跳到主内容
@wquguru
精选88RevenueCat 博客(RSS)产品与增长

缩短注册流程提升完成率但降低付费意愿的A/B测试

Everlove tested shorter onboarding on 24,000 users. Completion went up, trial starts went down.

原文
发到 X
推荐理由

给出了具体的A/B测试数据与漏斗诊断逻辑,读者可直接套用“先找流失点再调长度”的方法论,避免盲目优化。

Everlove, a couples app with 150,000+ users, A/B tested its 28-step onboarding against a 19-step version across 24,000 users. The shorter flow lifted onboarding completion by 8.6 percentage points and put roughly 700 more users in front of the paywall. But trial starts among paywall viewers fell 18%, and initial conversion didn’t move: the two effects cancelled out.

Everlove 是一款面向情侣的应用,拥有超过 15 万用户。它针对 24,000 名用户,对其 28 步的引导流程与一个 19 步的版本进行了 A/B 测试。更短的引导流程使引导完成率提升了 8.6 个百分点,并将大约 700 名更多用户推到了付费墙前。但在查看付费墙的用户中,试用启动率下降了 18%,且初始转化率没有变化:这两种效应相互抵消了。

The takeaway: onboarding length isn't the first thing to decide. Work out where your funnel leaks before you decide how long onboarding should be.

结论是:引导流程的长度不是首先要决定的因素。在决定引导流程应该有多长之前,先找出你的转化漏斗在哪里存在流失。

Here's a look at what we tested, and what shorter onboarding actually changed for Everlove.

以下是我们测试的内容,以及更短的引导流程实际上为 Everlove 带来了哪些改变。

Key findings

主要发现

  • Cutting onboarding from 28 to 19 steps raised onboarding completion by 8.6 percentage points (78.1% to 86.7%)
  • The shorter flow put roughly 700 more users in front of the paywall, despite a smaller cohort
  • Trial starts among paywall viewers fell 18% (3.83% to 3.13%)
  • Initial conversion showed no significant difference (3.35% vs. 3.11%)
  • Find where your funnel loses people before you decide how long onboarding should be
  • 将引导流程从 28 步缩短至 19 步,使引导完成率提升了 8.6 个百分点(从 78.1% 提升至 86.7%)
  • 尽管样本量较小,但更短的流程仍使大约 700 名更多用户看到了付费墙
  • 查看付费墙的用户中,试用启动率下降了 18%(从 3.83% 降至 3.13%)
  • 初始转化率没有显著差异(3.35% 对比 3.11%)
  • 在决定引导流程长度之前,先找出你的漏斗在哪里导致用户流失

A test was designed to settle the onboarding length question

设计了一项测试来解决引导流程长度的问题

The prevailing onboarding advice has swung back and forth over the years, but is currently heavily leaning toward longer onboarding being better. You only have to look at the success of Noom’s 113-step onboarding. Ask more questions, build more investment, let the user commit before you make the offer. The reasoning is sound, and there's real evidence behind it: a user who has spent four minutes telling you about themselves is more invested than one who spent 40 seconds.

多年来,主流的引导建议一直在反复摇摆,但目前强烈倾向于认为较长的引导流程更好。你只需看看 Noom 的 113 步引导流程所取得的成功即可。提出更多问题,建立更多投入,让用户在做出购买承诺前先进行投入。这种推理是合理的,并且有切实的证据支持:一个花了四分钟向你介绍自己的用户,比只花 40 秒的用户更有投入感。

Everlove, a subscription app for couples, had built exactly that flow. But it wasn't working. So we tested it against something much shorter.

Everlove 是一款面向情侣的订阅应用,已经构建了这样的流程。但它并没有奏效。因此,我们将其与一个更短的流程进行了测试。

Interestingly, the result came back split. The shorter flow won the top of the funnel decisively. The longer flow won at the paywall. Net conversion was flat.

有趣的是,结果呈现分裂状态。更短的流程在漏斗顶部取得了决定性胜利。较长的流程在付费墙处获胜。净转化率持平。

So the question became not which approach is better, but which one their funnel was stuck on at that moment.

因此,问题不再是哪种方法更好,而是他们的漏斗在那一刻卡在了哪种方法上。

Why we tested it

我们进行测试的原因

Everlove gives couples 15 minutes and something to talk about. Pick a topic, pick how deep you want to go, and the app serves up questions, quizzes and games about the things partners usually skip.

Everlove 为情侣提供 15 分钟的时间和谈话内容。选择一个主题,选择你想深入的程度,应用就会提供关于伴侣通常忽略的话题的问题、测验和游戏。

The onboarding was long, but length itself wasn't necessarily the problem. The flow included a substantial number of ‘About You’ screens focused on the individual user, alongside heavier questions about their relationship and what was or wasn't working.

引导流程很长,但长度本身不一定是问题。该流程包含大量专注于单个用户的“关于你”页面,以及关于他们关系以及哪些方面有效或无效的较重问题。

The hypothesis was simple: were these additional questions necessary to deliver the Everlove experience, or were they adding friction before users had experienced any value?

假设很简单:这些额外的问题是提供 Everlove 体验所必需的吗,还是它们在用户尚未体验到任何价值之前就增加了摩擦?

Two things made the onboarding worth testing:

有两点使得对入职流程进行测试变得值得:

1. Too heavy, too early: a long series of questions about what's broken in your relationship — asked before you've played a single game together — heavily front-loads emotional effort at the funnel's most fragile point.

1. 过早过重:在你还没有一起玩过任何游戏之前,就询问一系列关于你关系中存在的问题的问题——这在漏斗最脆弱的阶段过早地施加了情感负担。

2. A two-person product was spending a lot of time on one person: the flow included an extensive ‘About You’ section focused on person one, while Everlove's value depends on both partners joining and playing together. Users who bring their partner in convert at roughly 2x the rate of solo users. But every additional screen spent profiling one partner delays the behavior most associated with conversion.

2. 双人产品花费大量时间在一个人身上:流程中包含了一个广泛的“关于你”部分,专注于第一个用户,而 Everlove 的价值取决于双方伴侣加入并一起游玩。邀请伴侣加入的用户转化率大约是单独用户的两倍。但是,为单个伴侣建立档案所花费的每一个额外屏幕都会延迟与转化最相关的行为。

What we tested

我们测试的内容

We tested the existing long onboarding against a much shorter version. For the initiating partner, the flow dropped from 28 steps to 19, removing questions from the heavier ‘problems in your relationship’ and individual ‘About You’ sections to get users to the core experience and paywall sooner. The second partner's onboarding remained unchanged at 12 steps.

我们将现有的长入职流程与一个更短的版本进行了对比测试。对于发起方伴侣,流程从 28 步减少到 19 步,去掉了较重的“关系中的问题”和单独的“关于你”部分中的问题,让用户更早地接触到核心体验和付费墙。第二位伴侣的入职流程保持不变,仍为 12 步。

  • We measured success with: PostHog for onboarding start through to trial start or direct purchase, and RevenueCat for revenue, plan mix, retention
  • The test audience scope was: Worldwide — US volumes were too small to reach significance below the top of the funnel
  • Our test window was: from June 14 2026 onwards, to the point where both variants became equally exposed to a concurrent paywall test; onboarding length is the only difference between the two arms
  • 我们用以下指标衡量成功:PostHog 用于跟踪从入职开始到试用开始或直接购买的过程,RevenueCat 用于跟踪收入、计划组合和留存率
  • 测试受众范围是:全球——美国的数据量太小,无法在漏斗顶部以下达到显著性
  • 我们的测试窗口是:从 2026 年 6 月 14 日开始,直到两个变体都同等程度地暴露于并发付费墙测试的点;两个分支之间唯一的区别是入职流程的长度

The outcome

结果

The shorter flow won activation, definitively. Onboarding completion rose 8.6 percentage points — an 11% relative lift, statistically significant, and it holds in the US too (+5.3pp) where almost nothing else reaches significance. Despite a cohort that was 581 users smaller, the shorter flow put roughly 700 more users in front of the paywall.

较短的流程在激活方面取得了胜利。入职完成率提高了 8.6 个百分点——相对提升 11%,具有统计显著性,在美国也成立(+5.3pp),而在美国几乎没有任何其他指标达到显著性。尽管样本量少了 581 个用户,但较短的流程让大约多 700 个用户看到了付费墙。

The longer flow won the paywall. Trial start rate among people who reached the paywall fell from 3.83% to 3.13% — an 18% relative drop, and the one clearly significant negative in the set. This is the commitment argument showing up exactly where its advocates say it should: users who invested more were likelier to buy when asked.

较长的流程在付费墙方面取得了胜利。到达付费墙的用户中试用启动率从 3.83% 下降到 3.13%——相对下降 18%,这是数据集中唯一明显的负面显著结果。这正是承诺论点发挥作用的地方,正如其支持者所说:投入更多的用户在被要求购买时更有可能购买。

Initial conversion showed no significant difference. 3.35% against 3.11%, comfortably inside the noise, because those two effects roughly cancel each other out.

初始转化率显示无显著差异。3.35% 对比 3.11%,处于噪音范围内,因为这两种效应大致相互抵消。

If conversion per viewer had merely held, the reach gain alone would have produced roughly 444 conversions against 412. But instead, the rate fell far enough to absorb it.

如果每位观看者的转化率仅保持原状,那么触达率的提升 alone 就会产生约 444 次转化,而非 412 次。但事实上,转化率下降幅度足以抵消这一增长。

MetricLong onboardingShort onboardingDeltaSignificance
Onboarding completion rate (of users who started onboarding)78.10%86.70%+8.60ppStrong
Paywall reach rate (of registered users)74.26%83.92%+9.66ppStrong
Trial start rate (of paywall viewers)3.83%3.13%-0.70ppStrong
Initial conversion rate (of registered users)3.35%3.11%-0.24ppNone
Trial conversion rate (of trials started)14.29%15.91%+1.62ppNone
Average days to subscription4.013.58-0.43 daysNone
Average LTV per registered user$0.1163$0.1592+36.9%Weak (p=0.093)
Partner-joined rate31.26%31.98%+0.72ppNone
指标长引导流程短引导流程差异显著性
引导完成率(针对开始引导的用户)78.10%86.70%+8.60pp强
付费墙触达率(针对注册用户)74.26%83.92%+9.66pp强
试用启动率(针对付费墙观看者)3.83%3.13%-0.70pp强
初始转化率(针对注册用户)3.35%3.11%-0.24pp无
试用转化率(针对启动试用的用户)14.29%15.91%+1.62pp无
平均订阅天数4.013.58-0.43 天无
每注册用户平均 LTV$0.1163$0.1592+36.9%弱 (p=0.093)
合作伙伴加入率31.26%31.98%+0.72pp无

What we'll do next

我们接下来的工作

Here's what we're doing next. Each change can be applied to any funnel.

以下是我们接下来的行动。每项更改均可应用于任何漏斗模型。

Test the paywall, not onboarding again: the paywall now gets around 700 more users per window, and those users are less committed than the ones it used to get. The next opportunity is to make the paywall work harder: reinforce the value and address objections that the shortened onboarding no longer handles.

测试付费墙,而非再次测试引导流程:现在每个窗口付费墙能多获取约 700 名用户,但这些用户的承诺度低于以往。下一个机会是让付费墙发挥更大作用:强化价值主张,并解决缩短后的引导流程不再处理的异议。

Put the removed questions back later, not never: the questions were fine. Asking them early was the problem. Asked after someone has used the product, they should still make the experience personal without costing users at the start. We'll test that on its own.

稍后重新加入被移除的问题,而非永远不加入:这些问题本身没问题。问题在于过早提问。在用户使用产品后再询问,仍能使体验个性化,而不会在开始时增加用户成本。我们将单独对此进行测试。

Look at what people bought, not just how many users bought: the short flow may be producing fewer buyers who are each worth more. If that's true, it will show up in which plan they choose. Re-running plan mix will tell us whether that explains the revenue difference.

关注人们购买了什么,而不仅仅是购买人数:短流程可能产生的买家更少,但每位买家的价值更高。如果是这样,这将在他们选择的计划中体现出来。重新运行计划组合分析将告诉我们这是否能解释收入差异。

Run this on your own funnel

在你的漏斗中运行此测试

In this test, the longer onboarding built commitment: it won on trial starts among people who saw the paywall. But it also cost onboarding completions. So length isn't the thing to decide first — you need to understand where you’re losing people in the funnel, and then adjust onboarding to counteract that.

在此测试中,较长的引导流程建立了承诺感:它在看到付费墙的用户中赢得了更多的试用启动。但它也导致了引导完成率的损失。因此,长度不是首先要决定的因素——你需要了解用户在漏斗的哪个环节流失,然后调整引导流程以抵消这种流失。

Here’s what I’d suggest:

以下是我的建议:

  • Find where you lose the most people. Onboarding, or the paywall. Answer that before you touch length.
  • If you're losing them in onboarding, cut it. Shortening buys real volume, and that's the effect we measured.
  • If your paywall already gets plenty of traffic and converts badly, don't cut onboarding. You'll just send it more of the same problem.
  • Check every onboarding question against what the user has got out of the product by that point. If a question arrives before someone understands why it's worth answering, move it later or drop it.
  • Measure both halves. Onboarding completion rate and conversion per paywall viewer. One going up while the other goes down is the normal result, not a broken test.
  • 找出你流失最多用户的环节。是引导流程,还是付费墙。在调整长度之前先回答这个问题。
  • 如果在引导流程中流失严重,就缩短它。缩短能带来真实的流量增长,这也是我们测量到的效果。
  • 如果你的付费墙已经获得了大量流量但转化率不佳,就不要削减引导流程。你只会把更多相同的问题引向它。
  • 根据用户在使用产品到该阶段所获得的收益,来审视每一个引导问题。如果某个问题出现在用户尚未理解回答其价值之前,就将其推迟或取消。
  • 同时衡量两半数据:引导完成率和每位付费墙浏览者的转化率。一个上升而另一个下降是正常结果,并非测试失败。

更进一步:量化金融体系

看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力

进入量化体系 →

相似阅读

关联信息,但可能不是同一事件