Skip to content

Read in order ยท 9 articles

BigDocs

How to get complete, exact and repeatable results from AI on big-document work, and how to measure it. Read in order, starting with the first piece.

Benchmark results: structured pipeline against one big prompt
Structured pipelineOne prompt, changes onlyOne prompt, every paragraph
Knowing you have it all
Stopped early without saying so0 of 11Can't tell3 of 3
Paragraphs accounted for100%Not reported86 to 89%
New term no policy covers, sent to a person3 of 3NoNever reached
Getting it right
Planted departures classified correctly (of 15)141010 to 12
Same answer next time
Same classification in all 3 runs14 of 15Not tested7 of 15
Agreement on how serious each one is96%Not tested64%
Showing why
Decisions citing a checkable source85 to 87%Not reported1 to 17%
The redline
Words deleted per changed paragraph6.5About 22About 10
Section deleted without a markern/an/a3 of 3 runs
Effort
Cost per runUS$7 to 10US$0.52US$4 to 4.70
Time per run5 to 6 min3 minAbout 17 min

How the benchmark was run

  1. BigDocs: rolling the dice on big documents

    A lot of business still runs on big documents.

  2. The Agentic Wordhouse: a dimensional model for your words

    Most organisations keep their words the way their systems keep transactions: one document per occasion.

  3. Structure first: documents are written for human eyes

    A contract, a tender or an enterprise agreement is written for a person turning pages.

  4. The instruction file: write the job down before it runs

    Give an agent a big document and a library and ask it to do a thorough review.

  5. Never retype: if the model can't retype it, it can't mistype it

    The most natural request in the world is "give me the contract back with your changes".

  6. Output built for oversight: put people where judgement is needed

    A tidy redline is a strange thing to hand a reviewer.

  7. Maintaining the Wordhouse: keeping it true as it grows

    Every knowledge base starts with good intentions.

  8. Build the tests in: borrowing confidence from engineering

    Software teams stopped trusting "it looks right" a long time ago.

  9. BigDocs: guiding principles

    The BigDocs series is about one kind of work: big documents, often several at once, where the business transaction rests on getting the answer exactly right.

Working with me on big documents