is getting Beacons in the process and your team is ignoring them, your odds of victory will go down. But you should do everything you can to eliminate your opponent, and if this means risking someone getting upset you stole “their” kill, who cares? You’re in it to win it, and dead mech bodies win battles.

Beacons are beguiling. Looking at the data for Beacons, you almost wonder if there’s something of a divergence. If you get less than or equal to your opponent’s haul, you’re still only going to win 45.27% of the time. This is the worst correlation with victory of any of the three. In addition, as pointed out above Beacons are the only aspect that has the greatest results above 1.0. It’s not a huge difference, but a 1.25 coefficient has the best chance of victory.

What does this mean? Again, given our imperfect data and the overlap of the three different variables it’s impossible to draw an ironclad conclusion. But it does seem to indicate that if you’re going to go for Beacons, you really need to prioritize them. Otherwise, your best bet is to simply snuff out your opponent. I’ve had this decision point in many games, where I have to consider if a sort of “blitzkrieg” can carry the day, or do I need to leave my allies to go flip a distant Beacon.

There’s risks to both. Ignore Beacons, and you’ll lose games. Period. But I’ve had times where I’ve run off to go cap a Beacon, but come to feel that with a rush and a push we might have carried the day through attrition. Does that mean the game is bimodal, where your best bet is to optimize for one objective (Beacons) or the other (attrition) at the start of the match? Probably not. But it will convey some advantage to those adept at reading the battlefield, and knowing when to flip from one mode to the other.

As for those that forego that kind of situational awareness and would rather just rack up the damage?

The data’s not on their side.

Thanks for reading.


11 thoughts on “What Does it Take to Win in War Robots?

  1. I liked it. Maybe the statistic we need, that doesnt exist, is how long the beacon your capture is held? Do you get the beacon and die? Or does a strong force hold it for a time.
    Thanks for all your work.


  2. Nice article… wish you had found something more iron clad, but thanks for eliminating some questions… it would be awesome if Pix would open up the data vaults for more evaluation…


  3. Does it mean would be a good suggestion to Pixonic in reviewing the rating league points system at the end of the battle? Since that is based sotrongly on damages and beacons. Non mentioned kills.


  4. Nice article, it would help me a lot in battle tactics. Thanks for all the data! I’ll definitively share these informations in my clan. I hope that Pixonic plans to add a “battle replay” log, so that it replays the action in simplified version of the battle on a minimap with bot icons. That would allow stratergic reviews.


  5. >>As for those that forego that kind of situational awareness and would rather just rack up the damage?
    >>The data’s not on their side.

    Don’t forget the almighty Ruble. The fastest way to grind Silver is to focus on dealing damage, winning isn’t the primary objective; and at 5k per beacon, it is not profitable to play for captures.


  6. This was the conclusion that i came to independently over the last month or so, having gone beacons all the way to gold 1 until it punished me with mechouts (the inevitable consequence of being too focused on capping). Often much more succesful to wait for the big push


    1. Before that many players in the community where inferring on of the three as most important. Based on just their feeling. But it was only a feeling. Now we have a scientific answer


  7. I’ve never cared for the beacon metric. its not switching beacons that wins games, its holding them – something we don’t get any of the end-game metrics on (other than the binary which side won). The trouble with looking at how beacon switching correlates to win conditions it that its a non-causal correlation. Kills obviously degrade an opponents ability to hold beacons (and are their own win condition in the case of a mech-out).

    Damage is a weak indicator because Shields and different mech sizes (with different health pools) distort how much degradation the enemy actually suffered per unit of damage.

    Beacon switching as an indicator for winning is basically useless. If the ratio is more than 1.5 you won because you physically held more beacons. This only happens in about 1/4 of your data set, but doesn’t indicate a fall-off in “really going for beacons” being productive. It’s indicative of the fact that once you’ve hit the initial 3:2 equilibrium your ratio falls with each successive beacon switch. 1.5 (3:2) becomes 1.0 (3:3) becomes 1.33 (4:3) becomes 1.0 (4:4) becomes 1.25 (5:4) and so on. The harder you strive for and contest beacons, the more your ratio approaches 1.0. The other thing I’ll point out is evaluating beacon switching from an “equal to or less” basis understates the value of the “equal to.”. If you cap a beacon and then I take it from you, we have an equal ratio 1.0 but I win if the beacons are static and the game is less than half over.

    If I were of a conclusion drawing bent, I’d say not enough games are decided by mech-out to warrent ignoring beacons, but if you were inclined to give up on beacon and go a mech-out, the time to do make that switch is at half-time.


  8. Thank you for such an interesting analysis! Your data driven approach is truly illuminating, but it has only just begun to whet my appetite. Is it possible for you to post publicly or share privately the spreadsheet data that you and your team worked so hard on gathering and encoding? I would be very interested to run additional numerical analysis to look at other WR questions not covered here (e.g. win correlation with hanger strength or player rating). I would be happy to report findings to you for future articles, or post them here in comments, or on forums if you prefer.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s