Knowledge management at consulting firms: from shared drives to AI-ready records

What counts as knowledge at a consulting firm?#

Knowledge, for a consulting firm's KM program, is what the firm learns on engagements and is entitled to reuse: methodologies, proposal language, project lessons, benchmarks it is allowed to keep, and a map of who knows what. It is the firm's own know-how, not every client file the firm happens to store.

At many mid-size firms, KM amounts to a shared drive organized by client, a few partners who remember where the good material is, and a proposal library someone maintained for a while. That holds up while the firm is small. It breaks when new hires cannot find past work, when senior partners retire, or when an AI assistant is pointed at the drive and returns the wrong client's numbers.

The KM maturity table: where does your firm sit?#

The KM maturity table describes five stages, from scattered files to governed records, and what AI can realistically do at each one. Many firms sit between the second and third rows, often with one practice ahead of the rest.

The jump that matters most is from the second row to the third. Until firm-owned knowledge is separated from client material, every AI tool placed on top of the drive inherits the mix.

Why client-named folders defeat search and create risk#

Client-named folders defeat search because they answer whose work a file was, not what it was about. A consultant looking for prior pricing-strategy work for a distributor has to know which clients those were, open each engagement folder and guess which file was final.

The same structure creates confidentiality risk once AI tools arrive. Many enterprise search and assistant tools follow existing file permissions, so anything a user can open, the assistant can quote. If engagement folders are broadly shared across the firm, one client's data can surface in a draft for another. Restricting client folders to their engagement teams and building a separate, curated firm library addresses both problems at once.

Sort firm knowledge from client material first#

Sorting firm knowledge from client material is the first real KM task, and four ownership classes cover nearly everything. The engagement letter or MSA decides which class a document falls into, so treat the firm's standard terms as the default and flag engagements signed on a client's paper.

Borderline cases are common: a playbook refined during one client's engagement, or a benchmark built from several clients' data. Record who decided the classification and why, and send unclear items to whoever handles contracts. That note becomes useful evidence later.

What metadata makes a consulting record AI-ready?#

AI-ready metadata tells a retrieval system, and a human reviewer, what a record is, where it came from and how it may be used. A workable starting set fits on one row of a spreadsheet per item.

This mirrors how dataset documentation is developing outside consulting. The Data & Trust Alliance's Data Provenance Standards, an open specification, group dataset metadata into Source, Provenance and Use, and state that this metadata is needed to enable proper dataset selection for AI model training. Their Use group includes elements such as confidentiality classification, license to use and intended data use.

• 1Subject tags: service line, industry, problem type and engagement type.
• 2Source: originating system, folder path and engagement or pursuit code.
• 3Ownership class and confidentiality level, carried over from the sorting step.
• 4Status: current, superseded or draft, with the date it was last reviewed.
• 5Owner: the partner or practice lead who answers questions about the item.
• 6Restrictions: NDA coverage, return-or-destroy duties and the retention date.

A cleanup sequence that does not stall billable work#

A KM cleanup that protects billable time runs one practice at a time, with operations staff doing most of the sorting and partners deciding only the borderline cases. Trying to fix the whole drive at once is the usual reason these projects stall.

The other common mistakes are predictable: copying every folder unchanged to a new platform, letting IT decide what counts as firm knowledge, connecting an assistant to the whole tenant before permissions are fixed, and deleting old versions to make the library look tidy. Archived versions show how a pricing model or diagnostic changed across engagements, so keep them unless a retention schedule or client agreement says otherwise.

• 1Step 1: choose one practice with an engaged partner sponsor.
• 2Step 2: list the systems that hold its knowledge, such as SharePoint or Drive sites, Teams channels, Confluence or Notion spaces, the CRM and the PSA tool.
• 3Step 3: classify folders into the four ownership classes, starting with the most-used ones.
• 4Step 4: copy firm methods and de-identified project reviews into a curated library with named owners.
• 5Step 5: label superseded versions and move them to an archive instead of deleting them.
• 6Step 6: add pursuit IDs and project codes so library items link back to outcomes.
• 7Step 7: connect the AI assistant to the curated library only, then widen access as tagging matures.

Illustrative: a consultancy uses a drive migration to rebuild its library#

Illustrative: a fictional operations consultancy of about 110 people, serving utilities and transport operators, is retiring an old Box account and moving to SharePoint. Two founding partners plan to step back within a few years, and much of the firm's method lives in their folders and inboxes. IT proposes copying every client folder to the new tenant as it stands.

The managing partner stops the straight copy. The operations lead classifies the most-used folders into the four ownership classes before anything moves. Client confidential folders migrate as restricted engagement sites, and final deliverables are referenced from the engagement record rather than copied into the library. The founders set aside a few hours a month to confirm which of their playbooks are current, and each playbook gets a named successor owner.

When the curated library opens, every item carries subject tags, an ownership class, a status and its PSA project code, and archived versions stay retrievable. The firm connects its assistant to the library only. It also holds a record-family inventory, with dates and owners, that a later licensing review could start from without reopening the old drive.

From KM program to licensing candidate: the SourceX view#

A well-run KM program produces the same inventory a licensing review needs, which is why SourceX treats KM cleanup as a practical first step rather than a separate project. The ownership classes map directly onto the Rights step of the SourceX five-step transaction: Supply, Rights, Preparation, Approval and Delivery.

Preparation still needs care. Client names, figures and individuals must come out of the firm's records about the work, and automated tools help without finishing the job: Presidio, an open-source de-identification SDK, warns in its own documentation that there is no guarantee it will find all sensitive information. Preparation therefore pairs automated detection with human review, the result is recorded in the SourceX Evidence Packet, and the firm approves every step. Any license grants defined use of a prepared copy; the records are licensed, not sold, and the firm keeps ownership.