What document intelligence is, and when it is not worth it

The short version #

Document intelligence means a system reads your documents, understands what is inside them, and lets you retrieve that by asking a question instead of by searching for a file.

The difference from ordinary search: ordinary search finds documents containing a word. This answers the question — and, done properly, shows you the page it took the answer from.

What it does, step by step #

• Reads the file, including scans and photographs, not only clean digital text
• Pulls out what matters — dates, amounts, parties, obligations, terms
• Keeps the structure, so a 40-page lease is not treated as 40 pages of equal weight
• Answers questions in plain language, citing the source
• Watches the dates that matter and warns before they arrive

Where it earns its keep #

• Leases and contracts with escalation clauses, notice periods and renewal dates — the places money leaks quietly
• Invoices and receipts you need to check against something, months after you filed them
• Policies and manuals that people ask the same three questions about, over and over
• A portfolio that outgrew your head. Nine leases is already past it, and nobody notices the moment it happens
• Anything you signed years ago and have not opened since — the person who knew what was in it is you, in 2021

When it is not worth it #

This is the part most vendors skip. It is also the part worth reading, because it is your money.

• Documents you can already find in thirty seconds, with no dates in them that expire. A folder and a calendar reminder are enough. Volume is not the test — findability and consequence are
• Documents nobody ever consults. The problem there is not retrieval, it is that nobody needs them
• A team where someone else genuinely knows the documents and the process is written down. Then a system is insurance rather than efficiency — worth buying deliberately, or not at all
• A process that is broken upstream. Fix the process first, or you will search a mess faster

The one thing to insist on #

Whatever you buy or build, insist that answers show their source. A system that says "your renewal is in March" and cannot show you where it read that is asking for trust it has not earned.

The moment an answer is checkable in two seconds, the tool becomes something people actually use for decisions that cost money.

An answer you cannot check is a rumour with good formatting.

Full disclosure: I built one #

I wrote that last line first, and then could not ship anything that failed it. AskRoby answers with the exact page it read the answer from, every time.

It is priced in contracts because contracts are the case I know: fourteen years running a creative district with fifty-three tenants, where almost every answer was inside a document nobody wanted to open.

Trying it is free and takes about a minute: two contracts, twenty questions, no credit card, nothing to cancel. The smallest paid plan after that is three dollars a month. The address is www.askroby.io.