The innovation process is not a pipeline

The default picture of innovation is a pipeline. Ideas enter at one end, finished products emerge at the other. Between entry and exit sit stages, gates, reviews and milestones. The image is clean enough for a board presentation. It is also wrong in ways that matter.

Pipeline thinking encourages organisations to manage for throughput. Move the idea forward. Complete the gate. Hit the milestone. The problem is that innovation does not behave like manufacturing. The raw material is not well understood at the start. The specifications change as you learn. The customer you imagined at the beginning may not be the customer who cares at the end. Treating this as a logistics problem produces bureaucracy dressed as discipline.

That does not mean structure is the enemy. It means the wrong kind of structure is.

What the process actually contains

The innovation process is the full set of activities through which something new or improved is recognised, developed, tested, resourced, implemented and brought into use. That includes technical work, commercial work and organisational work. It includes product innovation, which is what most people think of, but also process innovation: changes to production methods, delivery systems, internal routines and management practices. In many industries, process innovation creates more value than product novelty. It just attracts less attention.

Two things follow from this definition. First, brainstorming is not innovation. It is one moment inside a much larger effort. Second, launch is not the finish line. An innovation is not complete when it ships. It is complete when it is actually used. That distinction sounds pedantic until you notice how many organisations celebrate launches and then abandon the learning that only real use can provide.

Why linear models persist

Stage-gate systems remain popular because they solve a genuine governance problem. Organisations need to allocate capital, set priorities and kill weak projects before they consume too much resource. A structured sequence of stages and decision points does that reasonably well.

The trouble starts when governance is confused with description. A stage-gate map tells you where the decision points are. It does not tell you how the work actually moves between them. In practice, development sends teams back to research. Production problems reshape the design. Market feedback rewrites the value proposition. Kline and Rosenberg made this point decades ago: innovation does not travel in one clean chain from knowledge to market. It loops, and the loops matter.

Van de Ven went further. His account of the innovation journey describes a nonlinear system of divergent and convergent activity, setbacks, shifting coalitions, revised criteria and partial restarts. That is not a failure of process. It is a feature of working under real uncertainty. Anyone who has watched a serious product or organisational change effort unfold will recognise the pattern immediately.

Pipeline logic versus learning logic

The distinction that clarifies most is between pipeline logic and learning logic.

Pipeline logic asks whether the team has completed the stage. Learning logic asks what the team has learned, which assumption has been tested, and which uncertainty is still dangerous to carry forward. The first optimises for orderly progression. The second optimises for the quality of understanding at each commitment point.

Experimentation-based approaches work from the second logic. Lean Startup compresses the process into build, measure, learn. Design thinking iterates through desirability, feasibility and viability by testing and reframing rather than executing a single pass from brief to solution. Thomke's work on experimentation makes the same argument in more operational language: firms improve by learning early and often, not by pretending they know enough at the outset.

These are not replacements for stage-gates. They are what stage-gates should contain. A gate is most useful when it asks what has been learned, what remains unknown, and whether the remaining uncertainty justifies the next level of commitment. Used that way, governance and experimentation are not opposed. They are the same system at different scales.

The tensions that shape the process

Several tensions sit permanently inside any serious innovation effort.

Exploration versus exploitation is the most structural. Organisations must search for new possibilities while refining what already works. Too much exploration and the firm becomes inventive but economically thin. Too much exploitation and it perfects yesterday's logic. March identified this trade-off in 1991 and it has not softened since. Innovation process design is partly the art of holding both activities in productive tension rather than letting one crowd out the other.

Product versus process innovation is the most neglected. Popular discussion still treats innovation as new things for customers. But changing how a firm produces, delivers, coordinates or decides can shift performance more durably than a new product launch. A better permit workflow, a redesigned supply chain, a faster internal decision cycle: these are innovations. They deserve the same rigour.

Idea generation versus commercialisation is the most misunderstood. Ideas are not scarce. The ability to turn one into something robust enough to be adopted, paid for and sustained is scarce. That is why the process does not end at launch. For many innovations, the highest-quality learning arrives only after real users, operators and organisational routines start interacting with the thing you built. Before launch you have hypotheses. After launch you have evidence. Treating post-launch as an afterthought discards the most expensive information the process can produce.

What follows from this

If the innovation process is a learning system, then the quality of an organisation's innovation capability is not measured by how many ideas it generates or how smoothly projects move through gates. It is measured by how well it learns at each stage, how honestly it confronts remaining uncertainty, and how effectively it translates what it discovers into better decisions about where to commit next.

Poor systems learn too late, after the organisation has already spent enough money, credibility or political capital that turning back feels impossible. Better systems force learning forward, concentrating the most consequential tests early, when failure is still cheap and useful.

The real question is not whether innovation should be structured. It should. The question is what the structure is for. If it exists to simulate certainty, it will become bureaucracy. If it exists to organise learning, it becomes a genuine strategic capability.

That is the better way to think about the innovation process. Not as a pipeline that converts ideas into outputs, not as a romantic space for creativity, but as a disciplined sequence of learning and commitment through which new products and new ways of working become real enough to use.