Skip to content

AI and engineering data

Why PDFs and Spreadsheets Are a Poor Long-Term Memory for AI

They are good documents. They are a poor place to keep the relationships between objects.

4 min read

A PDF is a good way to deliver a datasheet. A spreadsheet is a good way to work through a cost calculation. Neither is a criticism worth making, and this note is not making one.

The question is different: what happens when those two formats become the place where a company's engineering knowledge lives permanently?

The position. PDFs and spreadsheets preserve content and lose structure. They are excellent containers and poor memory, because the relationships between the things they describe are not inside them. For a human reader that is a minor inconvenience. For any automated reasoning it is the whole problem.

What each format is good at

FormatGenuinely good atQuietly bad at
PDFFixed presentation, fidelity, signatures, distributionBeing queried, being joined to anything else
SpreadsheetCalculation, ad hoc structure, fast iterationIdentity, versioning, concurrent truth
Shared folderCheap, universal, no training neededEncoding relationships a machine can follow

Nothing in the middle column is false. That is exactly why these formats spread. The right-hand column is what accumulates over a decade.

The specific thing that goes missing

Consider a BOM in a spreadsheet. It contains rows, and each row contains a part number. What it does not contain:

  • Which board revision this list belongs to, in a way that survives a copy.
  • Whether a row's part is the approved one or a temporary substitution made

during a shortage.

  • Which other products use the same part.
  • Which version of the layout these designators match.
  • Who changed row 48 last March, and why.

Every one of those is knowable at the moment the spreadsheet is edited, and none of them is recorded. Six months later the knowledge exists only if somebody remembers it.

The failure mode is silent

The reason this matters more than it sounds is that nothing visibly breaks. A spreadsheet with a wrong assumption still opens. A PDF of a superseded drawing still prints. The error surfaces at the only point where it is expensive: in manufacturing, in a field failure, or in an audit.

Document systems make this worse in one specific way. They are built around the file as the unit, so they faithfully preserve the file and have no opinion about whether its contents still describe the product. The NASA systems engineering handbook names the activity that is missing here: configuration status accounting, the discipline of keeping information about the product consistent with the product itself.

What a good hardware team should know

  • Keep documents. Stop using them as the index. A datasheet should hang

off a component object, not be the only evidence the component exists.

  • The spreadsheet is a tool, not a system of record. Use it to calculate.

Copy the conclusion somewhere with identity and versioning.

  • Any field you would miss if it vanished is a field that needs an owner.

If the only copy of a decision is in a cell comment, it is already lost.

  • Test it by export. If every file were exported to a new folder tomorrow

with its name intact but its location gone, how much would you still know?

Where structured engineering data helps

XKEMA keeps the objects and the relationships as data, and attaches documents to the object and version they describe. A datasheet remains a PDF. What changes is that it hangs off a component, that component is referenced by BOM rows, and those rows belong to a board revision with a release state.

None of that makes PDFs or spreadsheets wrong. It makes them what they are good at being: documents, in a place where the relationships are recorded somewhere else.

Sources

Where to go next

AI and engineering data

Why XKEMA Does Not Start With AI

Infrastructure before inference. The order is a design decision, not a limitation.

¿Te ha resultado útil?

Reddit LinkedIn

The XKEMA newsletter

Hardware engineering worth keeping

New courses, engineering notes and hardware guides, whenever there is something genuinely worth reading. Nothing else.

We will email you to confirm. You can unsubscribe from any issue.

XKEMA, in one sentence

Engineering knowledge management software for companies that develop electronic products: the product structure, its bills of materials, its documentation and its revisions, all in one place.