When people see what we're building, a common reaction is: "Isn't this just geometric search? Doesn't my CAD and PLM already group similar parts into clusters?"
Fair question. And the honest answer is — yes, increasingly, they do.
Finding parts that look alike is no longer a hard problem. Several CAD and PLM platforms now offer some form of shape-based similarity search. Give them a model, get back a list of geometrically similar parts, organized neatly into clusters. That capability is real, and it's becoming standard.
I think that's a good thing. And I also think it's where most of the conversation stops — as if finding similar shapes was the destination, rather than the first step.
The part number was never the problem on its own
Every enterprise system in manufacturing — ERP, PLM, procurement, quality — organises itself around the part number. It's the identifier for a purchase order, the primary key of a PLM record, the thing a supplier quotes against.
The part number is good at that job. It's an administrative identity, and administration needs identities.
But a part number carries no geometric meaning. Two parts with completely different numbers can be the same shape, designed years apart by people who never knew about each other. Two revisions of the same part number can drift apart geometrically while keeping the same identity. None of that is visible if the part number is the unit you're searching on.
Geometric clustering fixes this — it groups parts by what they actually are, not by what they're called. That's genuinely useful. It's also, increasingly, a feature rather than a product.
What a cluster tells you — and what it doesn't
A cluster tells you that a set of parts share geometric character. Similar form, similar features, similar constraints.
What it doesn't tell you is what to do about that.
Once you know that forty-seven parts across three product lines are geometrically related, the actual questions start: which of these can be consumed from existing inventory instead of reordered? Which suppliers have actually produced something in this family before, and how did that go? Does a quality issue on one of these parts apply to the others? Which of these designs represents the ‘right’ version to standardise around, given everything we know about cost, supplier capability, and manufacturing history?
A cluster is a set of parts that look alike. None of those questions get answered by looking alike. They get answered by connecting the cluster to inventory, to supplier history, to quality records, to manufacturing knowledge — the things that live in completely different systems, recorded against completely different identifiers, accumulated over years.
That connection is the part that's still mostly unsolved. And it's the part that actually changes a decision.
Search versus Intelligence
I'd put it this way: search tells you what's similar. Intelligence tells you what that similarity means — for this inventory position, this supplier base, this quality history, this program.
Search is converging toward commodity. Every serious CAD and PLM vendor either has it or is building it, and that's fine — it should be table stakes.
Understanding what similarity means in the context of inventory, suppliers, quality, and manufacturing decisions is the harder problem — and it's the one we've been working on. Not ‘here are forty-seven similar parts’ — but ‘here's what's actually reusable right now, here's the supplier who's already qualified for this geometry, here's the quality history that should make you cautious about this one specific variant.’
That's a different kind of system than a search index. It has to stay current as inventory moves, as suppliers are added and dropped, as quality findings come in. It has to connect across systems that were never designed to talk to each other. And it has to do this continuously, not just when someone runs a query.
Two ways manufacturing intelligence works
Most systems wait for someone to ask. We think it needs to work two ways.
On-demand intelligence answers a specific question when someone needs it. Can this inventory satisfy active demand? Which supplier has manufactured something similar to this before? Should this design be reused, or does it warrant something new?
Always-on intelligence doesn't wait to be asked. It continuously watches for newly discovered reuse opportunities, emerging duplication across programs, inventory that's quietly become recoverable, supplier concentration that's becoming a risk, and designs that are converging toward a standard nobody declared.
One helps someone make a decision they're already trying to make. The other makes sure they find out about the decisions they didn't know they needed to make.
Shape gets you the cluster. Sense is what happens next.
This is why the name is ShapeSense, and why both halves matter.
Shape is what lets us read geometry as ground truth and find the clusters — the parts that genuinely belong together, regardless of what they're called or which system they live in. That capability is foundational.
Sense is everything that happens after the cluster exists — connecting it to inventory, suppliers, quality history, and manufacturing knowledge, and turning those relationships into actionable decisions. On-demand, when someone asks. Always-on, when something changes that someone needs to know about.
Geometric clustering is becoming standard. We think that's a good thing. The real opportunity begins once those clusters are connected to the rest of the enterprise. That's where manufacturers discover reusable inventory, qualified suppliers, quality insights, standardization opportunities, and hidden value that would otherwise remain buried across disconnected systems. Shape reveals relationships. Sense turns relationships into decisions. That's where real value lies.