Building an innovation process that produces decisions instead of idea backlogs

The central idea

An innovation process is defined by its exits, not its intake. If there is no rule that ends an idea, the pipeline becomes storage and every review becomes a status update.

Most internal innovation systems are built intake-first: a submission form, a portal, a campaign to encourage ideas. Within a year the organisation has several hundred entries, a committee that meets monthly, and a growing sense that nothing is moving. Nothing is moving because nothing is allowed to stop. Each idea has a sponsor who would rather it stay open than be recorded as rejected, and a process with no exit criteria has no way to overrule that preference except by attrition.

What changed, and why it matters now

You can diagnose this without reading a single idea. Look at the ratio of items entering a stage to items leaving it in any direction. A healthy stage has a high exit rate, most of it downward: ideas killed, merged, or parked with a date. An unhealthy one has entries and almost no exits, which shows up as an average age that climbs quarter over quarter. When the average age of an open idea exceeds a year, the committee has stopped making decisions and started maintaining an inventory, whatever the meeting minutes say.

Build the operating model

Define each stage by the single question it answers and the evidence that would settle it, then attach a kill rule and a funding step to every stage boundary.

The evidence requirement is what makes the kill rule enforceable. A stage that asks whether anyone wants this should specify what would count as an answer — a number of qualified conversations, a stated willingness to pay, a signed pilot — and the amount of money released for gathering it. Without a stated evidence bar, every review becomes a debate about conviction, and conviction always favours whoever is most senior in the room. With one, the debate is about whether the evidence was produced, which is a question a junior person can win.

Measure what the decision produced

Measure the kill rate per stage, the average age of open items, the cost of reaching each decision, and the proportion of launched work that survives its first year.

Kill rate is the diagnostic that most organisations refuse to track, because it reads as failure. It should be read as throughput: a first stage that kills four in five is working correctly, and one that kills one in five is either receiving unusually good ideas or not deciding. Cost-per-decision matters more than cost-per-idea, since the purpose of early stages is to reach a confident no cheaply. Survival at twelve months is the only measure that tests whether the gates were calibrated or merely strict.

Where execution breaks

The main failure is a process that filters for articulacy rather than evidence, promoting whoever presents best and starving work that is harder to narrate.

The second failure is uniform funding. Every idea receives the same small sum at the same cadence, which is administratively simple and strategically useless: it is too little to test anything expensive and too much to spend on things that should have been stopped. The result is a portfolio of half-tested options, none of which has enough evidence to justify a real commitment. Funding should be uneven by design, and the unevenness should be visible, because that is the mechanism by which a portfolio expresses a view.

What this looks like in practice

A working process is usually smaller than expected. Three stages, not seven. A short written kill rule at each boundary that a person can apply without convening anyone. A single monthly forum whose only outputs are fund, kill, or park-with-a-date — no other verb permitted. Ideas that are parked carry an explicit review date and disappear from the active list, which is what keeps the list readable and the meeting short enough that decisions actually get made in it. The forum itself needs a rule about who speaks. If the sponsor presents, the discussion becomes advocacy; if an analyst presents the evidence and the sponsor answers questions, it becomes assessment. That single change in staging does more for decision quality than any scoring matrix, because it separates the person who wants the outcome from the person describing what is known, and it makes a no survivable for both of them. The second staging rule is that the kill rule is read aloud before the discussion, not after it. A rule recited at the end sounds like a justification for a conclusion already reached; the same rule read first frames what the group is looking for, and it makes the rare override visible as an override rather than as a judgement call.

The strongest argument against this

The reasonable argument against hard kill rules is that genuinely novel work often fails early tests. Evidence bars are calibrated on things that resemble what the organisation already does, and the ideas most worth pursuing are precisely the ones with no comparable precedent to be measured against. Applied rigidly, a stage-gate system will reliably eliminate them and leave a portfolio of safe incremental improvements that all pass their gates.

The mitigation is a small explicitly ungated allocation, kept deliberately modest and separated from the main process rather than smuggled through it as an exception. Naming it prevents the more common outcome, which is that every idea eventually claims to be the exceptional case. A process that grants exceptions informally has no gates at all; one that grants them from a fixed budget still does.

A 30-day implementation sequence

Write a one-sentence kill rule for each stage boundary, apply it to the current backlog this month, and publish how many items it removed.

Week one, count the backlog by stage and calculate the average age of an open item. Week two, write the question and the evidence bar for each stage. Week three, apply the rules retrospectively to every open item and produce the list of what would close — expect it to be uncomfortable. Week four, hold one forum, close them, and publish the number, because the credibility of the process is established by the first cull rather than by the announcement.

Track exits, not submissions

Keep a decision record for every closed item: the stage it reached, the evidence that settled it, the person who applied the rule, and the date. The value appears at around thirty entries, when the record shows whether decisions are being made on evidence or on seniority. A pattern where senior sponsors' ideas consistently survive stages that others' do not is not a governance opinion at that point; it is a documented fact, and it is far easier to change a process with that in hand than with an impression.

Review the portfolio monthly and the rules quarterly. The monthly review makes decisions; the quarterly one asks whether the bars are calibrated, using twelve-month survival of the things that passed. If everything that passed has survived, the bars are too high and the organisation is buying certainty it does not need. If most has not, they are too low. Adjust one bar at a time and record the previous value, so a change in outcomes can be attributed rather than guessed at.

Editorial conclusion

The scarce resource in innovation is not ideas and has never been ideas. It is the willingness to end things, on the record, while their sponsors are still in the room. A process that makes that ordinary produces decisions. One that avoids it produces a backlog and calls it a pipeline.