There’s a problem with how we commonly approach business model innovation. We tend to treat it as though it were an art. We try to innovate with more creative ideation, stronger phase gates or smaller trial-and-error cycles — and it’s not working.
This post is a companion to my O’Reilly Strata talk on making business model innovation more of a science than an art. I’m studying ways that data science can improve business model innovation (here’s the latest research paper).
What we’ve found is that our ability to spot and nurture the good ideas is not keeping up with our ability to generate and share bad ideas (see Figure 1).
What we need to climb out of this hole is a new innovation engine, and computing is the perfect fuel. Computing has gotten better as it’s gotten cheaper. As it stands, computing is the cheapest, most abundant resource that we can deploy against any problem. Why not use computing to turn business model innovation into more of a science? Something becomes more of a science when you start basing your decisions on better, more objective data. We can use stochastic simulation to generate future “what if” scenarios and use those scenarios as the basis of our plans to innovate.
The big idea here is that innovation is actually a technical and scientific problem. It’s the ability to systematically identify, prioritize and eliminate risks. Innovation is just what’s left over after you’ve eliminated all the ideas that won’t work. The key to making the big idea work is to think of risk management as a learning process rather than as a safety exercise. We have to approach managing risks as a series of controlled experiments. After we run each experiment, we learn something new.
For the purposes of managing innovation risk, we have to replace retrospective models. These models try to calculate the future based on past trends. Models like that take you away from breakthrough innovation and keep you stuck in optimizing. Instead, we need to build stochastic simulations, or prescriptive models. Take your business and model it using rules. In this research so far, the rules are in the form of simple, one-variable equations (see Figure 2). Add randomness into the rules to give you a good idea of where your risks might be. Run the simulations and generate prescriptive data. The data represents your “what if” and “what might be” scenarios.
Have faith in the power of simple models. It’s natural to think that complex systems have to be represented by complex models. But often they don’t. Simple approximations tend to be the most resilient when you are trying to generate plausible scenarios about something as uncertain as new innovations.
Let’s take this method out for a spin. Let’s rewind the clock back to 1998 and use prescriptive model simulation to manage innovation at Blockbuster. Does the video-by-mail business model pose a significant threat? I built a stochastic simulation using the model technique and one-variable equations I mentioned earlier. For each equation, I made simple assumptions (see Figure 4) about how to represent both business models. I ran hundreds of simulations and generated lots of interesting data.
Figure 5 shows revenue simulation results. The data is summarized into a box plot which makes it easier to see aggregate features like the expected, minimum and maximum performance and outliers. The simulations tell us that Blockbuster will start out in a commanding position. In 1999, Blockbuster was the market leader. They had $4.5 billion in net revenue, 7,000 locations worldwide, and 3 million videos rentals a day. Meanwhile, Netflix introduced a business model of Internet video rentals shipped by mail. The simulations tell us that revenue for the video-by-mail model will probably start very small. In 1999 Netflix did only a fraction of Blockbuster’s volume and made only a fraction of its revenue. So far, it doesn’t look like the video-by-mail model is that big of a threat. The maximum expected revenue for the Netflix model is much less than the minimum expected revenue for Blockbuster. In 2000, Blockbuster seemed to reach this same conclusion when they made it clear that they considered movie theaters (not video-by-mail) to be their primary threat.
We still have more experiments to run. Those experiments are going to tell us that, eventually, the video-by-mail model will become a credible risk. Figure 6 shows profit margin simulations. The timeline is the same as in Figure 5, but the Y-axis now shows simulated profit margins instead of revenue. The experiments tell us that there is a ceiling to the amount of revenue you can expect to squeeze out of the Blockbuster model. As you invest more in the business, the revenue limit puts profit margins at risk. In 2002 Blockbuster reported a 29% loss. This same drop is predicted by simulation, but for 2003. Experiments tell us that the Netflix threat isn’t going away. In 2003, Netflix reported its first profit. Simulation shows an outside possibility of this happening in 2003, but that it should be expected by 2004.
Our last set of experiments tells us that, eventually, the Netflix threat will be fatal. Figure 7 shows us simulated competition between Blockbuster and Netflix. The Y-axis shows a measure of survivability for each. Think of it, roughly, as being proportional to the likelihood that each business model will survive. The simulations show an expected decline by Blockbuster starting in 2001. In 2002 Blockbuster reported a 29% loss. In 2010 Blockbuster filed for bankruptcy. We were able to learn a lot about the video-by-mail risk through simulation. And what we learned brings us to a meaningful innovation.
To recap, it’s 1998 and we are in charge of innovation at Blockbuster. We’ve studied the risk of video-by-mail and come to an interesting conclusion. Two years from now (in 2000) when Netflix offers to sell their company…write ‘em a check! Purchasing Netflix minimizes a big risk and represents a significant innovation.
And that’s business model innovation as a science: build stochastic simulations, generate experiments, and use those experiments to minimize risk and innovate in the process. This work is part of the advanced analytics research I conduct at CSC. I study ways to innovate better and I love to collaborate. I’ve built a stochastic prescriptive scenario generation machine, and it’s hungry. I need interesting business model ideas to feed it.
Jerry Overton is head of advanced analytics research in CSC’s ResearchNetwork and founder of CSC’s FutureTense competency, which includes the Predictive Modeling Research Group, Advanced Analytics Lab and Predictive Modeling School. Connect with him on Twitter.