Consulting Project Folder Structure for AI: A Practical Team Template

Your team has the client brief, interview notes, spreadsheets and a growing collection of AI conversations. Yet each new piece of work starts with someone explaining the engagement again: what the client wants, which figures to use and what the team has already decided.

Split the engagement into a curated Shared area and Private workstream folders. Shared holds the authoritative knowledge the team uses together; Private holds work in progress, with one person accountable for each workstream. Keep unfinished material, raw inputs and superseded versions outside routine AI discovery, and deliberately select what each task should use.

That gives collaborators and AI a clear starting point without treating every available file as evidence.

Start with the context your team needs to share

A shared folder makes files available. Shared context explains how those files relate to the work.

For a consulting engagement, that context includes the client question, scope, working hypotheses, evidence, decisions and next steps. A market report becomes more useful when the team knows which question it supports. An interview summary becomes easier to interpret when its limitations and outstanding questions sit alongside it.

Think of the engagement folder as the place where a colleague can understand the work well enough to contribute. The same information helps you give AI a specific assignment: the question to answer, the sources to use and the output to develop.

Collivo is built around this shared-context approach: keep the material, guidance and ongoing work connected across the engagement. Start with the team’s information needs, then arrange the files to support them.

Curate what AI can draw on

An answer can look convincing while drawing on the wrong material. An exploratory slide might contain a hypothesis that was later rejected. An older workbook might use a different reporting period. If those files are available alongside current evidence, AI can carry their contents into an otherwise plausible analysis.

Keep a clear distinction between material the team has checked and accepted for a particular use, material still being developed, and historical records. A polished draft is still a draft; an AI-generated summary does not become evidence because it has been saved in the project. Curate the relevant information before using it as context for analysis.

Exclude scratch work, abandoned hypotheses, old versions and raw intake from routine AI search and discovery. Bring an item into a task deliberately when it is needed—for example, to analyse original interview transcripts, clean a raw extract or revise a draft. Identify its role and limitations, and check the resulting findings before using them in subsequent work. When a finding needs verification, supply the relevant original material too. Preserve the originals so the team can trace the work back.

Keep engagement analysis scoped to the project

For engagement work, avoid giving AI open-ended company-wide search access. Select the relevant client material, approved methods and reference documents, then bring them into the project space through scoped imports, uploads or references. If a task needs more evidence, expand that set deliberately.

Being allowed to access a document does not make it relevant or authoritative for this engagement. A proposal from another client or an internal brainstorm can resemble useful evidence while answering a different question. Have a team member select and review material from the wider firm's knowledge before adding it to the engagement's working context.

Separate Shared knowledge from Private work in progress

Make Shared and Private the two top-level areas. The distinction tells team members what they can develop within their own workstream and what others depend on as the engagement's accepted knowledge.

Engagement/
├── Shared/
│   ├── engagement-brief
│   ├── current-sources/
│   ├── agreed-definitions-and-decisions/
│   └── reviewed-findings-and-deliverables/
└── Private/
    └── team-member/
        └── customer-analysis/
            ├── task-requirements
            ├── progress-journal
            ├── working-files/
            ├── raw-inputs/
            └── archive/

Raw inputs and archives stay with the workstream so its owner can find them, but they are excluded from its default AI context. Apply the tool's source-selection or discovery controls and supply those materials only when a task calls for them; giving AI the workstream context should not mean giving it every nested file.

The example uses a team-member → workstream hierarchy. You can also organise Private directly by workstream or by team member. Choose the arrangement that makes ownership easiest to understand. Whichever structure you use, name one person accountable for keeping each workstream folder clean and current.

Shared: the knowledge everyone can build on

Shared contains the engagement brief, selected current sources, agreed definitions, decisions and reviewed findings. The engagement lead curates it so each workstream has a common foundation.

Team members should treat changes here with care: replacing a source or changing an agreed definition can affect several analyses. Develop changes in the relevant Private workstream, then bring them into Shared through the team's agreed review process.

Keep one authoritative location for a source and reference it from the workstreams that use it. When a replacement is accepted, update those references and retain the superseded Shared source in the engagement lead's Private curation archive, outside routine AI discovery.

Private: room to develop the work

Private holds the in-progress workstreams: analysis, drafts, raw inputs, experiments and session notes. Each owner keeps relevant working files together and moves abandoned or superseded material out of the active task context.

Deliberately select which Private material AI can use for each task. An analyst can ask it to examine raw interviews or develop a draft without making those materials sources for every other workstream. Give the task the relevant Shared knowledge, its own requirements and progress record, and the working files it needs.

The names Private and Shared describe the working convention; configure actual permissions to match who may read and modify the material. Agree what visibility the engagement lead needs into Private workstreams for progress updates and oversight.

Give every workstream requirements and a progress journal

Keep two short records inside each workstream folder. Task requirements explain what the work needs to achieve. A progress journal records where it has reached.

Together, they support daily progress updates and help a collaborator or AI agent resume the task later without reconstructing it from previous conversations.

For illustration, a customer-analysis workstream might have these requirements:

Identify the customer segments to prioritise. Use the reviewed September extract and agreed segment definitions in Shared. Prepare a recommendation memo that distinguishes evidence from hypotheses and makes gaps in the interview coverage clear. The customer-analysis lead owns the work.

Its latest journal entry could read:

Segment profiles drafted in the working memo. The recommendation remains open because smaller customers are underrepresented in the interviews. Next: assess whether additional evidence is needed before prioritising segments. The current memo and interview-coverage table are linked here.

The requirements remain the reference for the task; update them when the scope changes. The journal changes as the work progresses. Record durable engagement-wide decisions in Shared through the agreed review process, rather than leaving them only in one workstream's notes.

End each session by updating the journal

The person doing the work maintains these records, with AI helping to prepare updates. At the end of a session, ask it to summarise what changed, link the working outputs and record the unresolved questions and next step. Review the entry before saving it.

At the beginning of the next session, ask the agent to read the requirements, latest journal entry and relevant sources, then summarise where the task stands before continuing. Supply those records through the tool's supported context controls; storing them in a folder alone does not ensure they are read.

For example: “Read the task requirements and latest progress journal entry, then summarise where the work stands, what is still open and the next action you recommend.”

Keep the journal useful rather than exhaustive. It should record the state needed to resume the work, not reproduce the entire conversation.

Use the same records for daily progress updates

Before a daily check-in, review the journal against the requirements. What has moved forward, what is blocked and does the next action still serve the task's objective? The workstream owner can use those records to prepare a short update, with AI helping to summarise the relevant entries.

This gives the engagement lead a consistent view across workstreams without asking everyone to reconstruct their day from memory. Keep the journal in the working folder as the detailed record; share the relevant progress update through the team's agreed channel.

Agree how work becomes Shared knowledge

The engagement lead is accountable for curating Shared. The team decides the review controls that suit the engagement: a lead's review may be enough for a checked source update; a recommendation may need another reviewer before it becomes a Shared finding.

Make the route clear: develop in Private, review, then publish the accepted material or reference to Shared. Before publishing, confirm its source, purpose and current status, and identify anything it replaces. Everyone should know who may make that change and how affected collaborators will hear about it.

Keep the controls light enough that your team will actually maintain them, with more care where a change affects several workstreams or the client deliverable.

The useful distinction is between a workstream's developing interpretation and a finding the engagement has accepted. AI can help prepare and review the work; it should not turn an unreviewed draft into the team's authoritative knowledge simply by saving it in Shared.

What a handoff looks like with shared context

Imagine an analyst has drafted the customer segment profiles and another consultant is taking over the recommendation. The workstream folder contains the task requirements, latest journal entry and current memo, with references to the reviewed extract and agreed definitions in Shared. The handoff identifies which working files the next task needs.

The incoming consultant can see what has been done, why the recommendation is still open and which evidence gap needs attention. They can ask AI to review that gap using the requirements, journal and selected sources, then assess its suggestions against the underlying material.

When work changes hands, record the new accountable owner in the requirements and journal. If Private is organised by team member, move the workstream to the incoming owner's area and update its references and permissions as needed.

This is also where shared context supports review. The engagement lead can follow the recommendation back to its evidence and see which questions still require judgment.

Collivo brings engagement files, folder-level AI guidance and collaborative work into a shared workspace. That gives this way of organising an engagement a place to live, with the team’s working context alongside its deliverables.

Start with one active workstream

Choose a workstream your team is already developing. Give it an owned Private folder with task requirements and a progress journal, then link the current sources in Shared. Agree who reviews material before it joins that Shared knowledge. Check that the next AI task uses those selected inputs, then ask a colleague to pick up a small piece of work from that starting point.

Notice what they still need to ask you. Add the missing context to the requirements, journal or source references, and use that same starting point for the next AI task. You now have a concrete basis for extending the structure across the engagement.