I recently asked the CEO of a startup, one who is thoroughly an effectual thinker, what he thought of building an economic model similar to that found in "Developing Products in Half the Time." The answer was that he would not believe the model. I then asked a product manager of a company servicing a mature market the same question, and the answer was that not only do they build models, but they drive the model from data from past projects and industry analysis, and the CEO hammers every corner of the model until they believe it represents reality. Only then is a project approved. This CEO is a hard core causal thinker.
I was not at all surprised. The effectual thinker knows that data and assumptions are suspect and constant feedback is more reliable. The causal thinker knows that if the data is good, they can make better tradeoffs. Each has its place.
Regardless of ones thinking tendencies, a lot can be learned by experimenting with a model. By playing with assumptions and their effect on the economics, one can get a feel for which assumptions need better testing and how they relate to development tradeoffs.
I'll walk through and example and see what we can learn. Here is the story: we are a capital equipment supplier selling manufacturing equipment to electronics manufacturers in the US and Asia. The market size is $100M per year, our product is priced at $200K, and the market grows at 8% per year. The product is targeted at a newly developing market segment.
Let's get started on a model. The first thing we need to do is model revenue. Because we are targeting a new segment, we will model the market and sales separately. The market model has two components: the available market and the diffusion of the new product. The available market is:
Available = Market Size X Quality X Awareness X Buy.
The Quality factor is used to model the effect of entering the market and then improving quality. This is a simple way of accounting for learning through market experience. Awareness is one of the A's in the classic ATAR model. Buy models the fact that there are substitutable solutions and this is the portion of sales that can be expected.
The diffusion effect is the standard model that uses a Trial Rate and a Copy Rate.
Let's build out a spreadsheet for this and talk about its assumptions:
Market size is in units sold with an 8% growth rate. The market will only buy half as many systems in the first year and four years later quality does not matter. Awareness starts out at 25% and ends at 100% in year 7. The assumption is that there are many small customers all over the world that are hard to find. 60% of the market will buy this type of solution and 40% will by substitute solutions or adapt what they already use. The diffusion model uses a trial rate of 10% and a copy rate of 20% The adopter column displays the number of systems per year that the market will buy. The penetration column shows the total systems sold. At year ten 350 systems will be in use.
The important assumption is that there are multiple competitors entering the market at the same time, and this represents how the addressable market will play out as a whole. We now have to model how much of this market we can take:
This model says that we can make 80% of the addressable market aware of our product, but initially we can only support sales to 30% of the available market. By year 4 we can support 100% of the available market. Our sales conversion rate is 30%, which is a way of modeling market share. The model says we can take 30% of the market we can reach. The Quantity column represents repeat business on average for this type of product. Sales ramps up over the first few years as customers gain experience with the product in production and purchase risk decreases. The end result is a number of units sold.
Multiply these models together and with sales prices and we have a revenue model for the base case:
The goal is to model four sensitivities: COGS, Development Cost, Performance, and Delay to Market. We can model performance reduction and market delay in the revenue model. Performance is modeled by reducing the Buy column of the market model by 10%. If the performance is lowered, people will turn to substitutes. We could also model this in the sales conversion rate, but using Buy simplifies the model. Market delay is modeled by delaying availability one year and lowering the sales conversion rate to reflect a loss of market share. The assumption is that a one year delay will cause a 33% reduction in market share. If you play around with the spreadsheet, it becomes clear that it is loss of market share that causes the largest loss. If this is not the driving factor in a new product, you can model the delay in other ways. For example, if your competitor can lock up the input value chain you can model higher COGS. Here are all three models side by side:
The left graph is 10% performance reduction, the middle graph is the base case, and the right graph is the one year delay to market. Clearly in this case delay to market has the bigger effect.
Let's now model revenue and look at the other to sensitivities:
COGS is set to 55%, thus a 45% margin, which is a very conservative number. We model higher production costs by raising COGS 10%. Our SG&A is 25%. Development cost is $1M, and there is a $100K yearly development cost associated with product improvements. A cost overrun is modeled as a 10% increase in development cost. We use a 30% tax rate and make an approximate cash flow projection. We then calculate a IRR of 44%, a NPV, etc.
A note on the model: the base values are setup in a table so we can tweak the base assumptions.
The sensitivities are handled with switches:
Each sensitivity is tested using the switches and the result is graphed:
A one year delay to market has the greatest effect followed by manufacturing cost. Performance impact is much smaller and development cost is very much smaller! However, what gets a lot of attention in the heat of development? Development cost, especially if you are in a start up. Sometimes you just don't have the cash and you have no choice but to starve development. But, if your product development looks like this, your decisions should reflect it. This is the basis of product development in all Reinertsen's books on Lean Product Development.
How realistic is the model? Like most models the market model is the most difficult, and market share the hardest number to estimate. If you have developed similar products, you can use analogies. Performance requires some guess work, but development cost and manufacturing cost are usually fairly accurate, and when they are not, you can improve the numbers as the project develops. We can deal with the market delay estimation by testing, but guess what? You have to at least have a prototype product to do that? You can do better by concept testing before engineering development. There is no other choice than to do the best you can and refine the model as you go along.
Much of the value of economic modeling is in the thinking process. Are you a fast follower? Does the late entry effect market share? Does it effect development cost? Is this a winner take all market? How does performance affect sales? Even if the model is not perfect, you will probably gain a first order approximation of what is driving cost and what should be managed. The point is that your intuition may be inaccurate, so using models will test it. If you can build the model as a cross functional team, you will have a much better model. Modeling as a team forces you to justify assumptions and reach common agreement on how to manage the cost of product development so that everyone is aligned and not working at cross purposes and different assumptions on cost.
So, if this were your model, what would you do with it? What tradeoffs would you make? How would it affect your process and decision making?
Note: if you want a copy of the Numbers spreadsheet used in this post, send me an e-mail. See the contact page of the Website.