AI knowledge management: how can a firm reuse its past work and credentials?

For a consulting firm, an engineering consultancy or any services business, knowledge management means starting from what you've already done: a winning proposal, a design calculation, a similar engagement. An assistant reads the documents your teams already produce, links them by client, project, method and team, and answers with a citation to the source document. Two things are still on you: access rights that are set up properly, and a person who signs off on what goes to the client.

Published by Saména

Why aren't the intranet and shared drives enough?

A traditional knowledge base needs an extra step. At the end of an engagement, someone has to write up the project profile, file the deliverable in the right place and update the wiki page. By then the next engagement has started, so it gets put off and the knowledge base goes stale.

Shared drives keep everything, each team in its own way. You can find a file if you know its name or its folder. You can't find "the calculations for a structure like this one" or "our three best credentials in food manufacturing", because those questions are about what's inside several documents at once. And when the person who knew where everything was leaves, it gets worse: the page on knowledge transfer covers that case.

Atlassian, the company behind Confluence, surveyed 12,000 knowledge workers and 200 executives. According to that survey, teams spend 25% of their time searching for answers, time a firm that bills by the day never invoices.

What changes with an assistant that reads your documents?

A services firm's know-how is already written down, in its proposals, deliverables, meeting notes, calculations, reports and résumés. An AI assistant reads those documents in full, where they are. It knows that a given proposal, report and set of meeting notes belong to the same client and the same team, even when they sit in three different folders.

Every answer points back to the document it came from, so you open the original and check before you use it. The search itself is covered on the page about AI enterprise search, and the technical side on the one about the AI context layer.

A few questions teams ask:

  • Which studies have we done on school buildings since 2020, and who was the project manager?
  • Find the calculations for the footbridge we designed for a city three or four years ago.
  • Which restructuring engagements have we run in healthcare, and what came of them?
  • Put together the credentials page for this RFP: three comparable engagements, with client, year and team.
  • Who here knows this client best?

What are the options today, and where do they fall short?

Four kinds of tools address this, and they can be combined: a custom agent can be used from Claude, ChatGPT or Copilot. Platforms like Glean get a closer look on the page "Is there a European alternative to Glean or Dust?".

AI knowledge management tools (vendor pages read 11/20/2026)
OptionWhat it does wellWhere it falls short
Internal wiki with an assistant (Confluence with Rovo, for example)Answers from the pages people have written. Rovo comes with paid Confluence Cloud plans and also searches other apps.Only knows what someone wrote down and kept up to date.
Enterprise search (Microsoft 365 Copilot, Glean)Searches the connected tools and respects each person's permissions. Glean lists more than 275 connectors.Reads files as they're filed and shared. Microsoft advises reviewing sharing before rolling out Copilot, since SharePoint defaults to the most permissive setting. Glean doesn't publish its prices.
General-purpose assistant with projects (ChatGPT, Claude)Very good on a set of files you hand it.You pick the files by hand: up to 40 per ChatGPT project on Business and Enterprise, 25 on Plus. Claude switches a large project to retrieval (RAG), which expands its capacity up to 10x on paid plans. That works for one project, rarely for years of archives.
Custom agentBuilt on your folders and on the way you sell and deliver; links engagements, people and documents.Takes engineers to build and keep improving, so its value comes down to the gain you measure.

What does it take to make it work?

Whatever the tool, the result depends on five things:

  • Documents linked by client, project, method and team. There's no need to tidy everything up first: the tool has to link documents to one another, in the words your trade uses (job, phase, work package, engagement).
  • Access rights that are actually in place: who sees which client's engagements, and what stays with management (fees, personnel files). A tool that copies SharePoint permissions copies their gaps too.
  • A person who signs off. The assistant drafts the credentials page or the memo; a partner or project manager reviews it before it goes to the client.
  • One specific first use case: the credentials page in proposals, preparing a study, staffing an engagement. Measure the time saved there before going further.
  • A simple close-out. Filing the final deliverable with the rest of the project folder is enough; an agent can draft the project profile at close-out for someone to review.

How does Saména do it?

Saména builds custom AI agents for small and midsize companies. They have read-only access to your documents, in the folders you choose on SharePoint, OneDrive or Dropbox. Your files stay where they are and Saména keeps no copies. Connecting Claude or Copilot alone to SharePoint has them read files one by one; Saména first links each document to the others by client, project, method and team, in the vocabulary of your trade, which is what lets one question pull every comparable past engagement for a tender. A CRM, a time-tracking tool or industry software can be connected on request.

Your team asks its questions in the Saména app, and also from Claude, ChatGPT or Copilot. The agent cites its sources and drafts the credentials page, the proposal or the memo in Word or PowerPoint; someone on your team approves it before it goes out. The agents built for consulting firms, for construction and engineering firms and for RFP responses each have their own page.

At a consulting firm with 15 consultants, 250 GB of documents were made usable: consultants find a past engagement or deliverable in seconds. We estimate the gain at 30 minutes of searching saved per consultant per week, about 48 days a year, nearly €34,000 at a day rate of €700. At a 25-person environmental consultancy, preparing an audit file takes 40% less time, measured on its real files. The details are in the case studies.

Your first agent is up and running one month after the first call, on one specific use you chose, for example finding the past engagements closest to a new tender. The trial lasts four months, and the gain is measured in euros before any long-term commitment; the subscription that follows covers unlimited users. What Saména retains from your documents is encrypted with a key unique to your company, on servers in France with the database in the European Union.

Frequently asked

What's the difference between knowledge management and document management?
A document management system stores files, keeps their versions and controls who can open them. Knowledge management is about using them again: finding the engagement, method or person you need. An AI assistant connects read-only to the folders where your documents live and leaves your document management system in place.
Who should own the project on our side?
Someone who knows the engagements and can tell whether an answer is right, such as a partner, a project director or the person who puts proposals together. During the trial, they meet with Saména's engineers once a week.
Can we cite an engagement covered by a confidentiality agreement as a credential?
The contract with that client decides. The assistant finds the engagement; the person who approves the proposal decides whether it can be cited, and how: with the client's name, or only its industry and the size of the project.
Our archives are on a file server or Google Drive. Does that work?
Yes. SharePoint, OneDrive and Dropbox connect directly; a file server or Google Drive is connected on request, read-only like the others.
What about knowledge that isn't written down anywhere?
The assistant can only find what left a trail. The rest has to be captured: a short lessons-learned note at the end of an engagement, filed in its folder, gets read like every other document.

Sources consulted

The pages we read to write this one, and the day we read each.

Let's spend 45 minutes looking at what an agent could pull from your past engagements. No commitment.