Why 99.5% Accuracy Isn't Good Enough (And What to Do About It)

Every executive in manufacturing has heard the pitch by now. AI is transformative. Your competitors are using it. You need to get on board or get left behind.

Here's what they don't tell you: knowing AI exists and actually extracting value from it are completely different problems.

Over the past two years, we've been embedded with a specialized aerospace manufacturer, working through this transition firsthand. The real gap isn't awareness—it's implementation. Organizations get stuck because there's a massive disconnect between what the models can do and what they know how to do with them.

The challenge? A model that works 95% of the time looks incredible in a demo. But in production manufacturing, even a 0.5% error rate isn't just inconvenient—it can be catastrophic. You can't risk your delivery reputation on the assumption that edge cases won't hit you.

This is where most organizations freeze. They know they should be doing something with AI. They're afraid of falling behind. But they don't have a clear path from "everyone has ChatGPT" to "AI is materially improving our operations."

At xSkel, we've walked this path enough times to have a framework. And it's simpler than you think.

Stage One: Give Everyone Access (Responsibly)

If your people don't have access to a capable AI model, start here. Not next quarter. This week.

This isn't about enterprise AI infrastructure or multi-year rollouts. It's about making sure every person in your organization—from estimators to engineers to purchasing—can ask a model questions the same way they'd ask a colleague.

Do this responsibly. Put SSO in place. Set retention policies. Make sure there are controls around what data gets shared. But don't let compliance paralysis stop you from taking this step.

Why it matters: This is foundational hygiene. The same way everyone in your organization has an email address, everyone should have access to a model. Your competitors are already doing this. If your team isn't, you're starting from behind.

You'll see immediate lift in drafting, research, and problem-solving. That's great. But it's just the entry point.

Stage Two: Expose Your Data (Teach the Model Your Business)

Here's where most organizations either commit or stay stuck.

You've given people AI access. But right now, that AI knows nothing about your operation. It doesn't know what a work order means in your ERP. It doesn't understand that in your system, an RFQ becomes a quote, becomes an order, becomes a work order that rolls up into multiple purchase orders.

It can't reference your specs, your part numbers, your supplier lead times, or your change order history.

You're asking an employee to contribute without access to company systems. That's what it feels like when you stop at Stage One.

Stage Two is about intelligently exposing your data—ERP records, PLM systems, document repositories—so the model can actually function as a member of your team. Not dumping everything into an AI and hoping for the best. Structured integration: connecting the model to your sources of truth.

Think of it like onboarding a new hire. You explain what matters: "This is how we track programs. These are the objects that run our world. Here's where to find the current revision. Here's who owns what." The model needs the same context.

Why it matters: Without this step, you're just using ChatGPT with better branding. With it, you're building an AI that understands your business and can provide real assistance on your actual processes.

Stage Three: Build Tools, Not Just Prompts

This is where the real value shows up. And this is where xSkel lives.

Once your data is exposed, you start identifying the grunt work—the repetitive, necessary, error-prone tasks that eat hours and create risk:

• Comparing purchase order revisions to catch every change
• Extracting bills of materials from engineering drawings
• Tracking long-lead components across multiple programs
• Validating contract language differences between versions

Most organizations assume the answer is "get the AI to do it."

It's not. Not yet. The models aren't reliable enough to operate independently on high-stakes work.

The answer is building software tools, augmented by AI, that handle these processes with appropriate human oversight.

Here's what changed: Five years ago, building custom software for a small piece of your workflow made no economic sense. You'd need a development team, ongoing maintenance, infrastructure—all for a single process improvement. Enterprise software vendors weren't interested in contracts under seven figures, and you'd be adapting your workflow to their platform rather than the other way around.

Today? A competent developer can prototype a functioning tool for a fairly complex workflow in about a week. That tool can leverage AI where it's strong (understanding documents, extracting data, comparing versions) and rely on traditional software where reliability matters (enforcing rules, validating logic, maintaining audit trails).

Our design partner—a specialized aerospace manufacturer building complex assemblies—was already operating at 99.5% accuracy, an excellent track record built on skilled people and rigorous processes. But in aerospace, that remaining 0.5% represents edge cases that can destroy programs. A component with a 19-52 week lead time that doesn't get ordered on schedule? That's a six-month program slip waiting to happen.

The challenge with rare but catastrophic failures is that manual review processes can't catch them consistently. They slip through precisely because they're edge cases—they don't happen often enough for people to stay vigilant.

We helped them implement a simple, consistent process: every morning, automatically check every active work order and ask "Is this part on time?" To do that, we integrated their full ERP and built a tool that runs that check across all programs, alerting project managers to potential issues before they become problems.

Why does this need AI? Traditional ERP alerts work when you can define exact rules: "flag any PO where lead time exceeds X days." But real manufacturing edge cases don't fit neat rule definitions. You need a system that can read across a work order, understand which components are long-lead based on supplier history and current availability, compare that against the program timeline, and flag the intersection of factors that creates risk—not just any single threshold breach.

That kind of contextual reasoning across multiple data sources is where AI earns its place. It's not replacing the human decision; it's doing the tedious cross-referencing that makes the "should we be worried about this part?" question answerable at scale.

It's not sophisticated AI doing creative work. It's automation enforcing a simple question that humans should ask but can't reliably remember to ask for every part on every program every day. The tool turns "someone should probably check this" into "this gets checked automatically, every morning, without fail."

That's Stage Three: using AI not to replace humans, but to make building custom automation economically viable for mid-sized manufacturers. You're building the tools your operation needs—tools that were previously out of reach.

Why it matters: This is where you actually move the needle on productivity, quality, and cycle time. You're not asking AI to run your business. You're using AI to build the supports that catch the edge cases before they cascade.

Stage Four: Rethink What's Possible

This is where the real alpha lives—and where the gap between model capability and organizational exploration is widest.

Stage Four is first-principles thinking: redesigning workflows around what's actually possible with AI, rather than just automating existing processes. With the first three stages in place, you can start asking fundamentally different questions:

What handoffs could you eliminate entirely? What steps become unnecessary when the first pass is automated and reliable? What quality checks could you instrument at the beginning of a process instead of the end?

This is where you see the real possibility and fruition of AI. Not incremental improvements to existing workflows, but entirely new ways of operating that weren't feasible before.

The capability overhang here is massive. Models can already do far more than most organizations are exploring—but you need the foundation of access, data integration, and proven tools before you can effectively redesign around those capabilities.

This stage requires the discipline of the first three to be solid. It's where durable competitive advantage gets built, and where the manufacturers who move first will create separation from those who stayed stuck at "everyone has ChatGPT."

Where xSkel Comes In

We're not a chatbot vendor. We're not selling you generic AI platform seats.

We're an exoskeleton company. We come in, understand your highest-risk failure modes, and build the small, focused tools that prevent costly mistakes from propagating through your operation.

We do this because we've seen what happens when manufacturers try to implement AI without a partner who understands both the technology and the operational reality:

• They get stuck at Stage One (everyone has ChatGPT, nothing changes)
• They skip Stage Two and wonder why the AI isn't helpful
• They try to jump to Stage Three without the groundwork and get burned by unreliable outputs

Our role is to be your advisor and implementation partner—helping you move through these stages in a way that respects the fact that even 0.5% error rates can kill a program when those errors hit your critical path.

We work with whichever foundation models and infrastructure best fit your security requirements and use case—whether that's on-premises deployment, private cloud, or API-based. The specific technology matters less than the integration: connecting to your actual systems of record and building tools that operate reliably within your constraints. We're model-agnostic because the bottleneck isn't the AI—it's the implementation.

We build tools that do the first pass on critical work, enforce consistent checks that humans can't sustain manually, cite their sources, and flag for human review when they're uncertain. We deploy within your network to meet your security requirements. And we provide embedded training so people actually use what we build.

The gap between knowing AI exists and extracting value from it isn't closing on its own. But the roadmap is clearer than it looks: access → data → tools → redesign.

We help manufacturers walk that path without the edge-case failures that threaten your brand.

‍