Growth Models

How do you model and predict growth (and growth opportunities)?

There’s all this talk about “growth models” and “growth modeling” but not much talk about how to build them and get value from them.

As with many things, it’s easy to see why they’re important, but hard to put them into action.

This post covers how I think about growth modeling, how several impressive companies do growth models, and how you can build your own (even if you’re just starting out).

What’s a Growth Model? An Introduction

Growth models are a representation of the underlying mechanisms, levers, and reasons for your company’s growth.

They seem to have become popular recently, notably with the rising trend of ‘growth hacking,’ ‘growth marketing,’ ‘growth,’ or whatever we’re currently calling data-driven startup marketing (“a marketer by any other name…”).

Here are a few definitions of growth models from various sources…

Segment:

“Growth models…are feedback loops that project how one cohort of users leads to the acquisition of the next cohort of users. Viewing growth with this reinforcing model simplifies a complex system with tons of moving pieces to a set of functions and assumptions.”

HackerNoon:

“Every startup needs a framework/model for growth; a focused approach for scaling its organisation and user base.”

GrowthHackers:

“The concept of a growth model is both an old and a new one. It has a lot of similarities and connections to what’s traditionally called a “business model”, but companies and teams now focus much more specifically on growth and take a much more data-driven and experimental approach.

At its core, a growth model boils down to a way to conceptualize and summarize your business in a simple equation, which allows you to think about growth in a holistic and structured way.”

Translation: it’s a new concept for an old idea about creating a simplified high level model in order to make better business decisions. You’re trying to explain “how does this company grow?” What levers exist as inputs that contribute to “growth” as an output (however you define growth)?

In growth, there tends to be two different types of models:

  • Qualitative models
  • Quantitative models

Qualitative models are going to be more descriptive in nature. They’ll be high level descriptions of how your business is growing and plans to grow. You can draw pictures of these:

For example, it’s easy enough to analyze the growth of an ecommerce business on a qualitative level: look at your Google Analytics source/medium report. How are you currently acquiring visitors and customers?

You can say, at a high level, that you’re acquiring users from some main traffic sources, you’re converting them to subscribers, users, or customers at some rate, and some amount go on to repeat purchase (or tell their friends). You can get a pretty darn good idea of how your business is growing just from these numbers.

Even without considering attribution or looking at data, you probably have a good finger tip feel for how you’re getting customers. Is it virality? Word of mouth? Content marketing? PPC? Write all this down, and you’ve got a good idea of your primary growth levers.

Eventually, when you learn more about how you’re growing, particularly when you justify the use and implementation of a good analytics setup and can get granular data, then you can build out a quantitative model from this qualitative one.

Things can get pretty nitty gritty here, and every model is different (it really depends on your business concept). Here’s an example from Sidekick, a team that existed within HubSpot a few years ago that made features like Email Tracking and Documents that now exist within the HubSpot Sales suite:

As a hypothetical example, let’s say your ecommerce business is growing primarily from content marketing. In this case, you’d itemize the different steps of that customer journey and fill in corresponding metrics.

Some important variables in the context of content marketing might be sessions to the blog (overall top of funnel traffic), how many people click over to the store, how many people add something to their cart, how many people start checkout, how many people purchase, what the average purchase size is, how often people make return purchases on average, and possibly email subscriptions as well (which you could also build a micro-model out of).

You should make your spreadsheet prettier than mine, but here’s an example of modeling out that one growth channel over a few months.

Semi-related side note: I covered, in depth, how to model content growth and analyze results in this post on content marketing analytics.

The important part with a growth model is that you project out possible results to the future using your current trends. That way, you can tweak different variables to see what the biggest impact levers could be.

You can predict, based on current trends, where you’ll be in 3, 6, and 12 months, and if you’re okay with that projection. If you’re not okay with the projection, you can run sensitivity analysis to see which levers may be the most effective places to put your focus.

Is the add to cart to checkout page step low? Conversion optimization can help solve that. Do you simply need to lift your organic sessions because they’re stagnating? It’s easy enough to see that in the model.

Your model is only worth building if you’re going to use it to help you make decisions.

Growth Models Examples: 5 Ways to Model Growth

While most growth models are spoken of in general terms, they usually have similar ingredients. What acquisition channels are bringing in customers? How much is each customer worth? How long does each customer last? How many friends do they invite to your product or service?

In addition, they usually boil down to a few variables, which most of the time are further broken down into sub-variables. The main things are usually:

Acquisition Channels * Value of a User/Customer * Retention

One of the most lucid growth models I’ve seem comes from Drew Sanocki, who explains ecommerce growth in terms of three levers.

1. Drew Sanocki’s Three Levers of Ecommerce Growth

Drew Sanocki, ecommerce growth legend, teaches a concept in CXL Institute’s Ecommerce Growth Masterclass that I really like.

He explains that there are really only three growth levers we can pull to improve ecommerce growth:

  • Number of customers
  • Average order value
  • Customer retention

Here’s his full quote from the course (also, take the course):

“We get caught in tactical maneuver hell, where we look at all these tactical opportunities and get stressed out about optimizing this entire thing when it really only boils down to these three multipliers.

And the power of these three is that improving any one of them is good, but if you can improve all three, the results multiply.

For example, in a year, do you think you can increase your retention by 30%? Can you increase your average order size by 30%? Can you increase your total number of customers by 30%?

Any one of these in isolation, I think, is really doable. The trouble people get in is when they try to find the silver bullet that will double your total number of customers in a year. It’s really hard.

But if you look at only moving each of these only 30%, you’re going to more than double the business.”

Now, you can further break down each of these categories, right? Customer acquisition can be broken down into several smaller factors actually:

  • Customer acquisition channel (PPC, SEO, etc.)
  • Conversion rate (of the people that land on your site, how many convert?)
  • Word of mouth/virality (how many customers bring other customers to you?)

Each of those is sort of a mini-lever that you can pull within the number of customers category. That’s the magic of building out a quantitative model, too. Once you see that, while you’re bringing tons of customers to your site with SEO, but none of them are converting, you can begin to work on conversion rate optimization. It’s a great way to prioritize high impact growth opportunities.

Similarly, you can break down average order value into different factors:

  • Discounting
  • Operational costs (reducing the cost of shipping, merchandising, becoming more efficient)
  • CPA and advertising
  • Upsells/cross-sells/recommendations

This, I suppose, could be also put under the category of conversion optimization. But in reality, with this lever you’re optimizing for increased order size, instead of optimizing for increased purchase percentage.

Finally, the last category is around retention. How long do customers stay around (and in ecommerce terms, how frequently do they purchase from you?). To this end, you can break that down into channels like:

  • Email marketing
  • Purchase frequency
  • Customer satisfaction
  • Customer lifetime value

If you’re a subscription commerce company, this step is even more apparent and even more important.

Note: you can use models like this for things outside of ecommerce as well. For example, Shanelle Mullin used this concept to create a model for content marketing growth:

As Drew mentioned in the quote above, if you can increase one of these levers by a few percentage points, that’s great. But if you can increase every one of these levers by 10%, that’s compound value. That’s where growth happens.

This is a good macro-model for ecommerce. Let’s look at a similar model for SaaS (traditionally B2B, but works for B2C as well).

2. SaaS Growth Modeling

Similar to ecommerce, you’ve really got a few growth levers for SaaS:

  • Number of customers
  • Average order value
  • Customer retention

The only real difference here is how you define customer retention, and the steps that it takes to become a customer.

Often, in B2B, you’re going to break down your “number of customers” lever into distinct pieces:

  • Traffic
  • Subscriptions
  • Marketing Qualified Leads
  • Sales Qualified Leads
  • Customers

In this model, someone may find you via a search query (“customer feedback software”), and perhaps they land on a blog post. They find that you offer an ebook that explains how to accurately measure customer satisfaction, so they download that and they become a Marketing Qualified Lead.

Next time they visit, they sign up for a webinar on customer success, and they give you their company info and phone number. Now we can refer to them as a Sales Qualified Lead.

Because your model will be higher touch, most customers will require a sales touch, so you separate your stages into two distinct lead classes to reflect that.

If your B2B model is a lower touch model, like Dropbox, or if you run a B2C application, your model may look like this:

  • Traffic
  • Freemium or free trial users
  • Upgrade to paid customers

In this scenario, a customer may hear about HeadSpace on podcast advertisement, check out the website, and return later to give it a try. They sign up for the free app, use it for a couple days, then never returned. They dropped off before upgrading to becoming a paid customer.

For all intents and purposes, these B2B growth models have pretty much the same levers as ecommerce models, you just define stages differently and break steps down into micro-models differently (though how you break these down also depends on your acquisition strategy).

Image Source

Image source

One of the better worksheets I’ve seen was built out by Candace Ohm, data scientist and improv comedy legend. She offers an Excel spreadsheet (here) where you can  learn how funnel metrics, customer churn, user demand, and virality effect your growth curve. It’s worth playing around with:

We’ve got the high level models down now, so let’s dive into a few micro models that we can use to improve given channels or parts of the funnel, like word of mouth/referral and conversion optimization.

3. Optimizing Referral Marketing: A Model

Referral is usually a growth lever, regardless of business type or size. People telling other people about your business is one of the best ways to grow. Most smart businesses try to incentivize this in some way (in addition to building something worth talking about).

Though a lot of word of mouth is frankly un-trackable (if I tell a friend how awesome MeUndies is, they won’t know how to attribute that), you can track and optimize a good portion of referral traffic, especially if it’s incentivized (i.e. you give a discount or tracking code).

Like any other model, you’ll want to break it out step-by-step:

  • How many people are offered a referral link?
  • How many people accept the link?
  • How many people send the link to other people?
  • How many people do they send the referral link to?
  • How many of those people open the message?
  • How many of them click through to the website?
  • How many buy something?

Then it loops back, because you can offer that person a referral code as well. This is essentially known as a viral loop, and you measure its effectiveness using a ‘viral coefficient.”

We can also vastly simplify this model, like the following graphic shows:

Depending on your technology stack, you may have to pull this data and build the model by yourself. But you might also just be able to get the reporting from your tool, especially if you use something like Referralcandy.

4. Conversion Optimization Modeling

Conversion optimization should be an inevitable part of your model, because no matter what business you’re in, you’re going to bring a some amount of visitors to the website, and a certain percentage of them are not going to buy or become users.

If you can increase the percentage of visitors that convert, you get a compound effect over time (and increasing conversion rate increases the effectiveness of your other channels, which lets you bid more on ads, spend more on content, etc.).

Now again, if you’re in ecommerce, a lot of this is simplified due to simple prototypicality. In other words, almost all ecommerce sites follow a very similar pathway to conversion. Everyone has a cart, a checkout flow, a thank page after conversion, etc.

The simplest growth model you can construct for ecommerce CRO, then, is a sort of funnel. You can start at the broadest level and go all the way to the home run:
Homepage -> product page -> add to cart -> checkout -> purchase conversion.

Every time you can increase a step of this is a step in the right direction, though the end conversion is of course the most important (as well as the order size). But if you can systematically improve your funnel, all other marketing efforts will be improved by extension.

Everyone has their own approach to auditing and modeling conversion opportunities, as well.

I reached out to Luiz Centenaro, Optimization Manager at Optimizely & an eCommerce consultant, to see how he approaches CRO growth models, and he explains that he sets up a baseline with A/A testing and Google Analytics analysis:

“I typically approach an eCommerce site by running an A/A test on every touchpoint.

A/A test the Homepage, Category Pages, Cart Pages, Product Pages and Checkout and track clicks on everything. If you have a good marketing team this can all be accomplished within Google Analytics or Google Tag Manager but you can also do this with your A/B testing platform such as a Optimizely.

After you run your A/A test you’ll have a baseline with revenue per visitor and conversion rate for every page and you can segment to see the difference between mobile and desktop too.

Simultaneously while the A/A test is running you can research the demographics of the visitors using Google Analytics. One of my favorite reports is demographics by age.

Age and demographics should be taken into account when crafting hypothesis to A/B test. You won’t market to a 65 year old the same way you market to an 18 year old and they show significantly different user behaviors.”

It’s not much more difficult in other types of websites either, so long as you can define the discrete stages of your customer journey. With a B2B SaaS company, that may look something like: Homepage -> Pricing Page -> Get Started Page -> Signup Flow (several steps?) -> New User Created -> Activation Event -> Upgrade to Paid

Conversion optimization can help optimize the steps on a site that lead to a visitor becoming a user, and possibly even a user becoming an activated or engaged user.

5. Virality: A Micro-Model

Viral growth is one of the better developed channels for building out models. While some viral mechanisms may look different (Apple iPods, Hotmail, Dropbox, and Bird scooters are all examples of virality), they do include similar variables that allow us to model the system:

  • The Viral Coefficient (K)
  • Viral Cycle Time

The Viral Coefficient is just a name for the number of new users a current user brings in through virality. The formula is stupid simple: K = i * conv%, where i is the number of invites sent and conv% is the conversion rate of those invites.

That’s really all you need to model out the first part. ForEntrepreneurs even offers a free spreadsheet to help you build that out (here):

The other part of the equation is how fast you can acquire new users through viral loops. Clearly, the faster you can go through the viral loop, the better.

As with the Viral Coefficient, your Cycle Time includes several sub-variables as well. David Skok draws out an example here:

To the extent you can shorten the time length between any of those steps, you can increase your growth rate.

It’s important to model the time factors as well (which is included in the ForEntrepreneurs spreadsheet. Really, if you’re interested in viral growth, I’d just read that post, as this section is clearly just a summarized version of it):

With this, and every other model, the powerful part is that you can tweak different variables to see what happens to the output. Increase or decrease conversion rate of invites by 5%. What happens? Increase the number of invites sent. Does that move the needle?

This can help you make important decisions on what drivers to focus on when optimizing viral loops.

Everyone wants virality.

As David Skok put it, the perfect business model is “Viral customer acquisition with good monetization. However viral growth turns out to be an elusive goal, and only a very small number of companies actually achieve true viral growth.”

In my experience, this area has been the most over-exposed to bad content publishing and bad thought leadership. Most of the core of viral growth is a noteworthy product. It’s hard to make a bad or a commodity product viral (though not impossible, just probably not worthwhile).

Limitation With Models and What to Expect

As the statistician George E.P. Box famously said, “All models are wrong; some models are useful.”

No matter what model you use to represent and predict growth, it won’t be completely accurate once the rubber meets the road, once your plan meets the messiness of reality. As the great philosopher Mike Tyson once said, “Everyone has a plan until they get punched in the mouth.”

This echoes the common wisdom, probably first said by Helmuth von Moltke the Elder: “No plan survives contact with the enemy.”

That’s all to say: be fluid, and update your model as you get new information and insight.

When you look at, say, an inbound marketing funnel, you don’t expect it to exactly and linearly reflect reality, do you? Funnels are growth models; they’re ways to simplify the concept of how you’re growing, provide directions for measurement, and allude to opportunities for effort and optimization.

These types of things are most useful in planning; the execution of those plans is still to be determined. I’ve built out complex models only to find once I hit the ground that I didn’t actually have the resources to carry out some of my top prioritized plans. Whoops.

So, don’t expect these models to be perfect. Expect them to be useful and actionable.

Conclusion

Growth models give you an imperfect, yet helpful model to show you how you’re growing, what kind of growth numbers you can expect in the future, and some possible opportunities to impact that growth in a positive direction.

In most businesses, there are really three levers you can pull, at least at a high level:

  • Number of total customers
  • Average customer/transaction value
  • Retention/lifetime value

To each of these levers, you can break them down into, really, an unlimited array of possible channels and tactics. That’s where things get complicated (and there’s always a tradeoff between cost and complexity of modeling). A good model is both useful and accurate, to some degree of each, but no model can be both perfectly accurate and comprehensible/useable.

Build out models to give you a better understanding of your growth, and also a better way to communicate that with others. Don’t expect a perfect vision of reality, but expect them to help you prioritize and find opportunities you otherwise may not have.

Comments are closed.