---
id: KB-TD-019
url: https://app.codecontract.io/help/day-to-day/when-what-you-need-is-not-a-document
idioma: en
categoria: trabajo-diario
subcategoria: encontrar
audiencia: usuario
nivel: basico
actualizado: 2026-08-13
tambienEn: [es]
relacionados: [KB-TD-010, KB-DI-009]
citadoPor: [KB-TD-007]
---

# When what you need is not a document

_«When does it expire?», «what is the cover?». The answer is not a file: it is one line inside one._

**Responde a:** searching for a value inside a document · when does this supplier's insurance expire · getting a figure without opening the pdf · searching by amount or date

Half a dozen times a day the question is not «where is the contract» but «how long is it valid» or «what does it cover». Those are different questions, and searching for them the same way is what turns five seconds of work into five minutes.

## Two searches that get confused

| What you need | What to search for | How it ends |
| --- | --- | --- |
| The document, to send it | By its name or who issued it | You download and send it |
| A value from the document | By the value, not the file | You answer without opening anything |
| Everything meeting a condition | By the value, across all at once | You get a list, not a document |
| Whether something changed | By the value, comparing versions | You see what changed and when |

> [!IMPORTANT]
> The mix-up is costly for a not-obvious reason: **a file's name says nothing about what is inside**. `scan_2024.pdf` may be the policy expiring next month, and no naming scheme, however good, will tell you the sum insured. When the question is about a value, searching by name only leads you to a document you then have to open and read — and with thirty suppliers, to thirty documents.

## What has to be decided once

1. **Which values you are actually asked for each week** — Expiry, amount, number, holder: usually four or five.
2. **That those get extracted on upload** — A value that is not extracted cannot be searched or watched.
3. **And check them the first time** — A misread value spreads to every answer that depends on it.

> [!WARNING]
> The consequence of holding values outside the document is the one people do not anticipate: **what is extracted can be monitored; what lives only inside the PDF cannot**. An expiry date existing only as text on a scanned page can warn you of nothing, however neatly it is filed. Deciding what gets extracted is therefore not a configuration task: it is what separates a tidy archive from one that works for you.

## Tricks that work when it will not turn up

**En corto**

- Search by who issued it, not by what you called it.
- Try the reference number: official documents nearly always carry one.
- And if you are after a date, search the month: everyone writes years differently.

> [!NOTE]
> If a value is asked for often and is not extracted, it is worth adding even if the old ones need a manual pass. That is half an hour recovered in the first month.

**Can I search inside the document text?**

Yes, and it helps; but an extracted value can also be watched and compared.

**What if the document is a photo?**

It reads the same, though check the value the first time.

**Is it worth extracting everything?**

No: only what you are asked. Extra extraction adds review with no use.

## Ejemplos

**A company opens thirty policies to find which expire this quarter.**

- Extracts the expiry on upload and queries the value instead of the document

→ The list appears in a second and warns on its own when one comes close.

**What is needed is not a document but knowing whether something was done and when.**

- Checks the file's history
- Searches for the action rather than the file

→ The question is answered even though no paper answers it.

**A document that never existed is hunted for.**

- Checks whether what happened was a recorded action

→ You stop looking for what is not there.

**You need to know who decided something.**

- Checks that decision's log

→ The answer has a name and a date.

**You need to prove that warning was given.**

- Checks the send log

→ The warning is proved without a signed document.

**A report nobody ever produced is requested.**

- Checks the data and generates the report now

→ The answer exists even though the report did not.