Let’s Optimize: Azir Irelia

Dr.LoR goes over Mobalytics data to figure out statistically best build for Azir Irelia in Patch 2.9.

Dr. LoR here with an article from the ‘Let’s Optimize’ series, – the first one to be hosted on RuneterraCCG! My goal in this series is to use objective statistical data to figure out the best builds for popular meta decks.

Today we’ll dive into Azir Irelia, which is perhaps the deck that has registered the highest play rate ever in LoR history. Patch 2.9 brought a few changes, with nerfs to Inspiring Marshal, Blossoming Blade, and a “bug fix”/buff to Irelia. Due to its massive popularity, the archetype is largely considered to be “solved” – but that doesn’t mean there is consensus on the best list. That’s what we’re going to tackle today using Mobalytics data.

For the purposes of my analysis, I’ve pulled all versions of Azir Irelia decklists with at least 20 matches in Diamond and Master from the first 8 days of Patch 2.9. This made for a total of 249 different lists and 28k matches played, with the top five lists accounting for 39% of the matches. All Azir Irelia decks have an aggregate 55.5% win rate.

Cards Data

  • The Playrate super-column shows how often certain cards are included in Azir Irelia versions (‘Overall’), and in how many copies (‘0’, ‘1’, ‘2’, ‘3’). E.g., 75.2% of matches played are with lists that include 3x Shaped Stone.
  • The Winrate super-column shows the average win rates for decks that have that many copies of a certain card (‘0’, ‘1’, ‘2’, ‘3’). E.g., decks that run 2 copies of Shaped Stone average a 56.5% win rate, and decks with 3 copies have a 55.2% win rate. You will see that the percentage in the 1x Shaped Stone column there is written out in grey. It means there isn’t sufficient data behind it, so we can’t fully rely on it to accurately compare with other numbers.
  • The Drawn WR column shows the percentage of games won where at least one copy of that card was drawn; the Mulligan WR refers to the percentage of games won if this card is in your opening hand. Note that cards with lower play rates or played in fewer copies have less reliable Drawn WR and Mulligan WR data.
  • Keep % is the percentage of time players keep these cards in their mulligan. It’s important to understand when reading this data that when you mulligan a card away but draw a duplicate copy, Mobalytics still counts this case as a ‘keep’. Because of that, I believe that the ‘keep’ number even as high as 15% means that the card in reality is almost never kept.

Matchup Data

To figure out how to best go about optimizing a deck, we have to consider the meta we’re optimizing it for. Adding cards that help one matchup usually comes at the expense of another.

In the table below, you can see some of the most common matchups on Diamond/Master ladder in Patch 2.9 and Azir Irelia’s win rate versus each archetype. Margin of error means that the “true” matchup win rate is 95% likely to be +/- that number. For example, the “true” win rate of Azir Irelia versus Thresh Nasus is 95% likely to be between 38.1%-41.1%.

As you can see from the matchup data, Azir Irelia is weak to low-to-ground aggro decks, EZ Draven, and All-in Fiora. It is strong against midrange decks (except Dragons) and TLC.

Optimizing The List

For the purposes of this section, I narrowed down my initial data sample to 116 lists with at least 40 matches in Diamond and Master rank and calculated their win rates using Bayesian smoothing (it is a technique that allows avoiding statistical flukes. If you want to learn more about the concept, check out this explainer.)

Sadly I have to ignore a portion of the decklists that do not meet my requirements. This is a necessary evil since they would offer unreliable data: the win rates are easily skewed by a single good or bad pilot, and Bayesian Smoothing would also push these numbers toward the average so much that we don’t learn much from them.

Combo pieces

  • Azir needs little comment. It’s the highest WR card in the deck and you should always keep it in your mulligan. Obvious 3x.
  • Irelia used to be a middling card in this deck, but its buff made it the fifth-best card in the deck to draw. Obvious 3x.
  • Emperor’s Dais is a backup copy of Azir that’s even harder to remove (only Ezreal Draven commonly runs landmark removal) and has the second-highest Drawn WR in the deck. Obvious 3x.
  • Rite of Calling has started to make an appearance in decks since we’re more reliant on our champions now than pre-patch. We don’t have much we want to sacrifice outside of Dunekeeper, but the deck works reasonably well on little mana so giving up a gem is acceptable. Rite of Calling’s Drawn WR is right between Azir and Irelia’s, so having a little redundancy is good. 1x significantly increases WR over 0x, but 2x may be too much.
  • Ribbon Dancer is actually one of the lowest Drawn WR cards in the deck. It’s a 2-drop that you probably shouldn’t even keep in your mulligan. Yet, it seems to be a necessary evil and almost nobody cuts any copies of the card. The few lists that tried 2x have lower WR. I’ll stick to 3x but could see going to 2x.
  • Blossoming Blade was hit with a nerf. It also has a low Drawn WR and some people have experimented with cutting to one or more copies. In fact, the lists with 1x have significantly higher WR than 2x or 3x. I think we want to draw one copy but not two in most games, so I’m down with cutting to 2x for now but the adventurous should try 1x!


  • Dancing Droplet is the little 1-drop that could. It’s a prime bounce target for your recall effects and oftentimes the only way to draw additional cards in the deck. It’s an Elusive that can activate your attack effects when the ground is stalled. Yet, it has a mediocre drawn WR. There isn’t much experimentation with less than 3x, so we’ll have to stick with that, but 2x seems reasonable.
  • Dunekeeper is the third-best card to keep in the mulligan but has a medium Drawn WR. It’s a turn 1 Decimate or a two-blockers-in-one deal, but it doesn’t fit our general gameplan as well as many other cards in the deck. Lists running 2x have a significantly higher WR than those running 3x, so that’s what we’ll go with.
  • Sparring Student managed to dodge a nerf despite many calls for it. It has a high drawn WR and you should usually keep it in your mulligan. Obvious 3x.
  • Greenglade Duo has incredible damage potential after in conjunction with the Blade Dance mechanic. Yet, it has mediocre Drawn WR and it’s not great to keep in the mulligan. I think it’s a card you don’t mind drawing multiples of so I will keep it at 3x—the data suggests 2x may be better, but this difference isn’t significant.
  • Navori Conspirator is a nice combo with Droplet and works as an additional recall effect for your Blade Dancers. It has a medium drawn WR and decks running 1x have marginally higher WRs than 0x, but 2x has significantly lower WR. Since it can’t save your important units unlike your other recall effects, I chose to play 0x, but play 1x if you can find room for it. Decks with 5-6 self-recall effects (other than Homecoming) have the highest WRs.
  • Voice of the Risen has slightly above average Drawn WR and decks that run 1x or 2x have significantly higher WR than decks that run 0x. For decks that run 1x Rite of Calling, 3x Voice actually has the highest WR! Was Voice secretly better than Inspiring Marshal all along? I definitely think we need at least some of this effect in addition to Azir for matchups like Dragons. We’ll run 2x.
  • Inspiring Marshal was already cut by some players even before its nerf and it’s now pretty clearly underpowered. It has one of the lowest Drawn WRs in the deck, and lists running 2x or 3x have significantly lower WR than 0x. Most players have dumped it in favor of Voice of the Risen and I suggest you do, too.


  • Shaped Stone is an incredibly efficient combat trick, as long as you managed to draw Emperor’s Dais in your first few turns (which is ~56% by turn 3 if you hard mulligan for Dais and Azir). It has average Drawn WR and the stats suggest that 2x is marginally better than 3x. We’ll go with 2x but 3x is fine if you can find room for it.
  • Lead and Follow protects your units while generating more dances. It has medium Drawn WR but the decks that cut any copies have significantly lower WR so we’ll stick with 3x.
  • Retreat is the lesser of the two recall effects, with slightly lower Drawn WR. Instead of a Blade Dance, you get the occasional benefit of cheating a champion into play. Decks that run 2x or 3x have similar WRs and both higher than 0x or 1x. We only have room for 2x.
  • Nopeify! is important for protecting against removal on your champs and in the mirror but it has one of the lowest drawn WRs in the deck. Decks that include some copies have significantly higher WR than decks that run none. We also can run Deny, but Nopeify works for the spells you care most about. Interestingly, decks running 1x Deny show a significantly higher WR than 0x Deny. If we look at these spells as a package, the decks running 2x Nopeify, or 1x Nopeify and 1x Deny have the highest WRs. We’ll run the latter.
  • Syncopation is often run as a 1x, but its usefulness is relatively narrow, and it has the single worst Drawn WR in the deck. Decks that run 1x aren’t significantly worse off and I could see playing a copy in a tournament list to make your opponent play around it but we’re sticking with 0x.
  • Homecoming is the second coming of a pre-nerf Will of Ionia, especially when combed with Dancing Droplet. It has a medium Drawn WR and decks that run 3x have slightly higher WR than 2x, but this difference isn’t significant. I’ll stick to 3x but 2x is defendable. Actually, I’m a bit surprised nobody has tried 1x Will of Ionia in this deck…

Other cards


This is my final build based on the data-driven optimization is above. None of our choices were particularly revolutionary but we ended up with a unique list.

Comments or feedback are welcome, especially if you try this list or if you have a lot of experience with Azir Irelia in high-level play!

Good luck and have fun!

If you liked this type of post, follow me on Twitter and watch for future data-driven breakdowns of popular archetypes and meta reports.