AI for Parts & BOM Management
Why part search results show size beside the relevance score, and how reading both together lets engineers rule out parts that cannot fit before opening any model.
·
⏱
7 min read
Michelle Ben-David
Michelle Ben-David is a mechanical engineer and Technion graduate. She served in an IDF elite technology and intelligence unit, where she developed multidisciplinary systems integrating mechanics, electronics, and advanced algorithms. Her engineering background spans robotics, medical devices, and automotive systems.
BOTTOM LINE
A relevance score tells you how similar a part is to your request, and a size tells you whether it can physically sit in the space you have. Showing both in one result row lets you strike candidates that cannot fit before you open a model, then spend your attention on the few that can. It is a filter that saves time on dead ends, not a guarantee of fit, so you still confirm details, tolerances, and approval status on the part you pick.
You run a search for a bracket, and the top result comes back with a relevance score of 94. It looks like a near-perfect match. Then you open it, load the model, drop it into the assembly, and find it is thirty millimeters too tall for the pocket it has to live in. The score was honest: the part really is similar to what you described. It just said nothing about whether the part fits the space you have.
That gap is why part search results are more useful when each match shows its size beside its relevance score. This post explains what the two numbers each tell you, why one cannot stand in for the other, and how reading them together lets you rule candidates out before you open a single file. It is a small change in what a result row contains, and it changes how much time an engineer spends on dead ends.
Why a Relevance Score Cannot Tell You Whether a Part Fits
A relevance score answers one question: how closely does this candidate match what I asked for? It blends the words in your query, the metadata attached to the part, and, in a geometry-aware search, how close the shape is. A high score means the candidate resembles the request. It does not mean the candidate satisfies a constraint you never typed.
Most real searches carry constraints that never make it into the query. You search for "L-bracket, two slotted holes, aluminum" because that is how you think about the part. The space it must occupy lives in a different place: a housing drawing, a neighboring component, a clearance you measured last week. The search has no way to know about it unless the result itself exposes the one property that lets you check.
Similarity and size are also close to independent. Two brackets can share the same profile, the same hole pattern, and the same material, and differ by a factor of two in every dimension. A scaled family of parts scores almost identically against a text query, because the words describing them are almost identical. The score cannot separate them, and for a mechanical engineer that is exactly the separation that matters.
If you want background on how shape-based matching works in the first place, the post on geometric search for CAD covers it. The point here is narrower: even a very good match needs a size check before it earns your time.
IN PRACTICE
The geometry search has been invaluable, helping me find standard parts instead of designing new ones, saving a huge amount of time and effort. The search system is smart and CAD-aware.
- eytan s., R&D Engineer
What Size Tells You That Similarity Does Not
Size is the cheapest fit test there is. Before tolerances, before mounting patterns, before material or finish, a candidate has to physically fit inside the envelope you have. That is a comparison of three numbers, and it can eliminate a large share of a result list in seconds.
Think about what you actually know at the moment you start a search. Usually it is a space: a pocket depth, a maximum height under a cover, a width between two walls. Those are hard limits. Everything else, such as the exact hole spacing or the preferred fastener, tends to be negotiable. A size shown next to each result lets you hold the hard limit against every candidate at once, without opening anything.
Size also catches a failure that is easy to miss when you scan thumbnails. Thumbnails are rendered to fit a card, so a small clip and a large bracket can look the same on screen. The shape reads as right, the picture reads as right, and the scale is invisible. A number beside the thumbnail restores the scale that the picture removed.
There is a cost argument too. Opening a candidate means launching a viewer or a CAD session, waiting for a model to load, and often inserting it into an assembly to test it. Done for five candidates, that is a real chunk of an afternoon. The same elimination from a result list takes a glance. This is one of the quiet reasons engineers redesign parts that already exist: checking whether an existing part fits costs more effort than it looks like it should, so people skip the check and draw something new.
Ruling Candidates Out Before Anything Is Opened
Leo is an AI assistant for mechanical engineers that connects to an organization's knowledge base, including PDM and PLM data, and its part search results now show each match's size beside its relevance score. Leo offers integrations with leading PDM and PLM platforms (SolidWorks PDM, Autodesk Vault, PTC Windchill, Siemens Teamcenter, Arena PLM, and others), and it works as an intelligence layer on top of them rather than a replacement. The size column is the piece that turns a ranked list into a screening tool.
A practical way to use it looks like this:
State the space you have, in the numbers you actually measured, before you read any result.
Scan the size column first and strike every candidate that exceeds a hard limit.
Look at the relevance score only for what remains, and open the top one or two.
The order matters. If you read score first, you anchor on the top result and then look for reasons it might work. If you read size first, you remove the impossible candidates and let the score rank the survivors. The second habit is faster and produces fewer false starts.
This approach pairs well with searching in plain language. If you describe a part the way you would to a colleague, as in finding parts in minutes with plain-language search, the result list is broad on purpose. A broad list is exactly where a size column earns its keep, because you want to narrow it by a physical limit rather than by guessing better keywords.
Reading Size and Score Together
The two numbers are most informative in combination, and a few patterns come up repeatedly.
A high score with a size that fits is the easy case. Open it, check the details, and you may be done. A high score with a size that is too large is the instructive one: the part is the right idea at the wrong scale. That is a candidate worth noting, because a smaller member of the same family may exist, and a quick follow-up search for that family is often cheaper than a new design.
A modest score with a size that fits deserves a second look. Similarity can be low for boring reasons, such as sparse metadata or an unusual naming convention in a legacy file. If the envelope fits and the shape is plausible, it can still be the part you wanted. The size gives you a reason to look that a score alone would not.
A low score with a size that is wildly off can be dismissed on both counts. Having two independent reasons makes that call fast and easy to defend later in a design review.
None of this replaces judgment about the standard part you are really after. It helps to pair the screening step with a shared parts library, which is the subject of how to stop engineers from redesigning parts that already exist. The screening decides which candidates deserve attention. The library decides which ones your team has already approved.
What a Size Value Still Cannot Confirm
A size next to a score is a filter, not a verdict, and it is worth being clear about what it leaves open.
Overall size describes an envelope. It does not describe the shape inside that envelope. A bracket can fit within the pocket's width and height and still collide with a boss on the housing wall, block a fastener's tool access, or interfere with a neighboring part. Envelope fit is necessary for a candidate to work, and it is nowhere near sufficient.
Size also says nothing about tolerance, material, finish, load rating, or approval status. A part that fits and ranks well may be a superseded revision or a part that failed a qualification. Those facts live in the part record and the release history, and a person has to read them. When the part will carry a structural or safety-relevant load, the same discipline applies as always: confirm it against your standards and your own checks before it goes into a design.
Finally, the size is only as good as the source model. If the underlying CAD file was modeled at the wrong scale or the metadata is stale, the number will faithfully repeat the error. That is a data quality problem that exists with or without a size column, and it is one more reason to treat the final step as a human check. Teams that want to reduce that risk usually start with search across their PLM data so that records are consistent before anyone relies on them.
FAQ
Rule out parts that cannot fit
See each match's size next to its relevance score.
Leo shows a size beside every part search result, so you can strike candidates that will not fit the space you have before you open a single model.
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams
