你的应用图标好比是一个位于非常喧闹的杂货街上的店铺门面。

如果它足够温馨和友好，有趣并富有吸引力，你就能够鼓励人们进入店内走走看看，如果它了无生气或寒酸，那就难免无人问津的惨状了。

应用图标的确是值得开发者投入的一个元素。有家顶级游戏公司的联合创始人曾告诉我，他们花了3个月时间为一款怪物RPG游戏设计图标。而我们的首款游戏也曾测试了64个不同版本的图标，以便从中找到最佳方案。

app icons(from iteratingfun)

（这是我们过去数周测试的一些图标样本）

那么你如何执行有效的A/B测试以找到最出色的图标呢？

在此我将分享一些分享以助你进行判断。

1.良好的方法

有一个快速、简单而廉价的测试方法就是使用PickFu.com这类服务。它们可以执行简单的A vs B测试，通常可作为走向众包平台（如Amazon Mechanical Turk）的前端，询问参与者他们更喜欢哪一个选择。

优点：

*廉价

*非常快速

*能够根据用户群体特征进行筛选

缺点：

*带有测试者的主观偏见

*要服从参与者的自我意识

*缺乏真正的意图与情境

2.更好的方法

下一步你可以使用Facebook Ads这类平台。它可以让你在许多用户面前同时部署多个版本，以便衡量点击率情况。

优点：

*迅速

*能够根据兴趣、用户特点进行定向测试

*较少主观因素

*样本大小具有灵活性

缺点：

*用户意图和情况更为接近实情，但还不是真正的用户意图

*为获得统计数据，可能需要根据样本量投入许多成本

3.最佳做法

最好的测试方法就是，在你的应用可能被发现（游戏邦注：例如通过广告和应用商店排行榜）的这种情境下，找一名真正的用户进行测试。对iOS和Android平台而言，开发者可以选择AdMob、iAds、inMobi、Millenial Media等网络投放广告。而像付费墙等刺激性奖励渠道则无法准确衡量用户的真正兴趣所在。

除了点击率之外，你还可以通过绑定广告网站SDK等方法衡量安装及应用开启的转化率。这样才能考察玩家对你的游戏描述和登录页面的感觉是否与其产生的兴趣（这里指的是游戏图标）一致。

如果你的应用尚未发布怎么办？那你可以另外设置一个独立的测试帐号并在那里运行广告。Android平台就是一个很好的渠道。如果你想低调行事，还可以针对特定地区测试应用和推广活动。

优点：

*真正无主观偏见和具有代表性的用户

*准确的情境

*灵活的样本量

*衡量准确的转化率

缺点：

*需耗费一定时间绑定SDK，审核创意及衡量周期更长

*为获得统计数据，可能需要根据样本量投入许多成本

游戏邦注：原文发表于2012年7月6日，所涉事件及数据以当时为准。（本文为游戏邦/gamerboom.com编译，拒绝任何不保留版权的转载，如需转载请联系：游戏邦）

A/B testing your icon (good, better, best)

startup marketing

Your app icon is like a shop front, on a very busy retail street.

Warm and welcoming*, interesting and attractive and you’ll encourage people to wander inside and check it out  Drab or shabby (drabby!) and few will bother to look.

* I mean seriously, who could resist that cute a baby dragon?

It’s clearly something people think is worth investing in.  A co-founder of a top game company told me it took them 3 months to come up with the icon for one of their monster RPGs.  For our first game we tested over 64 different versions on the way to identifying a winning icon.

One of the many tests we did over the course of several weeks

So how do you actually conduct an effective a/b test to come up with a great icon?

There are several ways – here are a few approaches in order of increasing accuracy.

1/ Good

One easy way to run a very fast and cheap test is to use a service like PickFu.com.  They do simple A vs B tests, typically as a front end to a crowdsource platform like Amazon Mechanical Turk, where the participants are asked which they prefer.

Pros:

•Cheap

•Very fast

•Ability to filter by demographic

Cons:

•Bias of tester self selection

•Subject to participant self awareness

•Lack of real intent and context

2/ Better

Next up you can use a platform like Facebook Ads.  This let’s you deploy multiple versions in parallel out in front of many eyeballs and measure click through percentages.

Pros:

•Fast

•Ability to target by interests, demographics

•Less bias

•Flexible sample sizes

Cons:

•Intent and context is closer but not same as real intent

•Can cost a lot depending on sample size to get statistical significance

3/ Best

The best way to test is to actually test with a real audience in the context they would discover your app once launched (ie ads and appstore charts).  For iOS and Android this can mean ads on networks including AdMob, iAds, inMobi, Millenial Media etc  You don’t want to use incentivized channels like offerwalls as they aren’t an accurate measure of actual self-generated user interest.

Beyond just clickthroughs, you can also measure true conversion to installs and app opening, by integrating the ad networks’ SDKs. Ie- do the players feel your game description and landing page reflects what they were interested in, which was the icon.  (A whole another story on whether the game matches to what they expected when they installed)

What if you app isn’t already launched? Then you can setup a separate test account and run the ads to point there.  Android is good that way.  Also you can restrict the test app and campaign to certain geographies if you want to keep it fairly low key.

Pros:

•Truly unbiased and representative audience

•Accurate context

•Flexible sample sizes

•Measure actual conversion

Cons:

•Takes time to integrate SDKs, approve creatives and measure – cycles are longer

•Can cost a lot depending on sample size to get statistical significance（source：iteratingfun）