Knowledge & internal search

The answer that exists, in a document nobody can find

Somebody already solved this. In 2023. In a PDF.

Procedures, specifications, contracts, past jobs and email history — made genuinely searchable and answerable, with the source shown, so expertise stops walking out of the building.

How it runs today

You will recognise at least three of these.

  • Knowledge is spread across a shared drive, an intranet, email, and people's heads.
  • Search means knowing the filename, or knowing who to ask.
  • The same question gets answered from scratch repeatedly by different people.
  • Procedures exist in several versions and nobody is certain which is current.
  • When an experienced person leaves, a large amount of context leaves with them.
The tells

Signals it is costing more than anyone has measured.

  • New starters take far longer to become productive than the work itself justifies.
  • You have re-solved a problem you had already solved, at cost.
  • "Ask Dave" is a documented step in more than one process.
  • Contract terms are re-litigated because nobody could find the clause.
  • The intranet exists and nobody uses it, which everyone quietly accepts.
Rebuilt from the ground up

What it becomes.

The corpus stops being a filing system and becomes something you can ask. Documents, past jobs, specifications, contracts and approved procedures are indexed together, and a question in plain language returns an answer with its sources — the specific document, the specific clause, the specific past job — so the answer is checkable rather than merely plausible. That citation requirement is the whole design: an assistant that cannot show you where an answer came from is worse than useless in a business context, because it is confidently wrong at scale. Permissions are enforced at retrieval, so people can only get answers from documents they were allowed to read. And version currency is explicit, so a superseded procedure is labelled rather than quietly returned.

What the system does

  • One index across documents, past jobs, specifications, contracts and approved procedures
  • Plain-language questions with cited, checkable answers
  • Permission-aware retrieval — you only get answers from what you may read
  • Explicit version currency, so superseded documents are labelled as such
  • Gap reporting: the questions asked most often that nothing answers well

What it connects to

  • Your document storage — SharePoint, Drive, or a file share
  • The operational systems holding past job and case history
  • Identity, so retrieval respects existing permissions rather than inventing new ones
  • Chat or intranet, so answers arrive where people already ask

It augments the systems of record you already run. Nobody is asking you to replace your accounting package.

The first release

Where we would start, and how long it takes.

One well-defined corpus — usually procedures plus past jobs — searchable and answerable with citations, in one to two weeks. We measure it by whether people stop asking the human, which is a blunter and more honest metric than any accuracy score.

Others in data & decisions

Let's build

What's your AI Nirvana?

Tell us where you want to go. We'll bring the team, build the product, and grow it with you — and you own it.

  • You own the IP
  • US-based team
  • Reply within 1 business day
Get your free AI plan →