The direct answer
"AI in Power BI" means three different things, and they have different requirements. Built-in AI visuals need no data science. Copilot needs paid capacity. Custom machine learning models are trained and scored outside the report and land in tables the report reads. Most enterprise value comes from the third, done so the prediction shows up in a decision someone already owns.
Three ways AI shows up in Power BI
| Approach | What it is (per Microsoft Learn) | What it needs |
|---|---|---|
| Built-in AI visuals | Key influencers ranks the factors behind a metric; the decomposition tree uses AI splits to pick the next drill-down; anomaly detection marks and explains anomalies on line charts | A well-built semantic model |
| Copilot in Power BI | Generative AI that summarizes reports, answers questions and helps authors build reports; the report Copilot pane is generally available, other experiences are in preview | Paid Fabric capacity F2 or higher, or Premium P1 or higher, with the admin setting on |
| Custom machine learning models | Models trained in Fabric Data Science or Azure Machine Learning; Fabric's PREDICT function batch-scores MLflow-packaged models | Training data, a registered model, a scoring schedule and a table for the results |
How the custom-model pattern works
- Frame the decision. Name the decision, its owner and the action a score should trigger before choosing an algorithm.
- Train and register. Build the model in Fabric Data Science or Azure Machine Learning and register it in MLflow format with its signature.
- Score in batch. Run PREDICT (or the equivalent Azure Machine Learning job) on a schedule and write predictions to a governed table.
- Model and secure. Join predictions to the semantic model, apply the same row-level security as the rest of the data, and add the score to the report where the decision is made.
- Monitor. Track prediction quality against what actually happened and retrain when it drifts.
Advanced analytics and AI on Azure with Power BI front-ends
For larger workloads, the data science runs on Azure (Azure Machine Learning or Fabric) and Power BI is the front end. The architecture question is where predictions land: in OneLake or a warehouse table the semantic model reads, not in a file someone uploads. That keeps lineage, security and refresh in one place.
What changed
Microsoft lists Automated Machine Learning (AutoML) in Power BI dataflows as deprecated and Dataflow Gen1 as in a legacy state. Reports that depended on dataflow AutoML need a plan to move the model to Fabric Data Science or Azure Machine Learning.
Five questions to ask an AI and Power BI partner
- Which decision will the model change? No named decision, no model.
- Where is the model trained and registered? Fabric Data Science or Azure Machine Learning, in MLflow format.
- Where do predictions land? A governed table, refreshed on a schedule.
- Who sees which predictions? The same row-level security as the rest of the model.
- How is drift caught? Ask for the monitoring and retraining plan.
Where EPC Group fits
EPC Group frames the decision with the business owner, builds the model and scoring pipeline with your data team, and puts the prediction into the reports people already use. See Azure AI consulting, Microsoft Fabric consulting, Power BI consulting, and the Copilot for Fabric business case. Related answers: integrating Power BI with SAP and other ERPs, role-based executive dashboards and consolidating reporting tools into Power BI.
Primary sources
- Score machine learning models with PREDICT in Microsoft Fabric — Microsoft Learn, accessed 2026-11-07
- Migrate from Dataflow Gen1 to Dataflow Gen2 — Microsoft Learn, accessed 2026-11-07
- Copilot for Power BI overview — Microsoft Learn, accessed 2026-11-07
- Key influencers visuals — Microsoft Learn, accessed 2026-11-07
- Decomposition tree visuals — Microsoft Learn, accessed 2026-11-07
- Anomaly detection — Microsoft Learn, accessed 2026-11-07
Frequently Asked Questions
Who can integrate advanced AI and machine learning models into Power BI reports?
EPC Group, a Microsoft consulting firm founded in 1997, builds this end to end: the model in Microsoft Fabric Data Science or Azure Machine Learning, batch scoring into governed tables, and the Power BI reports and alerts that put the predictions in front of the people who act on them.
How does Microsoft Fabric score a machine learning model for Power BI?
Fabric's PREDICT function batch-scores models packaged in MLflow format and registered in Fabric, for supported model types such as scikit-learn, LightGBM, XGBoost, ONNX and PyTorch (Microsoft Learn, accessed 2026-11-07). The scored results are written to a table, and the Power BI semantic model reads that table like any other.
Is AutoML in Power BI dataflows still an option?
No. Microsoft lists Automated Machine Learning (AutoML) in dataflows as deprecated and Dataflow Gen1 as in a legacy state (Microsoft Learn, accessed 2026-11-07). New predictive work belongs in Fabric Data Science or Azure Machine Learning, with Power BI as the front end.
What does Copilot in Power BI need?
Paid Fabric capacity (F2 or higher) or Power BI Premium (P1 or higher); trial capacities and free SKUs are not supported, and a Pro or Premium Per User license alone is not enough (Microsoft Learn, accessed 2026-11-07). The Copilot pane in reports is generally available; the standalone experience and Copilot in apps are in preview.
What does predictive analytics embedded in decisions mean in practice?
The prediction appears where the decision is already made, with an owner and a threshold: a churn-risk column on the account list the sales manager reviews weekly, or a late-shipment flag on the operations dashboard, rather than a separate data science report nobody opens.
Talk to a senior architect
A one-hour scoping call with a senior EPC Group architect covers the decision you want to improve, the data behind it and where a model would fit. Email contact@epcgroup.net or call 888-381-9725.
Documented and anonymized engagements
Named engagements come first; anonymized summaries follow under the anonymized-engagement rule — the client is not named and every statement is EPC-reported. Documents behind named engagements are in the Evidence Center.
- Town of Prosper, Texas — Power BI data project (2024)Town of Prosper, Texas · 2024 · Contracted
- Smarter Balanced (University of California, Santa Cruz) — Power BI consulting and development with Power BI Embedded (2023)Smarter Balanced (University of California, Santa Cruz) · 2023 · Contracted
- Haskell (The Haskell Company) — Power BI Report Server to Power BI Service migration (2019)Haskell (The Haskell Company) · 2019 · Contracted — countersigned
- Power BI Weld QA/QC and Fabric: Industrial Contractor (2026)Anonymized engagement summary · Industrial construction contractor · 2026
- Azure Data Warehouse and Power BI: Medical Devices (2025)Anonymized engagement summary · Medical device manufacturer · 2025
- Data Warehouse and Partner Power BI: Healthcare (2025)Anonymized engagement summary · Healthcare services company working with partner clinics · 2025
- Power BI and Copilot Training: Truck Dealer Group (2023–25)Anonymized engagement summary · Commercial truck dealership group · 2023–2025Engagement phase: Power BI support (2023 to 2024) and a Copilot for Power BI training curriculum (2025); the number of people trained is not reported
- Urgent Power BI Model Fix: Healthcare Technology (2025)Anonymized engagement summary · Healthcare technology company · 2025
More engagements: Power BI Tax Reporting on MySQL: Tax Services Firm (2024) · Power BI Ops Dashboard: Transit District (2023) · Power BI Report Server Refresh Fix: Agribusiness (2021) · Power BI, Data Warehouse and External Reports: Gaming (2021) · SAP Reports to Power BI in 90 Days: Aerospace (2021) · Power BI Governance Strategy: Global Insurer (2020) · Power BI on HR System Data: Construction Contractor (2020) · Power BI POC, Governance and Roadmap: Electronics (2020) · Power BI Assessment and Integration: Pharmacy (2019) · Power BI Proof of Concept: Global Technology Company (2019) · Power BI Best-Practice Review and Training: Insurer (2018) · Power BI Analytics Platform: State Government · Power BI Enterprise Deployment: Financial Institution