In PART 1 (A), Chetan argued that every incoming RFQ ultimately requires three kinds of confidence:
The confidence to build.
The confidence to quote.
The confidence to commit.
Every manufacturing leader recognises these decisions immediately.
As engineers, however, we naturally ask a different question.
Why are these seemingly simple decisions so difficult for software to answer?
After working on this problem for over two years, I've come to believe the answer has very little to do with AI, model size or computing power. It comes down to three technical challenges that every manufacturing software stack faces.
Problem 1: Engineering Identity
The first challenge sounds deceptively simple.
How does a system know that two different part numbers represent the same engineering problem?
To an engineer, similarity is often obvious. To enterprise software, it doesn't exist.
ERP, PLM, QMS - every system faithfully records what happened. None understand that today's RFQ may represent almost the same engineering problem as something manufactured three years ago under a completely different part number. That's because part numbers carry no engineering meaning.
The question: "Have we made something like this before?" isn't a part number question. It is an engineering identity question.
Understanding that identity requires reading geometry — not as an image or a CAD file — but as engineering intent. That means understanding not only the part's geometric form, but also the engineering features it embodies, the manufacturing processes it implies, and ultimately the functional purpose it was designed to serve.
Many systems can identify similar parts. But grouping is not enough. Engineering decisions require understanding why two parts belong to the same engineering family—not simply that they appear similar. Identity must be deterministic before it can become useful.
At ShapeSense, engineering identity is established across four interconnected dimensions:
Geometric – How similar is the form, topology and overall shape?
Engineering – Do the critical features, tolerances and design intent align?
Functional – Were they designed to solve the same engineering problem, even if they look different?
Manufacturing – Would these parts be produced using similar processes, tooling and capabilities?
Together, these dimensions create a deterministic Engineering Fingerprint for every component. Similarity is no longer based on geometry alone, but on engineering meaning.
Problem 2: Enterprise Relationships
Knowing what a part is solves only the first problem. The next challenge is understanding everything the enterprise already knows about it. Historical cost is one source of evidence. The objective isn't to reproduce yesterday's price—it's to understand why similar parts cost what they did, and whether that evidence supports today's commercial decision.
Unfortunately, none of this lives in the same system. It is scattered across ERP, PLM, MES, QMS, supplier databases and spreadsheets. Each system uses its own identifiers. Each tells only part of the story.
The Engineering Fingerprint becomes the anchor. The Geometric Family Index becomes the organising structure. Together they connect decades of engineering, manufacturing and commercial history into a single body of evidence.
The Geometric Family Index organises related Engineering Fingerprints regardless of programme, customer or part number. This is far more than geometric clustering.
Clustering answers:
Which parts look alike?
Enterprise Relationships answer a far more important question:
What has the organisation learned from every part in this family?
Put differently, Enterprise Relationships transform isolated historical records into institutional engineering memory. Every successful production run, supplier qualification, quality issue, commercial outcome and engineering change becomes knowledge that can inform the next decision instead of remaining buried inside disconnected enterprise systems.
Without these relationships, geometry remains isolated. With them, Engineering Identity becomes enterprise intelligence. Building this index is neither glamorous nor instantaneous. But its value compounds.
Every drawing indexed improves the next query.
Every production run enriches future cost models.
Every supplier outcome strengthens future recommendations.
Every RFQ makes the next RFQ smarter.
The result is more than connected data. It is institutional engineering memory—knowledge accumulated over years of products, suppliers, manufacturing runs and commercial outcomes, available at the moment a new decision needs to be made.
This is what engineering intelligence that compounds means in practice.
Problem 3: Engineering Evidence
Even after engineering identity is established and enterprise relationships are connected, one challenge remains.
Can the answer be trusted enough to make a commitment?
Manufacturing decisions cannot rely on probabilities. A supplier either has demonstrated capability—or they haven't. A process either meets the required tolerance—or it doesn't. A delivery commitment is either supported by evidence—or it becomes a risk.
Manufacturing requires decision confidence, not model confidence. Decision confidence comes from traceable engineering evidence—not statistical inference.
Which suppliers manufactured this geometric family?
To what tolerances?
Using which processes?
With what quality outcomes?
When?
If the evidence exists, the system should present it. If it doesn't, the system should explicitly say so. There is no useful middle ground. This is the deterministic requirement.
Why Existing Software Doesn't Solve This
People often ask why ERP, PLM, CAD systems—or even modern AI assistants—cannot answer these questions already. The answer lies in what each system was designed to do. Every one of these systems performs its intended role exceptionally well. What none of them were designed to do is combine:
Engineering Identity
Enterprise Relationships
Engineering Evidence
into a single decision framework.
Bringing It Together
The three confidence decisions Chetan introduced in Part 1:
Can we build it?
Should we quote it?
Can we commit?
—appear to be independent.
Technically, they are not. All three depend on the same foundation. A system that understands engineering identity. Builds enterprise relationships. Returns engineering evidence. Only then can manufacturers make decisions with confidence.
When an RFQ arrives, engineers shouldn't have to search drawings, call colleagues, compare spreadsheets or rely on memory.
They should be able to ask one geometric question—and receive decades of engineering, manufacturing and commercial learning as traceable evidence.
Three decisions. One geometric query. Engineering intelligence that compounds over time.
That is what building decision confidence through engineering intelligence really means.
In PART 2 (A), we'll move from suppliers responding to incoming RFQs to OEMs issuing RFQs to suppliers—and explore how the same intelligence foundation transforms sourcing decisions.