If you run operations or growth at a small or mid-size manufacturer, I’d bet money you’ve had this exact conversation in the last thirty days: someone floats a new product idea, a line extension, a capability that could open a new market — and within five minutes the conversation isn’t about the idea anymore. It’s about who has the bandwidth to work on it, and whose job it’s going to come out of.

That’s not a people problem. That’s a systems problem. And it’s one of the most consistent, most underdiscussed constraints I see across manufacturing — whether I’m looking at it inside our own subsidiaries or talking with other operators in the industry.

Let’s get into why this keeps happening, and what the data actually says about it.

The Math Nobody Wants To Say Out Loud

Here’s the uncomfortable truth: in most SMB manufacturers, new product development doesn’t have its own resources. It has borrowed resources — the same engineers who are expediting today’s late order, the same ops leader who’s covering for a sick shift supervisor, the same owner who’s the only one who can approve a capital request. NPD gets whatever time is left over after operations takes what it needs and operations always needs more than it has.

This isn’t unique to any one company. It’s structural. Research on SME product development consistently identifies limited resources, informal strategic planning, and less-formal processes as defining characteristics of how smaller manufacturers approach NPD — not exceptions to the rule, but the rule itself. And it shows up in the numbers: an IndustryWeek analysis of small and medium manufacturers found that more than 60% of small manufacturers lack a defined new product development and introduction process at all. Not an imperfect one. None.

Think about that for a second. Six in ten of us are trying to develop and launch new products the same way we’d handle a rush order — react, expedite, hope it works out — instead of running it as a repeatable process with its own gates, its own owners, and its own resource allocation.

Ambiguity is The Tax You Pay For Not Deciding

When there’s no defined process, there’s no defined priority. And when there’s no defined priority, every day becomes a negotiation. Should the machinist spend the afternoon on the customer PO that’s late, or the prototype fixture for the product that might be revenue in eight months? Nine times out of ten, the thing with a due date today beats the thing that could matter later. That’s not because anyone’s making a bad call — it’s because ambiguity always resolves in favor of whatever is loudest and most immediate.

This is Theory of Constraints 101, and it applies just as much to where you point your development hours as it does to where you point your machine hours. If you haven’t identified your actual constraint — the one resource or step that’s genuinely limiting your throughput — you’re not managing NPD, you’re just reacting to whoever complained most recently. And most companies never do the diagnostic work to find out where the real bottleneck is. They assume it’s engineering capacity, or capital, or the sales pipeline, when it’s frequently something much less visible: a single approval step, an undocumented specification, one person who has to review everything before it moves.

Cooper’s decades of NPD benchmarking research backs this up from a different angle. Businesses in the bottom 20% of NPD performers don’t just have lower success rates — they run failure rates roughly 3.5 times higher than the top 20%, and both numbers track directly with whether projects stay on time and on budget. That gap isn’t talent. It’s process discipline and knowing where the real constraint sits.

Having A Process Matters More Than Having The Perfect Process

I want to be careful here, because I know what a lot of operators hear when someone says “you need a formal NPD process”: bureaucracy, forms, a system built for a company ten times our size. That’s not what the data supports. What separates the best performers from the worst isn’t process complexity — it’s whether a process exists and gets used consistently.

An APQC benchmarking study found that nearly every top-performing company in product innovation has implemented some form of stage-and-gate system to move ideas from concept to launch. Not because the framework is magic, but because it forces two things SMBs chronically avoid: an actual go/kill decision before real money gets spent, and a place where resource allocation gets discussed out loud instead of assumed. Cooper’s research shows kill decisions cluster right after the first real business case review — which is exactly the point. Killing a bad idea in week three is cheap. Killing it after tooling is ordered is not.

None of this requires a twelve-gate enterprise system. It requires a documented — even a lightweight — decision point where someone with authority asks “does this still make sense, and do we still have the resources to do it right,” before the project keeps eating hours nobody budgeted for it.

Leadership Has To Do More Than Approve The Budget Line

This is the piece that gets skipped most often, and it’s the one I feel most strongly about. Executive support for NPD can’t mean “I signed off on the project” and then going silent until someone asks why it’s behind schedule. Support has to mean active championing — visibly protecting the resources, making the trade-offs explicit to the rest of the org, and absorbing the risk of failure so the team doesn’t have to.

McKinsey’s research on innovation culture found that only 11% of companies with high-fear cultures are considered leading innovators, compared to 58% of companies with low-fear cultures. The three fears that do the most damage aren’t fear of the market or fear of the technology — they’re fear of criticism, fear of uncertainty, and fear of what a failed project does to someone’s career. That’s entirely within leadership’s control to change, and it’s one of the cheapest levers available. You don’t need capital to stop punishing people for a project that didn’t pan out. You need to actually do it, publicly, more than once, before people believe you mean it.

Separate research on NPD project leadership backs this up: leaders who normalize failure as part of the process — who treat a killed project as information rather than an indictment — get better outcomes and better learning out of their teams than leaders who treat every miss as a black mark. If your org still quietly treats a canceled product launch as someone’s failure rather than the system working as designed, you’re
training your best people to stop bringing you ambitious ideas.

And Yes — The Floor and The Front Office Will Both Push Back

If you’ve tried to formalize any of this, you already know the resistance doesn’t come from one direction. On the floor, operators see NPD work as the thing that interrupts real work — the stuff that pays this week’s bills. At the management level, it often looks like turf: a formal process means someone has to give up unilateral control over how their people’s time gets spent, and a resource allocation conversation means
admitting operations doesn’t automatically win every time.

Both reactions are rational responses to how these companies have always operated. Nobody’s being difficult for the sake of it. They’re protecting what’s measured and what’s rewarded. Which is exactly why leadership championing matters — because a process change without visible, sustained executive backing gets quietly starved to death by exactly this kind of pushback, no matter how good the framework is on
paper.

Now Layer AI On Top of A System That’s Already Stretched

Every one of these constraints gets worse, not better, when you introduce AI into the picture — and right now, every SMB manufacturer is being told they need to.

Here’s the tension: AI is genuinely one of the more promising tools we’ve had in years for closing the resource gap I just described. It can compress the time it takes to spec a part, draft a quote, research a material substitution, or build a first-pass estimate — work that used to eat the exact engineering hours that NPD and operations are already fighting over. But adopting it takes the same scarce resource everyone’s already out of: time from the people who are too busy to learn a new tool because they’re too busy.

The data on this is pretty stark. In manufacturing specifically, AI usage in any business function rose from 1.8% to 13.9% between late 2023 and early 2026 — real growth, but still a small minority of the industry and the pace of new adoption is already showing signs of plateauing. Zoom out to SMEs broadly and the gap between large and small companies is even more pronounced: in the EU, 55% of large enterprises were using
AI in 2025 compared to just 17% of small enterprises — a 38-point gap the OECD calls a “multi-speed adoption pattern.” That gap isn’t about willingness. Across EU, OECD, UK, and G7 surveys, 50–71% of SMEs that considered AI and didn’t move forward point to the same root cause: a lack of the relevant skills or expertise to actually implement it. Separately, 61% of supply chain decision-makers cite poor data quality and system integration — not cost, not skepticism — as the primary barrier, and that problem disproportionately hits smaller operations without dedicated IT resources.

In other words, the same companies that lack a defined NPD process, that can’t clearly say where their bottleneck is, and that are running development on borrowed time are now being asked to also evaluate, integrate, and govern a new category of technology — with the same headcount. If you don’t already know where your constraint is, adding AI tools doesn’t remove the constraint. It just gives people one more thing to half-adopt around it.

I’d also push back gently on the idea that AI adoption is a bolt-on decision separate from everything above. It isn’t. If NPD doesn’t have a defined process, AI tools get scattered across whoever picks them up individually rather than built into the stage-gate itself — used for a one-off estimate here, a drafting shortcut there, with no consistency and no compounding value. If leadership isn’t actively championing experimentation and tolerating the false starts that come with it, AI pilots get killed at the first bad output instead of iterated on, which is exactly the fear-of-failure dynamic from a few sections ago, just wearing a new hat. And if you haven’t done the bottleneck diagnostic, you’ll almost certainly deploy AI at the wrong point in your process —automating a step that was never actually constraining you, while the real bottleneck stays fully human and fully backed up.

The upside is real. Companies that treat AI as core to how they already work — folded into the process ratherthan layered on top of it — report meaningfully better outcomes than those experimenting at the edges. But that only happens for manufacturers who’ve already done the harder, less glamorous work: a defined process, a known constraint, and leadership that’s bought in enough to absorb a few failed pilots on the way to the ones that stick.

Where A Lot of This Gets Solved: Bringing In Capacity You Don’t Have To Build Yourself

If you’ve read this far and found yourself nodding along because you recognize your own shop in every section, I’d tell you the same thing I tell operators outside our own group: you don’t have to solve all of this by simply hiring your way out of it. That’s exactly why we built Big Rocks Engineering—to help manufacturers tackle these challenges, build stronger systems, and grow without constantly adding more people.

Most SMB manufacturers can’t justify a dedicated NPD function, a full-time process engineer to run bottleneck diagnostics, or an in-house team to evaluate and integrate AI tooling — not because the need isn’t real, but because the volume doesn’t support a full-time hire and the risk of getting it wrong on your own is exactly what keeps these initiatives stuck at the idea stage. That’s precisely the gap an outside engineering partner is built to fill: capacity and expertise you can bring in for the scope of work in front of you, without permanently changing your headcount or your overhead.

Concretely, that looks like:

• Standing up a lightweight, working stage-gate process rather than leaving NPD to run on whoever has a free afternoon — so go/kill decisions happen before money’s spent, not after.
• Running the actual constraint diagnostic instead of guessing where the bottleneck is, using the same Theory of Constraints approach that separates the top-performing 20% of NPD organizations from the bottom.
• Absorbing the engineering hours operations can’t spare — reverse engineering, design work, prototyping, documentation — as Engineering-as-a-Service, so new product work stops cannibalizing the team that’s keeping today’s orders on time.
• Piloting AI tools inside a real process rather than as a scattered experiment — evaluating where automation actually removes friction from your specific bottleneck, instead of adopting whatever’s trending.

None of that requires you to have it all figured out before you start. It requires being honest about where the constraint actually is, and being willing to bring in the capacity to work on it without waiting until you can justify a full-time hire to do it. That’s a conversation worth having before the next product idea dies in a resourcing argument.

---

Sources referenced:

• IndustryWeek (2007), “Four Questions, FourAnswers for Small and Medium Manufacturers”;
• Cooper, Edgett & Kleinschmidt, APQC Benchmarking Best NPD Practices series;
• Cooper et al. (2004) as cited in “Structuring and Managing the New Product Development Process”;
• McKinsey & Company, “Fear Factor: Overcoming Human Barriers to Innovation” (2022);
• Warwick Business School researchon failure normalization and NPD project leadership;
• Digit Software, “The State ofAI Adoption in US Manufacturing Report” (2026);
• OECD, “AI Adoption by Small and Medium-Sized Enterprises” (2025);
• Omago, “SME AI Adoption in 2026: What the Data Actually Shows”;
• ABI Research (2025) via Capsule CRM small business AI adoption statistics.

---