AI Data Readiness for Construction: A Microsoft Fabric Guide
Updated: 3 hours ago
Most construction leaders can picture the AI use case: ask "Which jobs are fading margin this month, and why?" and get a clear, trustworthy answer in seconds. The tools to do this now exist. Whether the answer is right depends almost entirely on the data underneath.
This guide explains what AI data readiness means for a construction business, why construction data is often not ready, and how Microsoft Fabric and Power BI help you get there, with a practical checklist to start from.
What does AI data readiness mean?
AI data readiness is the work of making your data accurate, structured, clearly defined and properly secured, so AI tools can answer questions about it reliably. In practice, AI-ready data has five things:
Trusted sources. Data is extracted, cleaned and reconciled, not copied from spreadsheets.
Clear business definitions. Margin, WIP, committed cost and over/under billings mean the same thing everywhere.
A well-structured model. Tables and measures are organised and named so a person, or an AI, can understand them.
Access controls. People, and the AI acting for them, only see the data they are allowed to see.
Tested questions. The questions people will actually ask have been tried and checked against known answers.
Why construction data is often not AI-ready
Key numbers live in spreadsheets that are rebuilt every month.
Job numbers, cost codes and customer names differ between the ERP, project management and finance systems.
Definitions of WIP or margin live in people's heads, not in the data.
ERP tables have cryptic names and codes that mean nothing to an AI without context.
There is little history, so trends and "what changed?" questions cannot be answered.
Point an AI tool at that and it will still produce confident-sounding answers. That is the risk.
How Microsoft Fabric supports AI data readiness
1. One governed foundation in OneLake
Fabric brings data engineering, storage and Power BI together on one platform. Using a medallion architecture (bronze, silver and gold layers), raw data from Vista and other systems is cleaned, conformed and modelled once, giving AI a single, governed source rather than scattered extracts.
2. Semantic models that speak your language
AI tools that query Power BI rely heavily on the semantic model's metadata. Microsoft's guidance for data agents recommends a star schema, explicit measures and business-friendly names, giving the example of "Total Revenue" rather than "TR_AMT". For a contractor, that means names like Projected Margin % and Over/Under Billings, with descriptions explaining how each is calculated.
3. Prep data for AI in Power BI
Power BI's Prep data for AI features (in preview at the time of writing) let model authors improve Copilot's answers. They are saved on the semantic model rather than individual reports:
AI data schema: choose which tables, columns and measures Copilot should focus on.
Verified answers: link common questions to a specific, approved visual.
AI instructions: explain your terminology and preferences, such as which date to use for WIP.
Microsoft is clear that these features cannot guarantee a specific output every time, because AI behaviour is nondeterministic. Testing still matters.
4. Fabric data agents
A Fabric data agent lets people ask plain-English questions of data in lakehouses, warehouses and Power BI semantic models. According to Microsoft, it generates SQL, DAX or KQL queries, is read-only, and runs with the requesting user's own permissions, so row-level security still applies. Agents can also be surfaced in Microsoft 365 Copilot, Copilot Studio and Teams.
5. Security and governance
Row-level security, workspace permissions and Microsoft Purview policies apply to AI access as well as to reports. Decide who can see which companies, jobs and payroll data before AI is switched on, not after.
A practical AI data readiness checklist
List 15 to 20 real questions leaders want to ask, such as "Which jobs have the biggest under billings?"
Identify where the data for each question lives today.
Land and clean that data in a governed Fabric layer.
Build a focused star-schema semantic model with business-friendly names and descriptions.
Agree and build explicit measures for margin, WIP, committed cost and cash.
Configure the AI data schema, verified answers and AI instructions.
Apply row-level security and check what each role can see.
Test every question, review the generated queries and compare answers with known results.
Keep a human review step for decisions that matter, and iterate as questions change.
For help with any of these steps, see our AI-ready data service or our Microsoft Fabric consulting.
Common mistakes
Pointing AI at raw ERP tables. Without a curated model and definitions, answers will be inconsistent.
Exposing everything. Large, unfocused models create noise; limit AI to the tables that answer real questions.
Skipping definitions. If finance and operations disagree on margin, so will the AI.
Treating AI answers as final. Check important numbers and keep people accountable for decisions.
No owner. Someone needs to own the model, the definitions and the test questions.
Key takeaways
AI data readiness means trusted, defined, well-modelled and secured data, tested against real questions.
Microsoft Fabric gives construction firms a governed foundation, and Power BI semantic models give AI the business context.
Prep data for AI and Fabric data agents help, but they work best on a clean model and still need testing.
Start with real questions, not with the technology.
Frequently asked questions
What is the difference between AI-ready data and normal reporting data?
Reporting data only needs to feed known reports. AI-ready data also needs clear names, descriptions, definitions and tested questions, because people can ask the AI anything.
Can an AI agent see data a user is not allowed to see?
Microsoft states that Fabric data agents use the requesting user's credentials and permissions, so row-level and column-level security still apply.
Will AI answers always be accurate?
No. Microsoft notes that AI behaviour is nondeterministic. Good data preparation improves accuracy, but answers to important questions should be checked.
What licence do Fabric data agents need?
Microsoft's documentation lists a paid Fabric capacity (F2 or above, or Power BI Premium P1 or above with Fabric enabled) as a prerequisite. Check Microsoft's current licensing before planning a rollout.
If Vista is your main source system, start with our guide to connecting Viewpoint Vista to Power BI and our Vista ERP analytics service.
Next step: check how ready your data is
Want to know where your business stands? Take our free 3-minute Construction Data Readiness Check. You will get a score out of 100 across reporting, data foundation, platform and AI readiness, plus practical next steps. Prefer to talk it through? Book a discovery call and we will look at your setup together.
Comments