BIMinAM AI and Data Engineering Workstream Meeting 02-04-2026

BIMINAM AI AND DATA ENGINEERING WORKSTREAM MEETING, 02-04-2026

Date: 2 April 2026

Purpose of the Session

The session focused on advancing the development of a structured “as-built” information framework and validation methodology, with a strong emphasis on how AI and data engineering can support automated checking, early-stage validation and scalable assurance. The group explored how to translate industry requirements into practical schemas, rules and automated workflows that improve information quality across the lifecycle.

Key Themes from the Transcript

  1. Defining “as-built” through a task-driven framework

The group is developing a structured methodology to define what “as-built” information should include, based on what a person needs to:

  • Identify the asset (location, ID, spatial context)
  • Understand it (specification, manufacturer, warranties)
  • Work on it (tasks, competence, risks, adjacency impacts)
  • Prove completion (checklists, sign-off, evidence)

This creates a practical, validation-ready schema, moving beyond traditional document-based definitions.

  1. From information definition to automated validation

A key objective is to:

  • translate this structured framework into rules and schemas
  • enable automated checking of delivered information
  • identify “digital defects” (missing, incorrect or incomplete data)

This supports both:

  • retrospective validation (e.g. O&M checks)
  • proactive validation during delivery
  1. AI as a tool for checking, not replacing expertise

The discussion reinforced that AI should be used to:

  • process large volumes of documents
  • extract and structure data (e.g. from drawings, O&M manuals)
  • identify gaps and inconsistencies

However:

  • AI must be guided by predefined rules and domain expertise
  • outputs require human oversight and validation

AI is therefore positioned as a productivity and assurance tool, not a decision-maker.

  1. Breaking information into structured data sets

A major challenge identified is converting:

  • unstructured documents (e.g. warranties, drawings, certificates)
    into
  • structured, machine-readable data points

Examples discussed:

  • decomposing warranties into key attributes (duration, conditions, ownership)
  • extracting metadata from drawings (title blocks, status, content completeness)

This is critical for enabling automation and interoperability.

  1. Early-stage validation (“shift left”)

A key insight was the need to move validation:

  • from handover stage
    to
  • authoring stage

Proposed approach:

  • AI-assisted tools that prompt authors in real time
  • “have you included this?” checks during data creation
  • automated identification of missing information

This reduces rework and improves first-time quality.

  1. Fragmentation of systems and data environments

Participants highlighted ongoing challenges:

  • multiple disconnected systems (CDEs, on-premise tools, CAFM platforms)
  • inconsistent data structures and naming conventions
  • lack of standardised outputs

This fragmentation limits the effectiveness of AI and automation.

  1. Importance of clear requirements and rule definition

A recurring theme was:

AI cannot compensate for unclear or undefined requirements

To enable automation, the industry must:

  • define what information is required
  • specify what “good” looks like
  • establish rules for validation (e.g. what must be present in a fire strategy drawing)

Without this, automation becomes unreliable.

  1. Development of rule-based validation frameworks

The group is creating structured checklists and rules, including:

  • validation of Regulation 38 fire safety information
  • expected content of drawings (e.g. escape routes, fire doors, key data)
  • completeness checks for O&M documentation

These rules form the basis for:

  • automated compliance checking
  • consistent, repeatable validation processes
  1. Data quality as a prerequisite for AI

A strong caution was raised:

  • AI effectiveness depends on high-quality, validated data
  • poor data leads to unreliable outputs

Therefore:

  • information must be verified before entering AI systems
  • “trusted data” should be clearly identified and controlled
  1. Collaboration and domain expertise

The session reinforced the importance of:

  • bringing together multi-disciplinary expertise (e.g. fire, MEP, asset management)
  • combining domain knowledge with technical capability

This enables the group to define:

  • robust standards
  • realistic validation rules
  • practical, industry-ready solutions

Overall Conclusion

The session marked a significant step toward creating a rule-based, AI-enabled validation ecosystem for asset information.

The emerging approach:

  • defines structured, task-driven information requirements
  • translates these into schemas and validation rules
  • uses AI to automate checking and identify gaps
  • shifts validation earlier in the lifecycle

Ultimately, the success of this model depends on clear definition of requirements, high-quality data, and collaborative expertise, enabling the Golden Thread to move from fragmented documentation to validated, usable and trustworthy information at scale.