Skip to content
KudosCRM

Founding offer: we set up your CRM for you — free for the first 100 teams. Book your setup

Help · AI in KudosCRM

Extract a document into fields

Hand Kudos AI a PDF or document and it reads the text and proposes structured field values — company, contact, amounts, dates — for you to review and confirm before anything is saved.

Updated August 2026

Re-keying a signed order, a vendor form, or a business card into the CRM is exactly the kind of work nobody should be doing by hand. Kudos AI reads the document for you and proposes the structured values it found — names, companies, amounts, dates — mapped to fields, so the tedious typing becomes a quick review.

This is a real Claude extraction, not a brittle template parser: it reads the actual text and infers the fields. And like everything in Kudos AI, it's human-in-the-loop — the assistant proposes, you confirm. Nothing is written to a record until you've checked it, which keeps a person between the model and your data.

Who this is for

Anyone who finds themselves copying details off a document into the CRM — ops, sales, and onboarding teams especially.

Before you start

  • A document to extract from (for example a PDF or a pasted block of text).
  • The record or object you want the fields to land on.
  • AI enabled for your workspace (usage is logged and quota-guarded).

What's real AI here — and how it stays safe
Document extraction is a genuine Claude (haiku-4-5) call that reads the document and proposes structured field values. It's human-in-the-loop by design: the assistant only suggests — you review and confirm before anything saves, so you can catch a misread before it touches the record. Everything is AI-labeled and usage-logged, and quota-guarded per workspace.

Extract and confirm

  1. Bring in the document
    Provide the document you want read — a PDF, an attachment, or pasted text. Kudos AI works from the text it can read in the file.
  2. Run the AI extraction
    Choose extract. The assistant reads the document and proposes the structured values it found, mapped to fields — the company, the contact, amounts, dates, and so on.
  3. Review the proposed fields
    Check each proposed value against the source. The assistant is reading, not guessing from a fixed template, but you're the one who confirms — so a misread never silently lands on a record.
  4. Fix anything that's off
    Correct or clear any value before you accept. This review step is the point: you stay in control of what's true on the record.
  5. Save the confirmed values
    Accept the extraction and the confirmed values write to the fields. Until you do, nothing is saved — the document hasn't changed your data on its own.

What you get

  • Structured field values pulled from a document instead of typed by hand.
  • A review step where you confirm every value before it saves.
  • A real model that reads the text, not a rigid template that breaks on a new layout.
  • AI-labeled, usage-logged extraction that never writes to a record on its own.

Frequently asked questions

Does the extraction save fields automatically?

No. The assistant proposes the values it found; you review and confirm before anything saves. Until you accept, the document hasn't changed your record. That human-in-the-loop step is deliberate — it lets you catch a misread before it lands.

Is this a real AI, or just a template parser?

It's a real AI — a genuine call to Anthropic's Claude (haiku-4-5) model that reads the actual text and infers the fields. That's why it handles documents it hasn't seen a fixed layout for, instead of breaking the moment the format changes. Every extraction is AI-labeled and logged.

What kinds of fields can it extract?

Common structured values — company and contact names, amounts, dates, and similar fields the document contains. You map and confirm what lands where, so you keep control over the result.

What happens if it misreads something?

You'll see it in the review step and can correct or clear it before saving. Because nothing writes until you confirm, a misread never silently becomes the truth on your record — that's exactly what the review step is for.

Is there a usage limit on extraction?

Yes — AI usage is logged and quota-guarded per workspace, the same as the rest of Kudos AI. You can see what's been used, and the assistant tells you if you reach the limit rather than failing quietly.

Learn more

  • AI assistant (feature) — Everything the writing and extraction assistant does, on the feature page.
  • Kudos AI overview — The AI journey stage — what's real, what's grounded in Claude, and what's coming.

Related articles

Have an account issue this guide doesn't cover?

Open a ticket Track your requests

Start free today

Ready to give your team a CRM they'll actually use?

Start free. Bring your whole team. Cancel whenever (you won't).