Blades / timothy leung

A Primer on Startup Growth

This primer is not original work but pulls together information from multiple sources including articles, slides and talks from Andrew Chen (AppSumo), Andy Johns (ex-Facebook), Dave McClure (500 Startups) and Paul Willard (Atlassian).

Growth is what turns a startup into a large company.

The metric used to measure growth is often active users or revenue/profit. However, it can also be taken from a diverse set of variables and will often change depending on the stage of the startup.

There are a few mantras that always get thrown around: “if you build it, they’ll come” and “you’ll win if you build the best product”. While sounding great in theory, these often don’t hold up in practice. After all, was Internet Explorer really better than Netscape? Excel better than Lotus 1-2-3? VHS better than BetaMax? Rather, whoever builds the best distribution wins.

Growth needs to be planned.

This post is divided into three sections: (i) choosing a suitable growth metric, (ii) monitoring the metric, and (iii) how to go about growing that metric.

A quick aside before diving in. Startup growth is very much dependant on the startup’s actual product. Optimizing for growth can only help so much and, in a lot of cases, time may be better spent on product/market fit. As Donald Knuth puts it, “premature optimization is the root of all evil”; always improve an inferior product before focusing on growth.

In fact, one way to test how badly people want a product (and hence determining the startup’s growth potential) is to erect barriers: make it difficult to invite friends, make it ugly and make it slow to use. If it still grows at a reasonable pace, optimizing for growth will only step up the pace.

Choosing a suitable growth metric

The key to choosing a suitable growth metric is understanding the user conversion funnel. These are the steps a user goes through while using the product. It is called a conversion funnel because people are lost with each step they are made to step through.

User Conversion Funnel

1. Acquisition: the user arrives in front of the product from a variety of channels.
2. Activation: the user has a great initial experience.
3. Engagement: the user keeps coming back because the product delivers value.
4. Revenue: the user conducts some monetization behaviour.
5. Virality: the user recommends the product to other people.

Many actions can have varying effects on user numbers and revenue/profit. Only by committing an action in an isolated section of the funnel ceteris paribus, can growth be accurately attributed to certain actions.

If the steps in the entire user conversion funnel are fixed, then the conversion rate (number of people entering the step vs. the number of people exiting the step) is a good metric to use. If the steps aren’t fixed, consider removing certain steps and seeing how the entire user conversion funnel performs as a whole. Reports suggest that each additional page or step in a website’s flow leads to a 20% drop-off rate.

The conversion rate, however, isn’t everything, and we use an example to illustrate this. According to Elliot Shmukler, LinkedIn decided to improve two sign-up channels in 2008. People who had reached the sign-up page were divided into whether they came from email invitations or from viewing the LinkedIn homepage. After some work, the following results were achieved:

1. Email invitation conversions jumped from 4% to 7%, taking 2 years.
2. Homepage view conversions jumped from 40% to 50%, taking 4 months.

There are a number of useful take-aways from this example.

1. The amount of effort is not necessarily proportional to the results. Email invitation conversions took considerably longer to improve yet were less effective in nominal percentage points than homepage view conversions. In certain cases, it’s better to double-down on what works instead of improving parts of the funnel that have a low conversion rate.
2. User conversions in a funnel only tell half the story. The quality of the users also matter and this can be assessed by segmenting users by channel and following them through the rest of the user conversion funnel. In this case, although email invitation conversions may take more work, what if converted users in this channel are more engaged, monetize easier and are more likely to invite friends? This is called cohort analysis and is a great way to understand the different types of users and their needs. For example, tell-tale signs of a later engaged user at Facebook was whether the user added seven friends in the first ten days, while for Twitter it was whether they followed thirty accounts.
3. The actual number of users entering the funnel is important. If 10M people enter the email invitation funnel, but only 1M people enter the homepage view funnel, then it may be more judicious to focus on the email invitation funnel since 3% of 10M is greater than 10% of 1M.

Monitoring the metric

All changes made to the product need to be compared objectively. A/B testing serves two different versions of the product to different users. These change can be small (the headline may be re-worded or the color of the button changed) or big (the user may be given the option to try the product before signing up). Once a certain number of users have tried option A and option B, the results are tallied up and a winner determined. As Andy Johns (Wealthfront) puts it: data is wonderful in never letting terrible ideas have a long shelf-life.

A few words of advice regarding A/B testing:

• It’s often important to find a balance between being data-driven and data-informed. A/B testing blindly is not very effective; the tester still has to come up with the options in an A/B test.
• A/B testing is often used to determine whether A is better than B, but it’s also an opportunity to understand the market. Perhaps B is better than A by 40% in the general NYC area, but in the rest of the country, it’s worse.
• Don’t A/B test your core repeat users, it will annoy them if the website looks looks different everytime they visit.
• Weekend traffic may be different from behaviour during the week, so consider run tests that span the course of a week.

Once a test is complete, the reason behind why users behaved a certain way should be analyzed. It often boils down to this relationship:

Conversion = Desire - Friction

To test for desire, start by changing headlines and the copy. Google AdWords can be a useful tool for seeing which words appeal to people. To test for friction, on the other hand, optimize for speed and change around the placement, size and color of the call-to-action. CrazyEgg produces a heat map and scroll map that helps in gaining a better understanding of how visitors engage with a website.

A/B test results are the conversion rates in the user conversion funnel. A tool like Kissmetrics or Mixpanel is able to link actions from the same user and plot the steps they take through the product alongside the relevant conversion percentages. Alternatively, this can be achieved by trawling the server logs and linking up requests by the same user.

Irrespective of the tool, ensure that a standardised report is spit out automatically for each change that you make. The report should look something like this:

Example Conversion Metrics

This will let you know whether your changes were effective and where next to concentrate your efforts.

Growing the metric

With the growth metric monitoring in place, it’s now time to focus on growing the metric. Unfortunately, there isn’t a systematic framework for coming up with the ideas that drive growth. What’s possible is to monitor their effectiveness and try to understand the rationale behind why certain changes work better than others.

Here are some things that are worth trying:

• Communicate clearly. Change the colour, shape and size of the call-to-action, copy, headline and tagline. The aim is to effectively communicate the value of the product to the user.
• Users want to use your product. Don Norman (ex-Apple) believes that everyone comes to a website with a reservoir of goodwill. Things that deplete it are: hiding information I want, punishing me for not doing things your way, asking me for information you don’t need, tricking me, putting obstacles in my way, amateur looking websites. Things that increase it are: knowing the main things that people want to do on your website and making these things easy and obvious, telling me what I want to know, saving me steps wherever you can, knowing what questions I’m likely to have and answering them, providing me with creature comforts like printer-friendly pages, and making it easy to recover from errors.
• Plot the engagement loop. According to Chen Li Wang (Dropbox), it can be very hard to resurrect people you’ve already lost without coming across as being pushy. Instead, if done right, emails, notifications and app store updates can help get people back to your product. Also, closely monitor engagement and unsubscribe rates of messages before sending them out to everyone.
• Plan for growth. Paul Willard (Atlassian) reveals that to show growth, only two graphs are needed: the Acquisition Exponential graph (customers vs. months) and Cohort Curves (life-time value vs. months as customer). Cohort Curves should shift higher and steeper as the company grows and their understanding of their customers deepen. This means that newer customers are buying more from the start as well as accelerating their pace of buying.
• Try different ways of understanding the user. Survey users to understand them better. For example, ask “What would you do if you could no longer use this product?”, “What was the primary benefit you received by using this product?” and “Would you recommend this product to someone else? Why?”
• Identify the magic feature. Andy Johns (Wealthfront) recommends dumping the session logs of twenty user who love the product and twenty users who never use the product. Then, manually piece together their histories and figure out their usage patterns towards the beginning of their experience. This will help identify the magic feature or behavior that gets people hooked.
• Make a great product rooted in human psychology. Facebook satisfies the curiosity of knowing what your friends are doing, Twitter, with its public tweets, retweets and favourites, appeals to one’s need for acceptance and popularity, and Instagram is a form of visual escapism.
• Move fast. Eric Florenzano (Twitter) explains that if you want to move as fast as possible, use HTML5 for components of your app that you aren’t sure about and only implement natively the parts that you’ve already nailed.
• Make it easy to share. Airbnb’s “Post to Cragslist” link allowed users to post their listings to Craigslist, which didn’t have a public API supporting this function. Have the option of importing the contact list.
• Make the product inherently viral. Hotmail added a “Get your free email at Hotmail” link at the bottom of each email after determining that 80% of signups were from referrals.
• Don’t make users wait. Sean Ellis (Qualaroo) recommends identifying the must-have experience of your product and to look for ways to front-load that experience. The sooner your customers experience the value of your product the better. Pinterest populates your feed by forcing you to follow a curated set of quality users when you sign up.
• Give users a reason to share your product. PayPal paid each referer and referee $10. Dropbox users could invite friends to increase their space by 250MB, helping them to grow from 100K users to 4M in under two years.