Here’s how to check Unit Economics before you scale.

If Unit Economics are critically important to a startup in 2016, but they change as a business grows, how can a startup test and prove Unit Economics before they scale up? Here are some practical suggestions inspired by my own experiences and that of others.

Image from Pablo by Buffer, Wisdom from Ice Cube.

Note: This is post 2 of a 2-part series. To recap what Unit Economics are and why they are so hot right now, click here to read the first post.

TL;DR:

Create a controlled simulation of all aspects of your supply and demand within a controlled market space.

Here are some methods that are proven to enable that…

Constrain test market by location

Use geographic constraints to create large penetration of both seller supply and buyer demand, but only within a certain area, such as a city.

Example: BankAmericard & The Fresno Drop

When I was consulting for Visa last year, I learnt of an excellent example of this, from nearly 60 years ago.

In 1958 Bank of America wanted to test their concept for the “BankAmericard” credit card, which would later become Visa. BofA knew they needed to create a two-sided market of both merchants and customers, so they decided to focus all of their efforts only on the city of Fresno, California, to create a controlled experiment that became known as “The Fresno Drop”, recently covered by the podcast 99% invisible.

Image borrowed from http://99percentinvisible.org/episode/the-fresno-drop/

This allowed Bank of America to simulate the full unit economics of targeting a given population, which could then be used to model the possible numbers for entering the next market. A similar strategy can also be seen by modern day companies such as Uber or AirBnB starting in San Francisco, and indeed by most dating apps that need to create a critical mass of singles that want to meet each other.

To make this work effectively for a business:

1. Choose the most winnable target market.

Bank of America chose Fresno because it was the city with the highest penetration of both merchants and customers with existing BofA accounts per capita. Therefore it represented the best opportunity to create scale quickly. Or in other words:

If it won’t work here, it won’t work anywhere”.

This might also mean choosing the area where you have the most advocates and partners. For example I have many years experience working in the automotive industry with clients including Audi, Ferrari, Fiat, McLaren and Mercedes-Benz. In the auto industry it is exceptionally difficult for a new initiative to succeed if the dealers don’t willingly embrace it. Therefore all of these companies have an inner circle of dealer partners that volunteer their support for tests in their local area.

2. Focus your supply

Ensure all logistics needed to deliver a lovable service to customers in that specific area are in place, including supply channel logistics and customer support. For B2B SaaS companies, this is perhaps not essential, but it is easy to see how important this can be for startups with complex delivery logistics, like Sprig, Zirx or Instacart.

3. Tightly target your demand generation.

Most digital marketing channels allow pretty accurate geo targeting. Google Adwords, Facebook, Twitter and many ad display networks allow targeting to specific metro areas. Location-aware services like Foursquare or Yelp could allow even more granularity, or for less digitally savvy demographics, even outdoor billboard ads and local radio can be considered. All of these options help reduce wasted ad spend by ensuring that you are not paying to reach customers outside of the boundaries of your test.

4. Measure and improve.

Obviously it is critical to measure the right metrics throughout the test, with a focus on those metrics such as CAC and magic ratio. I’d go further to argue that the test is not complete, however, until some level of optimization on the initial results has been performed, so that the company has some proof of which metrics realistically have room for improvement, and which should be accepted as-is for future planning.

5. Scale up and out

With unit economics at the scale of a single local market now proven, the startup is now better able to predict the unit economics of entering the next market. This can help set budgets for both supply and demand generation and help the startup scale efficiently.

A word of warning!

The idiosyncrasies of individual markets may create significant swings in the actual results. Everything from the traffic in the city, internet speed or cultural values could have a huge influence. Therefore, the results from one locality should of course only be considered a guideline for the next, and actuals should be checked and refined as often as possible.

Constrain market by other factors

Many businesses and markets transcend local areas. The Unit Economics of a B2B SaaS startup like ours for example are not really influenced by the location of the customer. So let’s look at some other ways to slice and dice a test market.

A. Demographics

In addition to targeting media by location, digital media also allows us to constrain targeting to many other demographics factors such as age and gender. Again, Facebook is great here, and probably the simplest way. But many other “DMP” (Data Management Platform) data exchanges and ad network products also allow you to target additional factors such as consumer affluence, professional interests and more. If you want to be freaked out by just how much ad platforms “know” about about a person, visit the Blue Kai Registry and see just one example — yours!

B. Technographics

Is your product or service currently only available to users of certain technologies or devices? If not, could that be a way to sub divide your user base and learn unit economics in a controlled way before scaling too fast? If you are working on a mobile app, could you release to iOS and create some benchmarks unit economics before scaling to support Android as well?

Example: My own startup is working on a product that currently supports software teams that use GitHub, Node.JS and GitFlow. We are not currently able to support teams that use other tools such as BitBucket or Jira, but our initial serviceable market gives us some great unit economic benchmarks that we can use to size the relative cost/opportunity of adding new integrations and support.

C. Psychographics

Or “Attitudinal” factors build upon demographics to define more characteristics of what a target market thinks, feels and does. For example, several years ago I was consulting for British supermarket Sainsbury’s, who were at the time battling with (Walmart-owned) Asda for the position of second largest supermarket in the UK, behind Tesco. While the potential market for Sainsbury’s was the ~62 million people that lived in the UK, Sainsbury’s could not afford the media spend to reach everyone.

Instead Sainsbury’s worked to define a subset “Mindset audience” of approximately 12mm people that they felt “should” be shopping at Sainsbury’s. The mindset audience was constrained to people that agreed with attitudinal values that Sainsbury’s exhibited, such as “I believe it is worth paying extra for organic food” and 10 or so other points that would identify an ideal shopper from among the masses. This psychographic classification was used to focus media spend, which at the time was largely “offline” media, but now could be done much more effectively with digital media thanks to ad platforms.

D. Previous behavior

Now I know that data exchanges and retargeting are not everyone’s cup of tea, but I love them. As a consumer I get to see more relevant ads, and as a marketer, I love the ability to reach only people that have previously done something that suggests that they might be a potential customer for my product.

Example: When I led digital for Audi of America at AKQA, we were able to use information collected from data exchanges and DMP’s to understand which type of car a visitor might be interested in BEFORE they even visited any Audi-owned presences at all. So if a person had configured say a BMW M3 on an auto industry site, we could show them an Audi RS5 on a display ad that could follow them around the web, or even their very first arrival to our site.

The opportunity here is to look for signals where customers may be unconsciously identifying and segmenting themselves in a useful way. For example, we could use this information to send a test message to only new customers that seemed interested in a small Coupé, and not include the larger markets of sedans and SUV’s.

Another warning!

When slicing a market by behavior, ensure that you don’t confuse buying stage, or purchase intent within the mix. For example, unit economics learnt from the coupé market may provide helpful benchmarks before scaling up to sedans, but the CAC for customers that have configured a car, will probably be much healthier than CAC for customers that have not yet configured a car, as the the former probably has much more purchase intent and is easier to convert.

Cut to the chase — set a fail condition.

One suggestion made by Tristan Kromer at a recent meeting of the Lean Startup Circle, San Francisco, was to start by stating a fail condition; that is, stating a worst-case, minimum acceptable scenario. Know the numbers below which you would not consider continuing with your business, then structure a test to see if you can at least achieve that.

  • What is the maximum cost under which your business model is still viable?
  • What is the least revenue you can afford to make to offset those costs and create acceptable profit?

This gives a worst-case start point for a volume you need to sell, and from this you can probably use industry benchmarks to work out any missing pieces of the equation. When you have a model worked out, you could choose any of the above approaches to constrain your test to a manageable level, and check that the numbers are not below your go/no-go levels. If they are below, and you can’t improve them with some reasonable optimization, maybe it is better to stop sooner rather than later.

One other recommendation that Tristan made: if you know in advance that your Unit Economics are not profitable, but you decide to do it anyway due to some other reason within your business model, (EG, being a loss leader to unlock another opportunity), it may be better to assign the whole initiative as a marketing expense within the broader business model. Because that is what it really is.

Summary

All of this is essentially using the advanced targeting capabilities of digital media to subdivide your huge full-scale audience, into a targetable smaller one that looks and behaves just like it, then learning from the unit economics of the subset to sharpen the assumed values of the full scale audience.

In the early days, a lot may be guess work. But it is better to have a plan that changes than no plan at all. Everything will change as you move from one market to another anyway.

Like Product/Market Fit, Unit Economics will always be a moving target, and so we always need to keep chasing it.

Questions or suggestions?

Let me know in the comments below if you have feedback, via twitter or by email.