DisruptiveInnovation.Org - Latest Discussions

Measuring Modularity

Horace Dediu, 11/15/2016


(This is Part 5 of a multi-part series by Horace entitled "Modular Revolution." The first entry is here, the second is here, the third is here, the fourth is here.)

If you were to plot the history of technological adoptions you would get a series of diffusion curves. Each curve begins with an invention, a period of gestation as the idea is built into a product and, if the product solves a need, the formation of a market. If the technology is widely useful then the products using it are “adopted” by a majority and the technology “diffuses” into the general population. Ultimately the technology is universally accepted and it “saturates” the addressable market.

Technological diffusions have been observed for at least two centuries. The earliest were for transportation canals built in 18th century England. In the 19th century, as industrialization took hold, diffusions became more common with transportation and communication alternatives. In the 20th century new enablers in the form of motors, fuels and production methods led to increasing consumerization and miniaturization. Today new technologies seem to appear and spread very quickly.

But are technologies being adopted more quickly? Has the speed of adoption matched the speed of introduction? Do we put ideas to use as quickly as we invent them? Are all technologies quick today? Will they be quicker in the future? Were they always slow in the past?

The historic data is sometimes contradictory. There were many older technologies which grew very quickly. In 1930s America, the radio was adopted very quickly. So was the TV in the 50s. Conversely, there are many technologies today which are slow to rise. Electric cars are diffusing slower than the Model T, new forms of energy production are growing more slowly than the original electric grid. Healthcare and education reform is far slower than the rise of hospitals and primary/secondary schools in the 19th century.

To test this further we broadened the sample to 88 US diffusions. The data shows significant variations in adoption speed for technologies that got started around the same time. For example the Kodak Brownie camera and the automobile both reached 10% of their markets around 1915. However the democratization of photography took 20 years while the automobile took over 70 years. Refrigerators and TVs were quick while their contemporary washing machines and dishwashers were slow.

Even toward the late 20th century and the rise of transistor electronics, the VCR was quick but the video game console was slow. The early 21st century is awash with internet/computing-based technologies which rise very quickly but there are innovations which seem slow and stubborn. We are facing a crisis of transformation in healthcare, education, energy, banking, transportation and government. Manufacturing and agriculture might also be reaching low-growth crises.

It's tempting to suggest that Moore’s Law, which correlates nicely with computing innovation rates, will solve these crises. But there seem to be some technologies which are disobeying the law. It we look back far enough we can see that well before Moore’s Law, there were other technological “laws” which led to rapid improvements in performance. In the 19th century steam engine efficiency grew exponentially. In the early 20th century innovations related to internal combustion and turbines meant that vehicle speed grew exponentially. In the mid 20th century, the miniaturization of electric motors meant their specific output increased exponentially. Steam, internal combustion, electric motors were all epoch-making enablers but, during those same epochs there were anomalous laggards. Not all diffusions conform to technological law governance.

To solve the puzzle we compared products with similar enablers but differing market development. In this pairing we saw a pattern rapid growth was correlated with the absorbability of the innovation in the adopter’s life. We call this “conformability” to user behavior. For example, the refrigerator rose quickly in popularity because it was easy to fit into a kitchen while a contemporary washing machine was slow because there wasn’t anywhere to put it in an urban apartment. We could test conformability by asking whether the adopter faced a minimum of five independences: purchase, assistance, time, space and learning.

We also observed a pattern of increasing granularity of value networks among the rapid diffusions. We call this “collaborative” producer behavior. Smartphones rose quickly because they leveraged modular software, components, distribution via network operators and pre-existing internet content while electric cars rose slowly because they require integration, defeat current distributors’ business models and need to be driven on roads built without charging infrastructure. We tested collaboration by asking whether the producer could recruit ecosystems, obtain network effects, distribution, supplier networks and leverage existing infrastructures.

In combination, conformability and value network creation were powerful accelerants. This has led us to build a construct stating that creating independence of purchase simultaneously with a dependence on partnership leads to rapid adoption. We thus define a “modular business architecture” as the combination of conformable demand creation and collaborative demand fulfillment. The duality of independence of purchase and dependence of supply explains the push/pull or market creation. We think of this as “modularity” since the product or service acts as a module which slots easily into an adopter’s world while simultaneously allowing partners to couple it to new business models that create additional points of value capture. The product has few dependencies as far as the buyer is concerned but has many dependencies as far as the seller is concerned.

To test the hypothesis we devised a way to score these 88 technologies on modularity and the results are shown in the graph below. This association between speed and modularity yields no anomalies. In other words it lets us hypothesize that there are no slow modular and no fast inter-dependent adoptions.

To test the model we applied it to 11 emergent technologies and used the conformability and collaborative tests to estimate adoption rates. We compared consensus opinions and modularity analysis. The results are in the table above. Six consensus estimates fell within the range our model predicts. Five did not. Wearable computers and cryptocurrencies model quicker than expected while 3D printing, electric cars and photovoltaic power appear to be slower.

Network effects and ecosystems tend to be underestimated which is why we expect wearables and cryptocurrencies to be quicker. On the other hand, physical infrastructure inertias are also underestimated. Large-scale industrial 3D printing dependencies on new materials and replacement of production methods currently in use, requiring re-learning and adaption will delay adoption. Electric cars suffer from fueling infrastructure, behavioral changes, ecosystem issues (grid capacity), a lack of network effects and distribution questions. Photovoltaic power lacks in all aspects of conformability to the current power production systems and does not benefit from ecosystems or network effects to offset this.

The methodology we propose is meant to shine a light on the deficiencies of the laggard technologies in order that they be remedied. There is no inevitability in the model. It is the actions of entrepreneurs and policy makers which ultimately determine which conditions prevail and cause the adoptions to accelerate. If the causes for delay are understood then action follows.