Innovation funnel: the stages, the model, a worked example, and the math that decides the output

An innovation funnel is a model of how a wide set of raw ideas narrows, stage by stage, into the few that actually get funded and launched. It is drawn as a funnel because the input is deliberately much larger than the output: most ideas are supposed to die, and the value of the structure is that they die early and cheaply rather than late and expensively.

The picture comes from Steven Wheelwright and Kim Clark, whose 1992 book Revolutionizing Product Development first drew product development as a funnel. Their point was portfolio-level: you cannot manage new product work one project at a time, because the interesting question is not whether any single project is healthy but whether the whole set of bets is balanced and whether enough is entering to sustain the mortality rate.

That framing is still the most useful thing about the model, and the most commonly ignored. Teams inherit the funnel diagram, spend their effort on the narrowing mechanics in the middle, and never ask the question the shape is actually posing: is enough going in at the wide end to justify how much we throw away?

What is the innovation funnel?

The innovation funnel is a way of describing innovation as a filtering process with known losses. Ideas enter at the wide mouth, pass through a series of screens that each remove a large fraction, and a small number emerge as launched products or implemented changes. Each screen costs more than the one before it, because each one is applied to work that has had more invested in it.

What distinguishes a funnel from a simple pipeline is the ratio. A pipeline implies things flow through; a funnel says explicitly that most things will not. That is not a defect being tolerated, it is the design. If ninety percent of what enters your funnel reaches launch, you are not running a funnel, you are running a queue, and the likely explanation is that only safe, pre-approved ideas were ever allowed in.

What are the stages of the innovation funnel?

Stage names vary by organization, but almost every version collapses to the same five movements. The table below is the shape that matters, along with the rough attrition at each step and the way each one fails in practice.

The mechanics of the decision points between these stages are covered in more depth in our guide to the stage-gate process, which is the most common way organizations formalize the screens.

How many ideas do you need at the top of the innovation funnel?

Far more than most teams assume. The most cited figure comes from Greg Stevens and James Burley, whose 1997 study in Research-Technology Management was titled, bluntly, "3,000 Raw Ideas = 1 Commercial Success." They assembled success curves from three independent sources: the project literature, patent data and experience, and venture capitalists. The three sources agreed to a degree that surprised them.

Their breakdown ran roughly: 3,000 raw ideas produce 300 that are worth shortlisting, which produce 125 small projects, which produce 9 that become significant developments, of which about 4 get close to launch, 1.7 actually launch, and 1 succeeds commercially.

Treat the precise numbers as a period artifact rather than a law. They come from industrial R&D in the 1990s, the definition of a raw idea is elastic, and a modest process improvement in a well-understood market obviously does not need three thousand candidates. What survives scrutiny is the shape: the losses compound, and they compound multiplicatively. Improve your screening accuracy by twenty percent and you improve the output a little. Double the quantity and variety entering the funnel and you roughly double the output, assuming the screens stay honest. That asymmetry is the single most practical thing the model tells you, and it is the opposite of where most organizations spend their innovation effort.

What is the difference between a closed and an open innovation funnel?

A closed funnel has solid walls. Ideas originate inside the company, are developed inside it, and reach the market through the company's own channels. Anything that does not fit the current business gets shelved, which is how large firms accumulate warehouses of unused patents.

Henry Chesbrough used the same funnel drawing to make his argument in Open Innovation (2003), simply by making the walls porous. In an open funnel, ideas and technologies flow in from outside at any stage: universities, startups, suppliers, customers, acquisitions and licensing. They also flow out at any stage, through licensing, spin-offs and joint ventures, so a project that does not fit your business can still create value rather than sitting on a shelf.

The practical implication is not "be open" as a slogan. It is that the top of your funnel is not limited to what your own employees think of, and the exit from your funnel is not limited to your own product line. Both of those relax a constraint that closed funnels treat as fixed.

What is the difference between an innovation funnel and a stage-gate process?

They describe the same thing at different altitudes and it is worth being precise, because teams routinely argue past each other on this. The innovation funnel is a portfolio-level picture: it describes the shape of the whole flow, the ratios, and the fact that input must exceed output. The stage-gate process is a project-level mechanism: it defines what a specific project must deliver, who reviews it, against what criteria, and what the four possible verdicts are.

You can run a stage-gate process without thinking in funnel terms, and most organizations do. The result is a well-governed process with a starving intake, where the gates work correctly and there was simply never much to choose between. You can also think in funnel terms without formal gates, which works fine at small scale and stops working the moment there is more money at stake than one person can sensibly wave through.

Innovation funnel example

Take a mid-sized US software company running one funnel per year. The intake stage collects 240 ideas: roughly half from an employee campaign, the rest from support ticket themes, sales loss reasons and two facilitated sessions with the product and customer success teams.

The first screen is deliberately fast. Two people spend a day and a half against three published criteria: does it serve a segment we have decided to serve, could a small team make progress in a quarter, and is there any evidence a customer wants it. That removes 205 ideas and leaves 35. Nobody writes a document to kill an idea at this stage; a one line reason is recorded so submitters get an answer.

The 35 go to a concept round where each gets a page: the problem, who has it, roughly what it would take, and what would have to be true for it to matter. Eleven survive. Six of those get a real business case with customer conversations attached, and four of the six hold up. Two are funded immediately, two are explicitly put on hold with a named revisit date, which is an honest verdict that most organizations avoid because it sounds indecisive.

Both funded projects reach validation. One ships. The other is killed after a paid pilot shows that the buyers who said they wanted it would not pay the price that made it viable, which is a good outcome that felt like a failure to everyone involved. So: 240 in, 1 shipped, roughly 240 to 1. That is a healthy ratio for incremental work in a known market, and it is only healthy because the first screen was cheap. Had those 205 ideas each consumed a business case, the same funnel would have cost a year of the organization's attention to produce the same single launch.

How do you measure an innovation funnel?

Four numbers tell you almost everything, and they are more useful as trends than as absolutes.

Intake volume and variety. Count what enters, but also count how many genuinely distinct directions it represents. Two hundred submissions that are all variations on "improve the onboarding email" is one idea with a large fan club, and a raw count will hide that.

Conversion rate by stage. Watch each screen's pass rate over time. A screen passing eighty percent has stopped filtering. A screen passing two percent is either fed badly or set impossibly, and both are worth knowing.

Cycle time per stage. Ideas that sit unreviewed for four months teach the organization not to submit. This metric predicts next year's intake volume better than any engagement campaign.

Post-launch outcome. The only number that validates the whole apparatus, and the one most consistently skipped. If nobody checks whether the launched thing hit its case, the screens have no feedback and cannot improve. Doing this properly means actually measuring what users did after launch rather than relying on the sponsor's recollection at the next review. Keeping the surviving-but-unfunded ideas visible between cycles is its own discipline, covered in idea backlog management.

Why do innovation funnels fail?

Three failure modes account for most of it, and only one of them is about process quality.

The first is a thin intake. The funnel is drawn wide at the top and run narrow, because idea collection is treated as always-on and therefore owned by nobody. The symptom is a review meeting where everyone quietly agrees the submissions are small, safe and obvious. No amount of workflow improvement fixes this, because workflow only sorts what arrives.

The second is screens that do not screen. Gates that never kill anything are usually described as collaborative. What they actually do is push the kill decision downstream to a point where it costs twenty times more, or worse, let everything through at reduced funding so that nothing has enough resources to succeed.

The third is a homogeneous intake, which is the subtlest. Two hundred ideas from people who share an office, a job function and a set of assumptions will cluster tightly, and a funnel cannot select for diversity that never entered it. This is why the structured techniques matter: SCAMPER forces transformations a group would not reach unprompted, and Six Thinking Hats forces the same challenge through caution, optimism and process in turn. Our write-up on why group brainstorming fails covers the anchoring mechanics behind it.

How to widen the top of the funnel without lowering the bar

The objection to widening intake is always the same: more submissions means more review time. That is true only if every additional idea gets the same treatment as the ones you already had, and it is why the cheap first screen matters so much. A one-day screen against three published criteria scales to several hundred ideas. A business case does not scale past a handful.

So the practical sequence is: generate widely, screen cheaply and publicly, and only then spend real analysis on survivors. Generating widely is the part that has changed most recently, because producing a hundred genuinely different angles on a challenge no longer requires booking a facilitator and a room. Typing one challenge into the studio on our idea generator returns dozens of tagged directions in about thirty seconds, each labelled with the angle it came from so you can see at a glance whether you have ten directions or one idea in ten costumes.

Then converge before anything enters the formal funnel: cluster the output into named themes and score it on impact against effort, so what reaches your first gate is already a ranked shortlist with reasoning attached. If you are evaluating platforms to run the governance half, we keep verified US list prices on idea management software pricing, and the category-level comparison sits on innovation management software.

The one thing to take from the model

The funnel is a picture of loss, and the loss is not a problem to be solved. Trying to make it narrower by admitting fewer ideas produces a tidy process with nothing in it. The organizations that get real output from the model treat the wide end as the controllable variable and the mortality rate as fixed, which is precisely backwards from how most innovation programs are actually run and improved.