Ask three people in your company how the CAD data is doing and you get three feelings, not a number. The engineer thinks it is "in reasonable shape", work preparation "can usually manage with it" and purchasing rattles off three recent misses from memory. Everyone is right, and nobody knows where things actually stand.
Without measurement, data quality remains an opinion, and you cannot steer on opinions. The good news: measuring the quality of your CAD data does not have to be a project of weeks. With a sample of a few projects, an export to Excel and three simple counts, you have a baseline in one afternoon. Everything you need is in Inventor plus an export facility like the one in Thundercad; the rest is counting.
In this article we build that baseline step by step: pick a sample, export the properties, count, turn it into a score and convert the three biggest gaps into targeted actions. Why this is worth the effort is covered in more depth in Clean, reliable data into production: every empty or wrong field surfaces sooner or later as an error in a quotation, purchase order or work order.
Why a feeling is not steering information
Discussions about data quality without numbers always run the same way: the most recent incident wins. If an order went wrong last month because of a wrong material field, material is suddenly top priority, whatever else is going on. A number takes the noise of the day out of it and makes three things possible: prioritising by size instead of by emotion, justifying hours of cleanup work to your manager, and showing after six months that the effort actually delivered something.
Just as important: a measurement makes the problem small. "Our data is a mess" is paralysingly big. "The material field is empty for a third of our manufactured parts" is a job you can schedule. That difference in wording decides whether anything happens or not.
Step 1: pick a small, honest sample
Do not measure your entire archive; that is exactly the scale that makes the project strand. Three projects are enough: a recently delivered machine, a running project and an older project that still regularly serves as a copying base. Together a few hundred parts: big enough to see patterns, small enough that you can still explain every result.
Resist the urge to pick your tidiest project. The measurement is not there to score well but to know where you stand, and it is precisely the older copying project that shows which data quality keeps spreading into new projects through reuse.
Step 2: export the properties and count three things
For each project, open the main assembly and put the BOM with all relevant columns into Excel: part number, description, material and the custom fields you have agreed on. With Export BOM that is one action per assembly: the BOM lands in your own Excel template, with the columns you have configured. Then, per column, you count three things:
- Empty fields: count the empty cells and convert to the percentage filled per field. This is the fastest and hardest indicator.
- Spelling variants: make a pivot of unique values per field. Where you expect five and find thirty, there is your gap: "RVS 304", "rvs304" and "1.4301" mean the same thing and count as three values.
- Duplicates: sort by description and look for lines describing the same part under a different part number. Count them per hundred parts.
A baseline stands or falls with quick exporting: one BOM per project with all the columns you want to count. Thundercad puts that export into your own template in one action with Export BOM.
Try 30 days freeStep 3: turn it into a simple score
Work out two numbers per field: the percentage filled and the percentage that follows the agreed spelling. Duplicates are counted separately, per hundred parts. For a baseline you need nothing more. An example of what that can look like (the numbers are made up):
| Field | Filled | As agreed | Biggest gap |
|---|---|---|---|
| Description | 98% | 71% | spelling differs per engineer |
| Material | 66% | 54% | empty for purchased parts, three spellings side by side |
| Part number | 91% | 88% | duplicates among sheet metal parts |
| Project code | 43% | 43% | hardly ever filled in |
If you do want a single figure for the department wall, take the average of the "as agreed" column. But steer on the rows beneath it: that is where you see which field hurts, and for whom. A score is not a school grade but a route map.
From measurement to three targeted actions
From the table, pick the three gaps causing the most pain further down the process, not the three easiest ones. Give each gap one owner and one concrete measure. The patterns usually fall into three kinds. A field that has been agreed on but stays half empty asks for easier filling: with a data card per document type in the iProperty Panel, the engineer sees at release exactly which fields are still empty. Varying spellings are not fixed with a memo, but with pick lists instead of free text. And a field that hardly ever gets filled deserves the question whether it should exist at all; how to make that trade-off is covered in Custom iProperties: when they help and when they hurt.
Close with an agreement on repetition: the same three projects plus the newest project, the same counts, one hour every quarter. From the second measurement onwards you have a trend, and that was the point: not knowing how bad it is, but seeing it get better.
Frequently asked questions
How big does the sample need to be?
Three projects with a few hundred parts between them is plenty for a first measurement; the patterns you do not see in those, you will not see in a thousand parts either. What matters more is picking the sample the same way at every repetition, so the measurements stay comparable. You can always widen the net if the first round gives you a reason to.
What score is good enough?
There is no general norm, and you do not need one: you compare against your own previous measurement, not against another company. Set the bar highest for fields that production and purchasing steer on directly, such as material, description and part number. An informational field may lag behind without anyone losing sleep over it.
What if the measurement mainly shows we have too many fields?
Then that is a perfectly good outcome. A field nobody fills only costs attention and pollutes every export; cutting it makes the remaining fields more important and the next measurement better. If you then want to make filling and exporting structurally easier, you can try Thundercad free for 30 days.