70 to 90 percent. That's the failure rate for new product launches in CPG and FMCG. One in ten makes it. The rest disappear from shelves, taking months of work, real budget, and a fair amount of internal credibility with them.
What's strange is that most of those failures weren't actually surprises. The signals existed. Someone, somewhere, probably had a nagging feeling. The problem wasn't a lack of information. It was that the information arrived too late, or in the wrong form, or got buried in a process that wasn't built for speed.
Here's what the personalization revolution in consumer marketing figured out years ago that innovation research still hasn't fully absorbed: understanding what consumers want is no longer the hard part. Acting on it before the window closes is.
McKinsey's research on personalization produced some numbers that became ubiquitous in marketing circles. 71% of consumers expect personalized interactions. 76% report frustration when what they get feels generic.1 Companies that excelled at personalization generated 40% more revenue from those activities than their slower-moving peers.2
But the stat isn't really the point. The point is what those companies did differently.
They didn't just collect more data. They built systems to act on consumer signals continuously, in near real-time, rather than in scheduled annual cycles. McKinsey found that doing this well could reduce customer acquisition costs by up to 50% and lift revenue by 5 to 15%, with marketing ROI improvements in the 10 to 30% range.3 Not because the insights were radically different, but because the speed of application changed everything.
Innovation research is still largely running on the old model. A concept gets flagged as worth testing. It enters a study. Six to ten weeks later, results come back. And somewhere in that gap, the retail window shifted, the buyer's priorities moved, or a competitor got there first. The research was good. The timing wasn't.
There's a structural paradox at the center of this work that doesn't get talked about enough.
The pressure to move fast is real. Retail cycles are compressing. Buyers at Walmart, CVS, and Whole Foods want a pipeline of validated concepts, not promising hunches. Shelf space is finite and contested, and the brands that show up without evidence don't stay long.
But the pressure to be right is equally real, and the consequences of being wrong are severe. A failed launch isn't just a P&L hit. Retailers delist. Category relationships that took years to build get damaged. The internal fallout is painful. And the next concept that comes through the Stage-Gate carries the weight of the last failure.
Traditional research was designed for a world where slowing down was an acceptable trade-off for certainty. That world is gone.
A single traditional concept study runs $20,000 to $80,000. At that cost, most teams can realistically test two or three ideas per quarter, maybe. Which means the vast majority of early-stage concepts get pressure-tested by internal conviction and whoever argued most persuasively in the last meeting. Not exactly a rigorous filter.
McKinsey's personalization research flagged something that cuts directly to the core of how innovation testing works: the gap between what consumers say they'll do and what they actually do.
This is not a small gap. Consumers are genuinely unreliable narrators of their own preferences, not because they're dishonest but because survey conditions don't replicate real decision-making. When you ask someone whether they'd buy a new product, they answer based on how they want to present themselves, what sounds reasonable, and what the framing of the question seems to invite. The research reflects those answers. Not the shelf behavior.
A concept can score an 80 in a traditional survey and fail at retail. It happens constantly. The survey measured intent. Retail measures action. Those are not the same thing.
AI-driven behavioral simulation approaches this differently. Instead of asking what consumers would do, it models what they're likely to actually do, drawing on behavioral variables, category-level dynamics, and real purchase drivers. The output is still probabilistic. But it's modeling the right thing.
The framing shift that changes how you think about this: it's not concept testing. It's decision rehearsal.
Before a product reaches a shelf, there are dozens of consumer decisions layered on top of each other. What competitive set will it appear next to? What price will it carry? Which claim gets the most traction with which segment? Who is likely to trial it versus who might repeat? These aren't abstract questions. They're the actual variables that determine whether the launch works.
Each of those questions used to require a separate study. Together, they represented months of sequential research and costs that made most early-stage pressure-testing financially unjustifiable.
Zibble compresses that into days. The platform's AI Signal Groups and behavioral modeling simulate real consumer decision environments, which means teams can test 10 or more concepts for the cost of one traditional study. Concepts that previously advanced on gut feel can be validated or killed before they absorb development resources. The ones that reach the gate have actually earned their place there.
AI-powered consumer intelligence tools have shown a 25% lift in marketing ROI and roughly 20% increase in sales conversion in downstream commercial applications.4 That's not a soft benefit. Deloitte research found that consumers receiving experiences tailored to actual behavioral drivers spend approximately 50% more than those who receive generic outreach.5
The translation to innovation is direct. Concepts validated against behavioral signals outperform concepts validated against stated preferences, because the research that shaped them was modeling how consumers actually make choices.
And the cost comparison is stark. A failed launch, factoring in delisting, lost shelf position, and brand damage, is orders of magnitude more expensive than better upstream research. The hesitation usually isn't financial. It's organizational. Research teams get evaluated on process adherence, not launch outcomes. That misalignment is where the real cost lives.
Stage-Gate was designed as a risk-reduction framework. In practice it often functions as a resource allocation mechanism, where the concepts that advance are the ones with the most internal support, not necessarily the ones with the strongest consumer signal.
That's not a criticism of the framework. It's a criticism of what happens when the cost of real consumer validation is too high to apply early.
When testing is fast and affordable at every stage, the dynamic changes. Teams can apply genuine consumer rigor from early ideation rather than saving it for later, when changing direction is expensive. The innovation department stops being a cost center that occasionally produces a winner. It becomes a system that generates predictable results because the filter is doing its job throughout.
The internal champion who walks into a retailer or a CEO review isn't just presenting research. They're presenting behavioral evidence. That's a different conversation.
McKinsey's conclusion in its personalization work was essentially a compounding argument. The brands that built continuous consumer intelligence systems would pull ahead over time. More iterations, better signal, faster learning. The gap between the organizations that made the shift and those that didn't would widen every cycle.
The same thing is happening in innovation research right now. The insights professionals who define the next decade of CPG and FMCG won't necessarily have the biggest budgets. They'll have operating models that generate consumer intelligence continuously and act on it quickly, turning Stage-Gate into what it was always supposed to be.
The 70 to 90% failure rate is not a fixed law of consumer markets. It's a product of research methods built for a slower world. Better methods exist. The teams that adopt them first aren't just improving their process. They're building an advantage that gets harder to close every year.
The companies that succeeded with personalization didn't necessarily have more customer data than everyone else. They simply became much better at using it. They built systems that helped them learn continuously, test ideas quickly, and make better decisions before their competitors did.
The same shift is starting to happen in product innovation.
Imagine being able to pressure-test a new product concept, package design, pricing strategy, retailer presentation, or marketing campaign against AI personas built from your own customer segments. Instead of waiting weeks for a research study, your team could explore dozens of ideas in a matter of days, identify the strongest opportunities, and move into qualitative or quantitative research with much greater confidence.
This is exactly why we built Zibble.ai.
Zibble is an AI-powered consumer intelligence platform designed for innovation teams in CPG, FMCG, and OTC Pharma. It helps teams create realistic customer personas from existing segmentation or research, explore how different audiences are likely to respond to new ideas, and uncover insights before committing significant time and research budgets.
We don't see AI replacing traditional research. In fact, we think the opposite. The biggest value comes from combining AI with qualitative and quantitative methods. AI helps you ask better questions, eliminate weaker concepts earlier, and spend your research budget on the ideas that deserve deeper validation.
Personalization has already changed how marketing teams work. We believe it will have an equally profound impact on innovation. The companies that build continuous consumer intelligence into their product development process will learn faster, make better decisions, and bring stronger products to market. That's where we think the next competitive advantage will come from.
Zibble.ai is an AI-driven consumer research platform built for innovation teams in CPG, FMCG, and OTC Pharma. AI Signal Groups and behavioral consumer modeling enable concept testing, pricing validation, and packaging research in days, not weeks.
