Models of Innovation: From the Linear Push Model to Open Innovation

Models of innovation are attempts to describe how an idea becomes a product in use. They matter because the model an organisation holds, usually without stating it, determines where it puts money, who it hires and what it counts as progress.

Roy Rothwell's five generations remain the standard organising scheme, and the useful way to read them is as a sequence of corrections: each generation exists because the previous one predicted something that did not happen.

First generation: technology push

The earliest model, dominant from the late 1940s, runs in one direction. Basic research produces discoveries, applied research turns them into technologies, development turns those into products, and marketing sells what emerges.

It is called technology push because demand plays no part until the end. The implied policy is straightforward: fund science generously and products will follow.

The model was not foolish. It describes the wartime and immediate post-war experience reasonably well, in a period of scientific abundance and unmet demand where almost anything that could be made could be sold.

Its failure is that a great many technically excellent products found no market. The model has no mechanism for discovering what anyone wants, so it cannot explain failure except as a marketing problem arriving too late to fix.

Second generation: market pull

The reaction, from the mid-1960s, inverts the arrow. Market needs are identified first; research and development is directed at satisfying them.

This is the model implicit in most marketing teaching, and it corrects the obvious flaw in the first. It also has a characteristic failure of its own: incremental products. Customers articulate needs in terms of what they already have, so a process driven entirely by expressed demand produces improvements to existing categories and very little else.

The two models together frame a debate that recurs in every generation since. Neither is wrong about its own half. The problem is that each treats one half as the whole.

Third generation: the coupling model

Rothwell and Zegveld's coupling model, from the later 1970s, joins them. Innovation is presented as a sequence of functionally distinct stages that are nonetheless interdependent, with feedback loops between them and with both the state of technology and the state of the market feeding in continuously.

The substantive claim is that innovation is neither pushed nor pulled but results from the matching of a technical possibility to a market need, and that the matching happens repeatedly throughout the process, not once at the start.

This is where the model becomes recognisably modern, and where the stage-gate processes that many firms still run have their origin: distinct phases, defined deliverables, a decision point between each.

Fourth generation: integrated and parallel

Observation of Japanese manufacturers in the 1980s produced the next correction. The stages were not, in the best firms, sequential at all.

The integrated model has functions working in parallel — development, manufacturing engineering and marketing engaged simultaneously and not in a relay — with suppliers brought in early and customers involved throughout. Rothwell's own phrasing is that it is less a sequence than a set of concurrent activities held together by intense communication.

The gain is speed, which by the 1980s had become the competitive variable that mattered. The cost is coordination: parallel work requires far more communication, and the failure mode is expensive rework when the parallel streams diverge.

Fifth generation: systems integration and networking

Rothwell's fifth generation, described in 1994, extends the integration outside the firm. Innovation becomes a distributed process running across a network of firms, suppliers, customers, universities and sometimes competitors, coordinated through information systems and strategic alliances.

The reasoning is that no single firm holds all the knowledge a complex product now requires, and that the ability to combine external knowledge has become a capability in its own right.

Open innovation, and what it changed

Chesbrough's Open Innovation (2003) is sometimes treated as a sixth generation and is better read as a sharper statement of what the fifth implied.

The closed model assumes that a firm's innovation comes from its own research, that it should own the results, and that first to market with its own technology is the route to return. The open model assumes that most relevant knowledge sits outside any one firm, that internal and external ideas should both be used, and that ideas the firm does not want should be licensed out rather than shelved.

Two flows follow. Outside-in brings external technology in, through licensing, acquisition, partnership or scouting. Inside-out takes unused internal technology to market through others, which converts a sunk research cost into a revenue line.

The concept has been enormously influential and attracts two serious objections. The first is that it describes what large firms in a few sectors — information technology, pharmaceuticals, consumer products — already did, and generalises from them. The second is that it understates the capability required: absorbing external knowledge demands substantial internal research capacity, so open innovation is not a substitute for a research function but a way of using one differently.

What the generations do not describe

Two omissions run through the whole sequence and are worth raising rather than accepting.

None of the models distinguishes types of innovation. A model that explains an incremental product improvement is being asked to explain a technology that destroys an existing industry, and the two behave differently. Incremental innovation rewards integrated, customer-coupled processes; radical innovation frequently comes from outside the incumbents entirely, and the customer-pull mechanism that works so well for the first actively suppresses the second, because existing customers do not ask for the thing that will replace what they buy.

None of them explains adoption. Every generation stops at market launch, and the history of technology is full of technically superior products that failed to diffuse. Rogers' diffusion work sits alongside these models, not inside them, and an answer that treats a launch as the end of the process has followed the models faithfully and missed half the outcome.

A third gap is worth noting where the assignment allows it. The models were developed to describe manufactured products, and they fit services awkwardly. A service innovation frequently has no development phase separable from delivery, no prototype, and no point at which it is finished, which makes stage gates difficult to apply and makes the fourth-generation emphasis on parallel working closer to the reality than anything in the first three.

Why the linear model refuses to die

Every generation since the 1960s has contradicted technology push, and it remains the model implicit in a great deal of practice and policy.

Three reasons are usually offered. It is easy to explain, and a model that fits on one slide travels further than one that does not. It matches how funding is structured, with research councils, development budgets and commercialisation grants as separate instruments applied in that order. And it assigns clear responsibility, which interactive models do not: when innovation is a network, no one owns it, and no one can be held to account for it.

The consequence worth naming in an assignment is that organisations frequently operate a linear process while describing themselves as open or integrated, and the gap is visible in the structure — stage gates, sequential handovers, marketing engaged after specification — rather than in the language.

What each model implies for how a firm is organised

Because the models are rarely stated explicitly, the quickest way to identify the one in use is to look at structure, and each generation leaves a different signature.

Technology push shows up as a research function with its own budget, its own site and little routine contact with commercial teams, evaluated on patents and publications and not on revenue from new products.

Market pull shows up as development reporting to marketing, a product roadmap built from customer requests, and a research function that has been reduced to applied work or eliminated.

Coupling shows up as a stage-gate process with named gatekeepers and formal review points, which is still the most common arrangement in established manufacturers.

Integrated shows up as cross-functional teams with co-located members and a single programme manager, and as suppliers holding design responsibility, not build-to-print contracts.

Networked and open show up as licensing and partnership functions, venture arms, and a metric nobody in the first two generations would recognise: revenue from technology the firm developed and chose not to commercialise itself.

The signatures are not exclusive, and most firms of any size show two or three at once in different divisions — which is itself a finding worth reporting, because the tensions between divisions running different models account for a large share of the friction attributed to personality.

Using the models in an answer

Two habits raise the mark.

Name which model the organisation is actually running, using evidence from how it is structured, rather than the one it claims. The claim is in the annual report; the practice is in the reporting lines, in who is in the room at specification, and in the sequence of the budget cycle.

And say what each generation corrected. A list of five models is description; an account of what the second fixed in the first and failed to fix itself is analysis, and it also makes the limitations easy to state, because each generation's weakness is the thing the next one was built to address. That reading also explains the order, which otherwise looks like a chronology of fashions.