04 · The forms of innovation

The oldest distinction measures the degree of discontinuity with respect to what exists. Incremental innovation progressively improves products, processes or services along established trajectories: it exploits knowledge already held, presents contained risk and predictable returns, and constitutes the great majority of innovative activity and of the aggregate economic value that derives from it. Radical innovation introduces a discontinuity founded on new knowledge: high risk, long timescales, high return potential, and the historical function of periodically redefining the trajectories along which incremental activity then proceeds.

Two clarifications avoid sloppy uses of the pair. The classification is relative to the point of observation: the same innovation can be radical for a firm (which does not possess the necessary knowledge) and incremental for the sector (where that knowledge is widespread); the correct question is always "radical for whom?". And the pair is not a scale of merit: a portfolio made only of radical innovation is a collection of bets, one made only of incremental innovation is an annuity running out; the composition is a choice of strategy, not of ambition (chapter 08 deals with the balance).

No term in the discipline is more used and more misunderstood than "disruptive". Clayton Christensen's theory (The Innovator's Dilemma, 1997) describes a precise mechanism, and it is worth reconstructing it before correcting common usage.

Christensen distinguishes two types of innovation with respect to the performance trajectory that the main customers of a market value. Sustaining innovations improve the product along that trajectory, and they may be incremental or even highly radical: on this ground incumbents tend to win, because they have the customers, the channels and the resources. Disruptive innovations, by contrast, enter with a product that is inferior on the traditional dimensions but superior on alternative ones (price, simplicity, accessibility, convenience), and for this reason of interest only to two audiences: the low end of the market, overserved and price-sensitive (low-end disruption), or non-consumers, excluded from the existing product by cost or complexity (new-market disruption). The dynamic point of the theory: the entrant's technology improves faster than the market's requirements grow, and when it reaches the performance threshold that is sufficient for the main customers, migration is rapid and the incumbent's response comes late.

The causal mechanism is not technological but organisational, and this is what makes the theory a dilemma: incumbents ignore marginal segments rationally, because their resource allocation processes (built by listening to their best customers and pursuing the highest margins) correctly classify those segments as unattractive. The practices of good management are exactly what makes them vulnerable: the theme will return in chapter 16, because it is one of the reasons for the existence of the venture builder.

The corrections to common usage, which Christensen himself had to publish (Christensen, Raynor and McDonald, "What Is Disruptive Innovation?", Harvard Business Review, 2015): disruptive does not mean revolutionary, nor high-impact, nor simply successful. A radically better product that enters from the high end of the market is not disruptive in the sense of the theory (the example discussed in the article is Uber with respect to taxis; likewise the first iPhone with respect to the smartphones of the time), and the distinction is not pedantry: the theory makes predictions about the behaviour of incumbents, and the predictions hold only for the mechanism it describes. An incumbent attacked frontally on the sustaining trajectory typically reacts and often wins; an incumbent attacked from below typically does not react until it is late. Diagnosing which of the two games is being played is among the most consequential questions in the competitive analysis of a new company.

The Netflix against Blockbuster case illustrates the complete mechanism. Netflix in 1999-2004 (DVDs by post, subscription, long catalogue, no late fees) was an inferior product for Blockbuster's main customer, the one making an impulse choice on a Friday evening: delivery in days against minutes. It was superior for a marginal segment: film enthusiasts, planners, those who resented late fees. Blockbuster assessed the opportunity (the offer to buy Netflix for 50 million dollars in 2000 is documented by the testimony of those involved) and declined it: a decision consistent with its own numbers, in which late fees weighed significantly on revenue. Streaming, introduced by Netflix in 2007, completed the improvement trajectory until it surpassed the shop on the dimension of immediacy as well; Blockbuster, which did launch belated responses, declared bankruptcy in 2010. Every element of the theory is present: entry from marginal segments, initial inferiority on the dominant dimension, rational refusal by the incumbent, improvement of the entrant beyond the threshold, rapid migration.

Business model innovation. The same technology, commercialised with different models, produces returns that differ even by orders of magnitude: this is the thesis documented by Teece (2010) and Chesbrough (2010), and it makes the business model (the logic by which the organisation creates, delivers and captures value) an autonomous design variable. The typological forms are well known: from ownership to subscription, from product to multi-sided platform, freemium, razor and blades. The conceptual point to retain: business model innovation does not require new technology, and new technology does not produce value until a business model captures it; chapters 09 and 11 will develop the financing side and the appropriation side respectively. The Netflix case itself is, as well as a disruption, a model innovation (subscription without late fees against rental with late fees) that arrived before the technological discontinuity of streaming.

Innovation in services. Services dominate advanced economies but innovation theory was born on manufactured goods, and the differences matter. A service is co-produced with the customer (quality depends also on the customer's behaviour), it is often intangible and cannot be stored (unsold capacity is lost), and its innovation is typically one of process, organisation and model rather than of product in the strict sense: for this reason it is harder to protect with intellectual property (little to patent) and easier to imitate in form, though not in execution. The defensibility of a service innovation usually lies in operational execution, in accumulated data, in scale and in the brand rather than in legal exclusivity: a difference that chapter 11 will take up when dealing with appropriability regimes.

The box on general purpose technologies. Some technologies do not belong to a sector: they cut across all of them. The category (Bresnahan and Trajtenberg, 1995) is defined by three properties: pervasiveness of uses, continuous improvement over time, and the capacity to generate complementary innovations in the sectors that adopt them. Steam, electricity and computing are the historical cases; contemporary artificial intelligence exhibits the three properties recognisably. Two historical lessons about GPTs deserve to be fixed. First: economic value materialises with long delays, because it requires organisational co-inventions; the electrification of factories produced its productivity gains decades after the dynamo, when plants were redesigned around the individual electric motor instead of replicating the layout of steam transmission shafts (the case is documented in the studies of Paul David). Replicating existing processes with the new technology is the initial and least productive phase; redesign around the technology is where the value lies. Second: during the diffusion of a GPT, much of the value goes to those who apply it to the vertical problems of individual sectors, not only to those who produce the general technology; for anyone building new companies, the consequence is that the correct question is rarely "shall we do AI?", and almost always "which vertical problem, validated according to chapter 02, does this GPT make solvable now that it was not before?".

• R.M. Henderson and K.B. Clark, "Architectural Innovation" (Administrative Science Quarterly, 1990): the original study on photolithography; the opening theoretical section is the part to read most carefully.
• C.M. Christensen, The Innovator's Dilemma (1997): the complete theory of disruption with the original cases (hard disks, excavators, steel).
• C.M. Christensen, M.E. Raynor and R. McDonald, "What Is Disruptive Innovation?" (Harvard Business Review, December 2015): the author's correction of improper uses of the term; twenty minutes well spent.
• T.F. Bresnahan and M. Trajtenberg, "General Purpose Technologies: Engines of Growth?" (Journal of Econometrics, 1995): the article that defined the category; for the historical lesson on electrification, P.A. David, "The Dynamo and the Computer" (American Economic Review, 1990).