Reuse past project work with AI in new engagements
5 minutes read
Every consultancy has solved a version of the current problem before. The work sits in old project folders where nobody can find it, and some of it was never the firm's to reuse. Here is how to sort it.
To reuse past project work with AI, a firm has to do three things in order: separate what it owns from what belongs to a client, approve the pieces that may be reused, and make sure each reused piece is cited wherever it turns up in new work. Most of the effort goes into deciding what is allowed, and very little into the technology.
Consultancies redo work for a simple reason. The method that solved last year's problem lives in a folder named after a client, written up for that client, and the person who built it is on another project. Starting again is faster than finding it.
Why old project folders are the wrong source
Old project folders mix two kinds of material that must be treated differently. One is the firm's own thinking: the method, the framework, the interview guide, the structure of the analysis. The other is the client's: their data, their documents, and the findings that were specific to their situation.
Pointing an AI tool at the whole archive treats both the same. The risk is easy to state. A draft for one client quotes a number, a phrase, or a finding that came from another client's engagement, and nobody notices because the sentence reads well. The guide to connected knowledge covers why the firm's knowledge has to be connected deliberately instead of all at once.
What can be reused and what cannot
Sort past work by who owns it before any of it goes near a new engagement.
| Material | Owner | Reusable on other clients? |
|---|---|---|
| Methods, frameworks, and templates | Firm | Yes, once approved |
| Interview guides and survey designs | Firm | Yes, with client-specific questions removed |
| Anonymised lessons and patterns | Firm | Yes, if no client can be identified from them |
| Client data, documents, and correspondence | Client | No, only on that client's own work |
| Findings and recommendations for a client | Client | No, unless the contract and the client both allow it |
The middle row is where judgement is needed. A lesson such as "market-entry studies in this sector need at least two rounds of interviews" is the firm's. The same lesson with a client's name, market, and numbers attached is not. Stripping the detail is a person's job, and it should happen once, when the lesson is written up, so nobody has to redo it under deadline later.
Turn finished projects into library entries
The practical unit of reuse is a short library entry. A copied folder does not do the job. At the end of a project, the team writes up what the firm learned in a form that stands alone: the method used, when it fits, what went wrong, and the template or guide that came out of it. A page or two is enough.
Each entry needs an owner who approves it. Approval is the check that the entry contains nothing a client would object to seeing reused, and that the method is one the firm is willing to repeat. An entry that nobody has approved is a draft and should be labelled as one, so anyone who finds it knows how much weight to give it.
This is slower than indexing everything. It is also the only version in which the firm can say what its AI tools draw on.
Keep client material inside its own engagement
Client files stay useful, but only on that client's work. A returning client's earlier reports, data, and decisions are often the most valuable context for the next project with them, and there is no reason to lose that.
The control that matters is the one the files already have. If a consultant cannot open a client's folder in the firm's document system, an AI tool acting for that consultant should not read it either. Carrying existing permissions over is simpler and safer than building a second set of rules that will drift from the first. Connecting your firm's tools and data to AI explains how that works in practice.
Cite reused work where it is used
Reused material should arrive in a new draft with its origin attached. When a paragraph draws on a library entry, the reader should be able to open that entry. When it draws on a client's earlier report, the reader should be able to open the report.
There are two reasons to insist on this. A reviewer needs to know whether a statement is new research or something carried over, because carried-over material can be out of date. And the citation is the proof that nothing crossed from one client to another: if every reused passage points to a firm library entry or to this client's own files, the question answers itself.
How this looks in Hebno
Hebno keeps the firm's methods, standards, and notes in one library, where an owner approves each entry and unapproved entries are marked as drafts. Client files come from Microsoft 365 and Google Workspace, and Hebno reads only what the person's account can already open. A folder can be linked to a client project, and every answer cites the entry or file it used. The knowledge feature page shows the library and the client file connection.
