How can a consulting firm reuse what it learned on past engagements with AI, without mixing client data?

Short answer

Keep two layers apart. Firm know-how (methods, templates, anonymized lessons) goes into a project shared with the whole firm. Client material goes into one project per engagement, shared only with that team; in Claude you also turn off Use account memory there, and in ChatGPT shared projects use project-only memory by default. Personal AI memory is the risk: it carries one client into every other chat.

The two kinds of knowledge in a firm

A consulting firm sells two things: what it knows in general (frameworks, benchmarks, the way it runs a workshop, what went wrong last time), and what it learns inside one client (their numbers, their politics, their plans). The first should travel to every engagement. The second may never travel at all.

AI tools blur that line in one specific place: personal memory. A consultant who spent a week with Claude or ChatGPT on client A carries that into every chat outside a project, including the one where they draft a proposal for client B.

Layer Example Where it belongs Who sees it
Firm know-how Pricing model, interview guide, anonymized case lesson One project shared with the whole firm Everyone
Engagement material Client data, interview notes, draft report One project per engagement Only that engagement team
Personal AI memory Whatever the AI picked up in someone’s chats Ideally nowhere near client material Only that consultant

Setting it up in Claude

  1. A firm-wide project for methods and templates, created with Share with your broader organization (available on Team and Enterprise, where admins allow it).
  2. One project per engagement, created as Keep it private, then shared via Share project with the engagement team only, as Can view or Can edit.
  3. Turn off Use account memory under Memory on each engagement project. By default a project uses its own memory alongside the consultant’s account memory, and new project memory can show up in chats outside projects. With the setting off, the project is closed in both directions.
  4. Put what everyone must know in project knowledge, not in chats. Anthropic notes that context is not shared across chats in a project unless it is in the project knowledge base.
  5. Close the engagement properly. Archiving keeps members and access; remove people explicitly when the work ends.

Setting it up in ChatGPT

  1. Shared projects on Business, Enterprise and Edu can be shared with individuals, workspace groups, or the whole workspace via a link (workspace links join with chat access by default).
  2. Memory is already separated. OpenAI states that shared projects automatically use project-only memory and cannot access a member’s personal memories or context from outside the project.
  3. Limits to plan for: up to 100 collaborators and 40 files per workspace project.
  4. Offboarding: someone who leaves a shared project can copy their chats into a new project of their own. Owners and workspace owners can remove collaborators at any time.

The common misunderstanding

Many firms think the risk is a consultant pasting client A’s file into client B’s chat. The quieter risk is memory: a phrase, a number or a strategic detail that the AI remembered on its own and offers again elsewhere. That is why the memory setting per project matters more than the sharing setting.

What this means for the firm

The firm’s real asset, what it learned across engagements, ends up where it is least reusable: spread across personal memories that cannot be transferred, curated or audited. Someone has to turn engagement lessons into firm know-how, deliberately and anonymized, or it leaves with the consultant. The same tension shows up in law firms and marketing agencies.

Where Bombos fits

Bombos is a plugin that gives every AI tool a firm uses, Claude, ChatGPT, Copilot and your own agents, one shared memory of the firm instead of the person. Knowledge and learned rules can be scoped, firm-wide or to one engagement, and every answer shows what it rests on. It runs in our cloud or on your own server, and one consultant can try it first with their own agents.

Related questions

Does archiving a Claude project remove the client team's access?

No. Anthropic states that archiving keeps all members, permission levels and project knowledge. To revoke access, remove people in the project's sharing settings. An archived project cannot be deleted until you unarchive it.

What happens to a consultant's chats when they leave a shared ChatGPT project?

They get the option to copy their chats in the project before leaving; the copies move to a new project in their own sidebar. Decide in your offboarding whether that is allowed for client engagements.

Can the firm see what a consultant's AI remembers about clients?

In ChatGPT Business, memories are tied to each account and cannot be transferred, even within the workspace. In Claude Team and Enterprise, owners control whether memory is available at all; members review and edit their own. In neither tool is there a firm-level place to curate what each consultant's AI remembers.

Is it enough to anonymize before sharing a lesson?

It is the usual practice for firm-wide know-how, but check your engagement letters: some clients restrict reuse of deliverables, not just names. When in doubt, keep it in the engagement project.

Sources

  1. Claude Help Center: How can I create and manage projects?
  2. Claude Help Center: Use Claude's chat search and memory
  3. OpenAI Help: Projects in ChatGPT
  4. OpenAI Help: Memory FAQ (Business version)
The first half hour · free

What does your AI forget?

  1. 01You tell us which AI you use and where it keeps starting over.
  2. 02We show what a shared memory changes there, on your own examples.
  3. 03You keep the plan, with or without Bombos.
Piet Baudoin
Piet BaudoinYou talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe
We reply the same working day.