Formic’s 99.98% Rule: Why “Good Enough” Kills Hard Tech Adoption

In a recent episode of Category Visionaries, Saman Farid, CEO and Founder of Formic, explained why his robotics company rejects algorithms that academics consider breakthrough achievements. The reason comes down to a calculation that most tech founders never run.

An academic researcher approaches Saman with a new algorithm achieving 98% accuracy—a significant improvement over existing solutions. In their world, this is publication-worthy work.

Instead, Saman does the math.

The Math That Changes Everything

“If you’re doing, let’s say, 10,000 operations per day and you’re only 98% accurate, that means you’re dropping hundreds of parts per day, making hundreds of mistakes per day, which is just completely unacceptable in a manufacturing environment,” Saman explains.

Let’s break this down. At 10,000 operations daily with 98% accuracy, you generate 200 failures per day. That’s 1,000 failures per week, 4,000 per month. Many facilities run two shifts—double those numbers.

At scale, 98% accuracy doesn’t mean occasional issues. It means systematic disaster.

When Saman shares this math with academics, “they’re typically shocked,” he says. “They’re like, that’s unrealistic. That’s impossible. There’s all these different issues and challenges.”

But unrealistic or not, these are the actual standards production environments require.

Why Software Thinking Fails in Hardware

The performance gap reveals a deeper problem: most people building technology don’t understand production environments. “A lot of these robotics companies that you’re talking about, they come out of academia or they come out of large tech companies or even coming out of software companies,” Saman notes.

Software companies assume iteration. “In software, there’s this mentality that you can kind of iterate really quickly,” Saman explains. Ship at 80% complete, get feedback, fix bugs in the next release. This works because failure costs are low—users restart crashed apps, you push updates for bugs.

Manufacturing doesn’t work this way. When your robot fails, production stops. Materials get wasted. Orders get delayed. Revenue disappears.

“With robotics, you don’t really get the chance to iterate quickly once you deploy something, if it doesn’t work, then you’ve just lost that customer,” Saman says. There’s no “push an update and try again.” You get one shot.

The Academic Trap

Academia brings different but equally problematic assumptions. “In academia, there’s this mentality that the coolest technology wins,” Saman explains. Research labs optimize for novelty. The goal is advancing the state of the art, achieving results nobody has achieved before.

When cutting-edge approaches achieve 98% accuracy, that’s genuinely impressive from a research perspective. It might represent years of work and significant breakthroughs.

But manufacturing doesn’t care about technical impressiveness. It cares about whether production keeps running.

“I think in the real world, what we found is neither of those are true,” Saman says, referring to both software iteration and academic novelty. Production demands reliability over innovation, proven performance over cutting-edge capabilities.

The Real Performance Standard

So what does manufacturing actually require? Formic targets 99.98% accuracy or higher—not aspirational, but minimum threshold.

At 10,000 operations daily with 99.98% accuracy, you generate two failures per day. That’s 10 per week, 40 per month. Still not zero, but approaching tolerance levels where facilities can work around issues.

This forced fundamental choices. Most robotics startups chase the newest technology. Formic does the opposite.

“Let’s not necessarily go and find the newest technology out there for every kind of robot in the world. Let’s go find the things that are the most reliable and choose the path that leads to the highest robustness for our customers,” Saman explains.

This means proven sensors over cutting-edge ones. Well-tested components over latest releases. Boring beats sexy at 99.98% accuracy.

Why Ten Minutes Destroys Deals

The requirement becomes even more extreme considering deployment context. “If your robot is down even 2% of the time, what that means is you’re dropping hundreds of boxes a day or hundreds of parts a day,” Saman notes. Two percent downtime sounds minor—in practice, it’s 38 minutes of failure per day across two shifts.

That’s 38 minutes of paying idle workers, delayed shipments, and explaining late orders to customers.

The threshold is even lower. “Even if it’s down for ten minutes, you’ve just lost that customer,” Saman says.

Ten minutes. Not hours or days. Ten minutes of unexpected downtime and the customer is gone forever. They’ll never try robotics again. They’ll tell other manufacturers. They’ll contribute to industry-wide skepticism about automation.

The Deployment Reality

This unforgiving standard explains why Formic’s deployment strategy looks nothing like typical robotics companies. There’s no “deploy and iterate.” No “minimum viable product.” No gradual improvement based on field data.

Formic frontloads everything into pre-deployment. They invest heavily in reliability engineering, test extensively, vet projects carefully, and build monitoring systems that predict failures before they happen.

By the time a Formic robot reaches a factory floor, it must already perform at 99.98%+ reliability. Not eventually. Not after iterations. Immediately, from day one.

This requires capabilities most robotics companies lack: extensive testing infrastructure, predictive maintenance systems, nationwide service networks, and the discipline to say no to deployments that aren’t ready. It’s slower and more expensive than “ship fast and iterate.” But it’s the only approach that works when ten minutes of downtime kills your business.

The Lesson for Hard Tech Founders

Formic’s experience reveals a crucial principle: understand the actual performance requirements of your target industry, not the performance standards of your source industry.

If you’re from software, your instincts about iteration will actively hurt you in hardware. If you’re from academia, your focus on novel technology will lead you to build things that impress researchers but fail customers.

The gap between 98% and 99.98% sounds small—just two percentage points. But in production environments, those two points represent the difference between technology that works and technology that destroys your business.

You can’t compromise across this gap. You can’t explain that 98% is pretty good. You can’t ask customers to be patient while you iterate. You can’t convince them your technology is exciting so they should accept lower reliability.

You either hit the standard the environment requires, or you fail. There is no middle ground.

As Saman discovered: production environments require perfection, not innovation.