They Bleed: How to Conquer SK Telecom T1
SK Telecom T1 arrived at MSI as gods and left as mere mortals. Somehow, unexpectedly, a European squad full of rookies and a supposedly inferior Chinese team were able to prove that the Koreans were less than immortal, that they could, in fact, bleed.
Coming into MSI, SKT's skin seemed impenetrable. They had captured the LCK crown handily, with a 3-0 sweep of the GE Tigers. The followed that with a 5-0 Round Robin, leading SKT's Faker to state that he was "eager to get [a] clean sweep". That didn't seem like an unrealistic goal, especially after SKT had beaten EDward Gaming so convincingly in their Round Robin matchup.
But when the dust settled after Game 5 of the MSI Finals, EDG stood victorious. And along the way, European underdogs Fnatic had also, very unexpectedly, managed to push SKT to a fifth game.
How did SKT's LCK and MSI playoff opponents manage to take seven games and the MSI championship away from them? The statistics below explore three keys to victory over SKT:
Get That Snowball Rolling
Push the Pace
Keep it Messy
Note: This analysis is based on statistical evidence and my personal perspective on watching SKT's games. I welcome alternate interpretations of, or explanations for, these stats.
The Korean champions showed some weakness in the very early stages of games, and their successful opponents worked to capitalize on those weaknesses as much as possible. In the LCK and MSI playoffs, SKT gave up 61% of First Bloods and 56% of first Towers. Through those 18 playoff games, SKT was behind, on average, by 0.5 kills and 134 gold at the 5 minute mark.
But more often than not, a 5-minute or 10-minute lead wasn't enough: SKT still won six of the eleven games where they were behind in gold at 5 minutes, and three of the nine games where they were behind at 10 minutes. The essential thing for SKT's opponents was to create an early-game snowball and turn their 5-minute leads into much bigger 15-minute and 20-minute leads. SKT only won a single game where they were down in gold at 15 minutes (Game 3 vs. Fnatic).
To be clear, the point is not that snowballing in the early game leads to wins. That's self-evident, and applies to every game of League of Legends. The point, rather, is that teams never beat SKT without snowballing: during the LCK and MSI playoffs, no one beat SKT without a gold lead of at least 2,400 at 20 minutes. In fact, in the two MSI Finals games where SKT beat EDG, SKT was still down in gold at 20 minutes (by 1,106 and 1,956), and even those leads weren't big enough snowballs to prevent SKT comebacks.
Looking at this from a different angle, while SKT was able to make comebacks fairly often, they completely shut down any comeback attempts by their opponents: the Korean champions only lost one of the nine games where they had a 10-minute gold lead of their own (a loss to CJ Entus, in which CJE had evened the gold up completely by the 15-minute mark).
In short, teams generally couldn't win by going even with SKT early on and counting on beating them in the mid game through good team fighting or better map play. It was vitally important for SKT's opponents to get their snowball rolling as they exited the laning phase.
The Source of the Snowball
An important question is what the anti-SKT snowball looked like. Where did the leads come from?
One important source was early kills (pre-10-minutes), whether through ganks, tower dives, skirmishes, or team fights. In SKT's losses, they had suffered an average of 3.29 deaths by 10 minutes, while only responding with 1.57 kills of their own. Compare that to SKT's wins, where at 10 minutes they averaged only 1.00 deaths with 1.91 kills.
The biggest targets for SKT's opponents in the first 10 minutes were MaRin and Wolf: overall, the Top laner had been killed an average of 0.78 times by 10 minutes, and the Support had been killed an average of 0.56 times. The Top and Support players represented 71% of SKT's pre-10-minute deaths.
When SKT's opponents were able to secure early kills, it gained them opportunities to pick up map objectives, such as early Tower damage, Dragon kills, or even just lane pushes that resulted in denying CS. All of those advantages helped to build the size and speed of their snowball.
The other very important source of snowballing for SKT's enemies was Tower kills. At 15 minutes, SKT was behind by 0.28 Tower kills, on average. In losses, that number was a much larger average deficit of 0.71 towers, which grew to 1.14 by 20 minutes. That translated into average SKT gold deficits (in losses) of 4,109 at 20 minutes.
For SKT's part, in their wins they only averaged a tower lead of 0.64 at 20 minutes, and a gold lead of only 1,981. This suggests that SKT preferred to play a more controlled game with less snowballing, allowing themselves to take care of business in the mid and late game with superior teamfighting coordination and map movement, either executing a comeback or methodically pressing their own leads home.
The Size of the Snowball
Because of SKT's ability to to win games that were close at the 20-minute mark, their opponents needed to secure big enough gold leads (and item advantages) to overwhelm them by sheer math. How big of a snowball was "big enough"?
As mentioned previously, nobody was able to beat SKT without a gold lead of at least 2,400 at the 20-minute mark. Very large snowballed leads were needed to prevent SKT from coming back in the mid game. Those leads turned into very large end-game gold differences, as well.
Gold Spent differential is a good way to measure how big a team's snowball grew by the end of a game. Bigger gold spent differentials suggest blowouts, while smaller differentials suggest that the game remained close and neither team was able to develop a meaningful snowball.
In SKT's wins, they averaged a gold spent differential of 8,438.
In SKT's losses, they averaged a gold spent differential of -11,912.
In short, SKT was able to win relatively close games where they hadn't developed massive leads. But when they lost, they tended to lose big, due to huge opponent snowballs.
Starting a snowball and getting it big enough to keep on rolling was very important, as we've seen above. But it was also somewhat helpful to choose team compositions that were able to make good use of mid game leads through appropriate power curves.
In SKT's losses, their opponents made frequent use of champions with strong mid-game effectiveness, such as Hecarim, Rumble, Maokai, Gragas, Sivir, and Lucian. In SKT's wins, however, their opponents were somewhat more likely to pick late-game scaling champions like Ryze, Azir, Cassopeia, and Jinx.  This wasn't a hard-and-fast rule: Fnatic and EDG still won games with a mid lane Cassiopeia, for example, and the one time EDG picked Corki, the pre-eminent mid game ADC, SKT managed to pick up a win.
This suggests that while mid game team compositions were useful, the champion choices were less important than how they were used.
With the exception of one of CJ Entus's wins, SKT's opponents were only successful if they were able to turn 5-minute and 10-minute leads into much larger 20-minute leads through lots of early kills and Towers, building as large a snowball as possible along the way. That meant an all-or-nothing approach to the early game, regardless of which types of champions were selected, though some champion picks seemed more conducive to snowballing than others.
Statistically speaking, SKT preferred slower, more controlled game play in the early stages. A controlled, slow-paced, low-action game may stay in the laning phase for 15 minutes or more, with more farming and less skirmishing or tower diving. Controlled games may be more likely to see standard lane setups (1v1 Top, 1v1 Mid, 2v2 Bot), which leads to later Tower kills and generally fewer champion kills.
That kind of control was exactly what SKT's opponents needed to avoid in order to build their snowballs. SKT's losses tended to be faster-paced games that advanced more quickly through the laning phase into relatively early mid games, via earlier tower kills and more combat in the opening minutes, leading to larger snowballs.
Tower kill times are perhaps the clearest indicator the fast pace that SKT's opponents pushed for. Look at how much more quickly the first Tower went down in SKT's losses (see below).
Towers tend to fall much sooner in 2v1 lane swap scenarios, suggesting that lane swaps were an important early tactic for SKT's opponents.
On top of the earlier tower kill times, there were simply more towers killed earlier in the game in SKT's losses. In SKT's wins, on average there were 2.5 combined Tower kills at 20 minutes. In their losses, however, there were 4.0 combined Tower kills at 20 minutes.
Compare those numbers to Fnatic: in their EU LCS and MSI playoff wins, there were an average of 4.4 Towers dead at 20 minutes, compared to 4.0 Towers dead in their losses. For EDG's MSI playoffs, the numbers were 4.7 Towers dead at 20 minutes in wins, and 3.0 Towers dead in losses. Some teams clearly benefited more from having more global gold in both teams' pockets earlier in the game, but SKT preferred to keep the gold totals lower for longer, reducing the potential for big leads on either side.
In addition, there was a lot more combat in the early stages of SKT's losses. Look at the Combined Kills at 10 Minutes numbers below.
At MSI, Fnatic drove the pre-10-minute action mostly through their Top laner, Huni, who averaged 1.4 kills and 0.8 assists at 10 minutes over the five-game series, and their Jungler, Reignover, who averaged 1.0 kills and 0.8 assists at 10 minutes. When EDG faced the Koreans, they ran their pre-10 action through their Support, Meiko (0.6 kills and 1.6 assists at 10 minutes), and through their ADC, Deft, and Jungler, Clearlove (both with 0.6 kills and 1.4 assists at 10 minutes).
Both Fnatic and EDG were able to rack up these early kills by getting stronger early-game pressure from their Junglers than SKT was able to get from Bengi. And EDG's Meiko, in particular, outdid SKT's Wolf in early-game contributions.
Because of the faster pace, SKT's seven losses came in accelerated games, with an average game length of only 35.8 minutes. Their eleven wins, however, averaged 39.2 minutes. The longer game times were either because SKT was able to stay in control and slow the game down, or because they needed the extra time to mount a comeback from an early deficit.
When SKT's opponents were able to dictate the pace with constant, aggressive early action and quicker Tower kills, the result tended to go the opponents' way. In SKT's losses there was 67% more bloodshed in the first 10 minutes, Towers fell much sooner, and the mid game arrived before SKT was ready for it. As a result, opponents had a better chance to secure their snowball (and admittedly a better chance to make a mistake and get snowballed on themselves). This led to shorter games in SKT losses, and longer games in their wins.
SKT's opponents found more success when they were able to play more aggressively early in the game. But even beyond the first 10 minutes, heavy aggression and simplified (some might say sloppy) tactical play were a key part of the path to victory over the Korean champions.
By aggression, I mean the frequency of combat leading to kills on both sides. During the LCK and MSI playoffs, SK Telecom T1 averaged a combined kills per minute (CKPM) of 0.81. At MSI only TSM had a lower CKPM than SKT, by a difference of merely 0.02. The Korean team clearly preferred more controlled games, relative to their international opponents.
In SKT's LCK and MSI playoff losses, however, they were involved in much more action than they seemed to prefer, with 21% more combined kills per minute in their losses (see below).
For context, AHQ e-Sports Club also had lower CKPM in their LMS and MSI playoff wins than in their losses (0.92 vs 1.03), while Fnatic had much higher CKPM in their EU LCS and MSI playoff wins than in their losses (0.92 vs. 0.68). EDG, in their MSI playoff games against AHQ and SKT, had 1.11 CKPM in their wins and only 0.79 CKPM in their losses.
SKT comes from Korea, a region that is highly respected for its opportunistic map play. They don't take fights without a reason; every fight has its place and its purpose. For North American LCS fans, it may sound like I'm describing Cloud9's tactical, controlled style of play. (For the record, C9 had the lowest CKPM of NA LCS teams last split in both the regular season and the playoffs.)
In other words, SKT prefers to slow the game down and make smart plays with low risk. This makes a lot of sense for a team with superior shot calling and coordination.
But SKT's opponents at MSI both had other ideas. Both Fnatic and EDG likely knew that they had a smaller chance to win in a controlled, tactical showdown. Instead, they apparently opted to bring the tactical level of play down a few notches, reducing the predictability of the games so that their individual skill and mechanical play could play a larger role.
It was a risky approach: it is inherently more risky to attempt a gank or a tower dive than it is to just keep farming. Tower dives especially have built-in risks if the enemy Jungler or Support shows up unexpectedly to counter-gank, and that can lead to disaster. But through excellent mechanical play during tower dives and well-chosen gank paths and target selection, the actual level of risk was decreased, and the rewards came flowing in. The plan worked out excellently for EDG, and almost worked for Fnatic.
Essentially, both teams knew they couldn't win at chess, but figured they might win at poker, so they embraced the randomness and placed their bets. They lived and died by their aggression, especially in the first 10 or 15 minutes, taking big risks with ganks, roams, and tower dives in order to find advantages. Those plays could lead to success, or they could lead to failure, but Fnatic and EDG accepted and even embraced that risk, because it gave them the highest chance to win.
We should note that the exception to this pattern was SKT's series against CJ Entus. In those five games, CJE worked to keep the kill pace even slower than average for SKT, suggesting that CJ Entus was the only opponent capable of competing evenly with SKT in a slow-paced, tactical game. But in the end, SKT triumphed in the Blind Pick fifth game, and SKT went on to prove against the GE Tigers that even other Korean teams might be wise to adopt a more bloodthirsty approach to unseating the LCK Spring Split champions when Summer rolls in.
In general, teams had more success against SKT when they engaged constantly and kept the kills flowing. This prevented SKT from controlling the game tactically, and likely put them off balance. With less tactical play and more constant skirmishing, individual player mechanics become more important than team movements and coordination, resulting in more volatile games with higher chances of an upset.
SK Telecom T1 may be the best team in the world right now. They were certainly considered the best going into MSI, and it's an open question whether EDward Gaming has actually won that title away from them, or is simply borrowing the championship belt until things get really serious at Worlds later this year.
But what MSI demonstrated is that SKT can bleed: they can drop games to European teams, they can drop series to Chinese teams, and they even came close to dropping a game to the LMS's AHQ.
The weapons required for drawing SKT's blood appear to be massive early-game snowballs, earned through heavy aggression and an accelerated pace that rushes into the mid game as soon as possible.
These weaknesses may not remain by the time the next international competition takes place. They may even already be resolved by the time the LCK Summer Split begins. Korea's coaching infrastructure and tactical play are the best in the world, after all.
But we could have said the same thing after IEM Katowice, during which a different Chinese team, World Elite, defeated a different tactically superior Korean team, the GE Tigers, through relentless early aggression and fast-paced play. Fnatic and EDG may have even been intentionally drawing on the WE model at MSI when they drew up their game plans.
It's up to SKT and the rest of the LCK now to prove whether an established pattern has been found for defeating them, or whether they will be able to prepare themselves to outlive the assault on their heavenly thrones.
Interested in more LoL analytics? Player and team statistics from the NA LCS and MSI are currently available on OraclesElixir.com, and all major regions will be covered over the course of the 2015 Summer Split.