January 11, 2012 by startupengineering .

As discussed in previous posts, the business model canvas works differently for existing businesses than it does for startups. For an existing business, the contents of each box need to be a clear description of how business is actually conducted. It doesn’t follow, though, that for a startup, it’s sufficient to describe how the founders anticipate that they will conduct business.

To be maximally useful, the propositions (hypotheses, guesses) in each box need to meet at least three tests: coherence, binary testability, and framing.

Here are Joe and Amy working on the canvas. Joe’s an entrepreneur who believes in the benefits of organic food, wants to share those benefits with his dog, but can’t afford premium organic dog foods. He meets an organic farmer, Amy, at the farmers’ market, who tells him that frequently she can’t sell all of her produce, and she’d love to be able to get rid of aging goods, even at a discount price. An idea is born: discount organic dog food made from farmers’ excess inventory.

Joe and Amy start OrganicDogFood.com and construct the following business model canvas:

CoherenceThis canvas is a reasonable-seeming description of a business model, but is it a useful tool? A useful canvas is one with testable propositions which, as you validate them, take you through a process of developing and discovering a repeatable, value adding business model.

A proposition must be coherent in the sense that it works in the context of the business model as a whole. Amy is an expert, so her opinion that organic farmers will partner with OrganicDogFood.com to manage excess inventory carries some weight. But that opinion is only a starting point. What validation would be sufficient? Further evidence along the same lines (ten more organic farmers corroborating Amy’s experience) wouldn’t solve the problem, because a valid partnership with organic farmers has to work in the context of the entire business model. For example:

• The key activity of “developing recipes” implies that different ingredients may be needed. What are they? Are the organic farmers willing to sell at a discount in a position to provide those specific ingredients?
• Are ALL the necessary ingredients available from organic sources? (If not, the value proposition is in question.)
• Are ingredients available in all seasons or with a short enough lead time to meet the needs of potential channel partners? How about manufacturing partners?
• What volumes will be required to keep manufacturing and distribution costs low enough to maintain the cost model? Will organic farmer partners guarantee enough produce to meet these requirements?

The final formulation of the proposition, might be something like “organic farmers and ranchers producing all the ingredients needed for several different dog food products, will have sufficient excess inventories at the right times and for the right prices, to satisfy the requirements of the value proposition, key activities, and cost structure, related partnerships, and other aspects of the business model.”

Some of these questions may occur to Joe and Amy beforehand; others emerge during research and can’t be anticipated. Most attempts to validate a proposition lead to a set of new propositions. These sets aren’t infinite, but they’re not immediately knowable either. Coherence-testing the questions leads to finer grained questions.

Testing

Peter Thiel’s advice for venture capitalists “force yourself into a binary mode” applies here too. Propositions are not meaningful if they cannot be invalidated. Daniel Kahneman talks about the “illusion of validity.” People jump to conclusions based on available evidence, however partial that evidence happens to be, and all without recognizing the size of the jump. Worse, our intuitive sense of whether something is true or not responds much more sensitively to a good story than it does to a lack of evidence. If it’s a bad story, it smells wrong and we can tell. If it’s a good story based on bad evidence, it smells fine and we can’t.

With their extremely limited resources, startups need to ruthlessly invalidate. This means figuring out what constitutes evidence, and designing metrics around that evidence, BEFORE you start gathering it. It also means making sure the questions and tests are designed to elicit binary judgments.

In the “customer relationship” box, Joe and Amy have listed generic marketing vehicles. Spending money on any one of these channels will have some result. Some people will make a purchase based on an email link; but that doesn’t make email with links in the right channel. A valid proposition about a marketing channel is one that’s designed in the form of a metric that is tied to a performance level required by other aspects of the model.

Coherence pushes the business model canvas in the direction of greater resolution – behind every question several new questions are hidden. Test-ability is designed to invalidate propositions, and therefore push the canvas in a new direction – toward a somewhat different picture. The two processes work together — invalidated propositions make way for new iterations to be tested, and new propositions call for new rounds of coherence testing. For example, less efficient channels might raise the cost to acquire customers, which might impact manufacturing or packaging contracts, which might raise prices, which might impact the value proposition. The ultimate goal is a high resolution picture of a valid business.

Framing

Broad framing is an important heuristic to support both coherence and testability. When asking yourself about a product of feature, or asking customers or partners, you get better information if you avoid narrow framing “What do you think of X?” and look for broader-framed equivalents. “What do you think of X or Y or Z?” There are two reasons for this – the first is that it’s too easy for you or your team or your potential customers or partners to make a decision and then through rationalize it, rather than to make thoughtful judgments. The second is that multiple options expose nuances, while fewer options hide them.

In Joe and Amy’s canvas, customers are listed as “pet owners.” Will pet owners buy their organic pet food? This is a tough question to validate, and worse, conclusions about its validity are tough to turn into a business activity. Better to test propositions about lots of customers:

• Pet owners in the Northwest
• Pet owners in big cities
• Boarding kennels
• Holistic vets
• New pet owners
• Owners of geriatric animals
• People who shop at Whole Foods and have pets
• People who shop at Costco and have pets

Having a lot of options will guard Joe and Amy against investing too much in validating a single customer segment, and subconsciously putting their thumbs on the scale. Plus, thinking through different types of customers is a great way to narrow and concretize your hypotheses. Validating the proposition that the customers are “pet owners” wouldn’t tell them much about the rest of the canvas. But good evidence that “pet owners in the Northwest who shop at Whole Foods” are interested, reverberates through the rest of the canvas, with implications for manufacturing, pricing, communicating, distributing, and partnering.