Skip to content
Project data

When project knowledge is scattered across drawings, spreadsheets and email

Many studios spend too much time finding the current version of a project fact. A project data foundation connects drawings, product choices, decisions, revisions and documents so the team can see what is reliable, what is missing and what needs professional review.

Short answer

Many studios spend too much time finding the current version of a project fact. A project data foundation connects drawings, product choices, decisions, revisions and documents so the team can see what is reliable, what is missing and what needs professional review.

AI can prepare, compile and flag uncertainty. The architect validates consequence, judgement and responsibility.

architecture

Here, architecture does not mean software architecture. We mean the built environment: architecture studios, renovation, local plans, BR18, materials, building data and architectural decisions.

Control basis

What must the team be able to trust?

Before agent output enters project work, the team must see sources, assumptions, gaps and the next control point. Otherwise AI becomes one more place where project knowledge can turn unclear.

Source-fixed extraction

A useful output for project data foundation should show which information comes from CAD/BIM extracts, spreadsheets, PDF outputs, product data, and which points are based on project assumptions.

Professional sorting

The agent should not only reproduce text. It should help the studio sort what matters for the case, what can wait and what requires human assessment.

Validation track

The output should point to who checks the next step. In this workflow, that especially means that the architect assesses whether the agent's synthesis makes professional sense in the concrete case.

Decision log

Important findings should be traceable to source, status and next action. At minimum, the team should see why a recommendation was included or rejected.

Pilot in practice

How to test without making AI the answer.

The first goal is to test whether the system can gather the information already used in the project and show where it comes from, while the studio checks whether the output actually improves the workflow.

  1. 01

    Start with a real case where the studio knows enough of the answer to assess quality.

  2. 02

    Compare the first output with your manual workflow, and note where it saves time, misses something or becomes too certain.

  3. 03

    Keep the pilot scope narrow: start with one real case and one piece of preparation that currently requires many manual lookups.

  4. 04

    End the test with a decision about where the workflow should enter practice, and which parts are still owned by architect, adviser or leadership.

The need

Where does the need appear in the studio?

The need does not appear because the studio needs yet another system. It appears when project knowledge is spread across drawings, spreadsheets, PDFs, emails and experienced colleagues, making the team spend time deciding what can be treated as reliable.

What can the agent prepare?

  • check_circle Gather the information already used in the project and show where it comes from.
  • check_circle Point out relations between rooms, codes, products, documents, status and revisions.
  • check_circle Prepare drafts for selected outputs where sources, assumptions and gaps are visible.
  • check_circle Flag where the studio must stop and validate before material is used with a client, supplier or authority.

What must the architect validate?

  • verified The architect assesses whether the agent's synthesis makes professional sense in the concrete case.
  • verified The project lead approves which information can be used externally and which must remain internal preparation.
  • verified Specialists or suppliers validate technical, commercial and authority consequences when the agent can only surface a question.
Anonymised example

When the agent's answer must be checkable

In an anonymised project workflow, the same product choice appeared in a drawing, a spreadsheet, a price appendix and a manual. When the variant changed, the point was not to make the system choose again. The point was to make visible which documents still relied on the old information.

  • sync_alt The project basis had to show product, variant, price, documentation and source basis.
  • sync_alt The architect and project lead had to see what was certain, uncertain and ready for approval.
Method

Data sources and uncertainty

The source basis must be visible so the studio can distinguish between data, interpretation and decision.

Data that can be included

  • CAD/BIM extracts
  • spreadsheets
  • PDF outputs
  • product data
  • datasheets
  • project folders
  • revision history

Working method

  • Start with one real case and one piece of preparation that currently requires many manual lookups.
  • Define what the agent may treat as a source, assumption, working hypothesis and stop rule.
  • Use project data as a control basis, not as a hidden decision machine.

Uncertainty and responsibility

A project data foundation does not make old information right. It makes relations, sources and uncertainty visible so the studio can judge whether the basis is strong enough to use.

First pilot

Start with one concrete case.

Choose one project type and one piece of preparation, for example a price appendix or manual. Map which information the agent must find, document and flag as uncertain before an architect uses the output.

FAQ

Frequently asked questions

What is a project data foundation?

add

It is a shared basis for the project's important information so an agent can prepare work with visible sources, relations and uncertainty.

Do we need everything in one database?

add

No. It is often better to start with the relations that already create manual follow-up work in a concrete case.

Is a project data foundation the same as an AI agent?

add

No. The foundation is the basis the agent works from. The agent becomes useful when it can show its sources, assumptions and stop rules.

Editorial responsibility

Author
By
Published / reviewed
Published
Basis
Method and source basisPrivacy Policy: Privacy