Nine Forecasts and No Evidence. Why product innovation actually fails.
The usual explanation is that too many people stand between a concept and a consumer. The data says something more uncomfortable: the number of people is not the problem. What they are being asked to do is.
Walk a new product through a large consumer packaged goods company and count the judgments it has to survive. A brand manager. A brand director. An innovation lead. Finance. Sales. A channel manager. A distributor. Somewhere in there, two or three rounds of commissioned research. Then, at the far end of all of it, a person in a store who has never heard of any of them and owes the project nothing.
The obvious conclusion is that the layers are the problem. Strip them out, get to the consumer faster, and the hit rate improves. It is a clean argument, it flatters everyone who has ever been slowed down by a gate review, and I believed a version of it myself.
It does not survive contact with the evidence. But the reason it fails is more useful than the argument was.
Almost every conversation about innovation failure opens with a statistic: 80, 90, sometimes 95 percent of new products fail. It gets attributed to Clayton Christensen, usually alongside a claim that 30,000 consumer products launch each year.
Christensen denied saying it. George Castellion and Stephen Markham asked him directly; he told them the rate was lower and observed that the field had never properly validated the figure. Their review found nineteen peer-reviewed studies between 1945 and 2004 putting market failure for launched products in the range of 30 to 49 percent. In consumer packaged goods the number runs worse than the cross-industry average, but nowhere near the folklore.
This matters more than a footnote. If you believe almost nothing works, you either stop trying or stop thinking, and both are expensive. The real rate is bad enough to take seriously and good enough to be worth getting right.
Here is the fact that reorganized how I think about this.
NielsenIQ finds that the split between new items whose sales grow and new items whose sales decline shows up as early as four weeks after launch. The approval chain that put the product there took months, and often more than a year, to reach that moment.
Four weeks. Whatever the gates were arguing about, the market resolves it almost immediately once the product is in front of someone with a basket and a budget. Everything expensive, slow and contested happens before that moment, in a period where nobody has the answer and everybody has an opinion.
And the opinions are not good. NielsenIQ ran 91 concepts for a large manufacturer that had been using an internal concept-pass system and found 62 percent of them had been sorted into the wrong category. The gates were not merely slow. They were wrong about which ideas were which, at a rate close to a coin toss. Worth naming the obvious: NielsenIQ sells the alternative system, so treat the precise figure as directional rather than settled.
Approval is also not the same thing as demand. Catalina studied the top-selling new food and beverage launches in the United States — the winners, not the strugglers — and found that of the shoppers who tried one in its first six months, just 11 percent were still buying it a year later. For the average new product in that set, 0.7 percent of shoppers accounted for 80 percent of volume. One shopper in 143 was deciding the fate of products that had cleared every internal review.
So: fewer layers, faster contact, better products. Except the evidence goes the other way, and it goes there four separate times.
The companies that removed the distance, then stopped
2018–2025Large food companies built exactly the machine the argument recommends. General Mills stood up G-Works in 2019 to develop and launch its own products outside the normal chain. Kraft Heinz launched Springboard. Mondelēz built SnackFutures and then CoLab. These were purpose-built distance-compression units, funded and staffed by people who understood the problem intimately.
In March 2025 General Mills scrapped G-Works and paused investment through 301 INC for the foreseeable future. Mondelēz had already wound down CoLab by 2024 and redirected SnackFutures to straightforward minority investing. Trade coverage described the retreat as an industry-wide pattern rather than a company-specific one.
Abandonment is not proof that something failed. Budgets get cut for reasons unrelated to merit. But a lever that reliably raised the hit rate would not have been among the first things cut.
Second, the denominator. If distance were the variable, the businesses with almost none of it should win. They do not. Research cited by James Richardson puts roughly 80 percent of premium food and beverage brands as never reaching a million dollars in retail sales. Coca-Cola's own venturing unit reported that around 3 percent of beverage brands reach ten million. Founders sit inches from their customers and fail at rates that would end careers inside a large company.
Third, structure. Robert Cooper's research puts the success rate for companies running ad hoc development at 24 percent, with structured, gated processes performing substantially better. More gates, better outcomes. That is the opposite of the prediction.
Fourth, and most damaging to the case for speed:
Accelerated development carries hidden costs, including "undertaking less innovative projects."
Robert G. Cooper, on compressing the idea-to-launch cycle
Compress the loop and you do not simply get the same ideas sooner. You get systematically smaller ones, because the ideas that survive a shortened process are the ones that needed the least explaining. Cooper also notes that the research on acceleration's benefits remains inconclusive — and in a separate study of 103 projects, that solid up-front work produced projects that were both more profitable and faster to market. Front-loading did not cost time. It bought it.
Put the four together and the picture inverts. Removing layers does not help. Adding structure does. Speed helps only sometimes, and it quietly narrows what you are willing to attempt.
But the original instinct was pointing at something real. The people approving the product are not the consumer, and they are not thinking like one. That part holds. The error is in what follows from it.
The problem is not that nine people stand between a concept and a consumer. It is that all nine are being asked to predict a consumer rather than observe one. Every gate is a forecast. Stack nine forecasts in sequence and you have not built rigor, you have built nine consecutive opportunities to be confidently wrong, with no contact with the ground at any point. The chain is not too long. It is unanchored.
That explains what distance cannot. It explains why founders with no layers still fail: they reach the market fast but cannot afford to learn from it twice, so their single contact with reality has to be right first time. It explains why structure beats its absence: gates ruling on evidence are genuinely useful, and gates ruling on opinion are theatre. And it explains why the compression units were shut down. They removed judgments without grounding the ones that remained, which changes how quickly you arrive at an unsupported decision but not whether it is unsupported.
Cooper's 692-project study lands the last piece. Speed to market improves outcomes, but market uncertainty moderates the effect — where uncertainty is low, speed barely matters. Read that as a statement about information rather than time. Going faster helps when you do not know something. When you already know, it buys you nothing at all.
- Your gates are ruling on opinion — the reviews turn on seniority, articulacy or who framed the deck. Where a gate already turns on live market evidence, the process is working and the fight is elsewhere.
- You can get real contact cheaply — a regional test, a farmers market, a single retailer, a direct channel. Where the minimum viable test is a national rollout, as it is in some regulated or heavy-manufacturing categories, this is not available and pretending otherwise wastes a year.
- You can afford to be wrong more than once — grounding gates means running tests that sometimes kill your favorite idea. A business with one shot needs conviction more than it needs evidence, which is an uncomfortable thing to say and true anyway.
- The uncertainty is genuine — you do not know whether people want this. Where the answer is already known and the delay is political, faster contact with consumers will not fix a problem that lives in the org chart.
Take the next project through your gate process and, at each gate, write down whether the decision rests on evidence from a real consumer or on somebody's forecast.
Two columns, one line per gate. Do not change the process, do not remove anyone, and do not announce it as an initiative. You are measuring one thing: how far into the sequence a product travels before anyone in the room is holding an observation rather than a projection. In most chains the answer is that it never happens until launch, and seeing that written down is more persuasive than any argument about layers.
- What it costs
- An hour of your own attention per gate, and the discomfort of showing the list to people whose judgment it puts in the forecast column, including your own.
- How you'll know
- By the end of one project cycle you can name the earliest gate that could have been grounded with real market contact for less than the cost of the meeting that replaced it. If there is no such gate, your process is already evidence-led and the problem is somewhere else.
The people in that chain are not the obstacle. Most of them are good at their jobs and being asked to do something nobody is good at, which is to know in advance what a stranger will do in a shop. The market will tell you in four weeks. The only real question is how much you spend guessing before you let it.
Related from Sound Decisions: Distribution Doesn't Create Demand; It Captures It · Hope Is Not a Strategy · Purely Elizabeth Took Eight Years to Raise $3 Million
One conversation. The decision your process is about to make on no evidence.
Not a process redesign. Finding the one gate worth grounding first, and what it would take to ground it.
Request a conversation Or read the frameworks →