Power BI earned its place as one of the most widely deployed analytics tools in the enterprise, and for many teams it remains a solid choice. But as organizations tighten data-residency requirements, rethink per-user licensing, and look for AI that queries data in plain language, some BI leads are evaluating a move to a self-hosted platform. AI Insights 365 On-Prem runs entirely inside your own infrastructure on Windows, which changes the calculus on where your data lives and how your team interacts with it.
This guide is deliberately balanced. Power BI is a capable product, and a migration is a real project — not a weekend swap. What follows are the honest reasons teams make the move, what carries over, how core concepts map across the two platforms, and a practical plan to execute the transition.
Why Teams Consider the Move
Three motivations come up most often.
Cost per user
Per-seat licensing scales linearly with headcount. As viewer counts grow across departments, the cost of giving everyone access can climb faster than the value any single viewer extracts. A self-hosted model shifts the cost structure toward infrastructure you already operate, which can be attractive once an organization has many occasional viewers rather than a few power authors.
Data residency and control
Because AI Insights 365 is self-hosted on Windows and runs inside your environment, your data and credentials stay within your security boundary. For organizations in regulated industries — or those operating under contractual or sovereignty requirements about where data may be processed — keeping analytics fully on-premises removes an entire class of questions about where datasets are cached, processed, or stored.
The deployment model reinforces this: because the platform runs inside your network, it connects directly to your in-network SQL Server and PostgreSQL databases. Where a database sits in a remote subnet or another site, you reach it over private networking or a VPN rather than the public internet, and you point the platform at a least-privilege, read-only account. Credentials for your most sensitive systems never leave the network, and source data is never exposed beyond your boundary.
AI-native querying
This is the capability that most often tips an evaluation. AI Insights 365 is built around natural-language-to-SQL: business users ask questions in plain English and get governed answers, while the platform can generate AI auto-reports automatically. The AI layer runs on Azure AI Foundry or, for fully air-gapped deployments, a local Ollama model — so even the intelligence layer can stay inside your walls.
What Carries Over
A migration is far less daunting when you do not have to rebuild everything. Two interop points matter most.
- Connect existing Power BI semantic models. You do not have to abandon the modeling work already invested in Power BI. AI Insights 365 can connect directly to your existing Power BI semantic models, letting you layer natural-language querying and AI reports on top of governed models you already trust.
- Power BI export interop. The platform supports Power BI export, so the transition does not have to be all-or-nothing. Teams can continue to share outputs into Power BI workflows during a phased migration, which keeps stakeholders productive while you move authoring over.
Together these mean you can adopt AI Insights 365 incrementally — connecting to what exists, exporting where needed, and rebuilding only where there is a clear benefit.
Mapping Power BI Concepts to AI Insights 365
Migrations go smoothly when the team shares a mental model of how familiar concepts translate. The mapping below is the one to internalize.
| Power BI concept | AI Insights 365 equivalent | Notes |
|---|---|---|
| Dataset / data source | Datasource | Created via the 24-connector catalog; read-only, sandboxed, AES-256 encrypted credentials |
| Semantic model | Connected Power BI semantic model (or native datasource) | Reuse existing models directly, or model natively |
| Report | Report | AI auto-reports plus authored reports |
| Dashboard | Dashboard | Built from 72+ chart types |
| Row-level security (RLS) | RBAC (Admin / Editor / Viewer) + workspaces | Governance shifts from row filters to role- and workspace-scoped access |
| Workspace | Workspace | Same organizing concept for grouping content and access |
| Scheduled subscriptions | Scheduled email reports | Native scheduling to email recipients |
| Azure AD / Entra ID | AD + Entra ID SSO | Sign-in maps directly |
Datasets become datasources
In Power BI you connect to data and often build a semantic model on top. In AI Insights 365 you create a datasource from the connector catalog (24 connectors spanning relational databases, cloud warehouses, files, SaaS/APIs, and Power BI). Every datasource validates its parameters, introspects schema, and runs read-only queries with row limits — the safety model is consistent across all sources.
Reports and dashboards
Authored reports and dashboards map closely to their Power BI counterparts. The notable addition is AI auto-reports and a library of 72+ charts, so much of the visual rebuild can be accelerated rather than recreated pixel by pixel.
RLS maps to RBAC and workspaces
This is the most important conceptual shift. Power BI's row-level security filters data within a dataset by user or role. AI Insights 365 governs access through RBAC — Admin, Editor, and Viewer roles — scoped by workspaces, with all activity captured in audit logs. Plan this mapping carefully: rather than translating every RLS rule one-for-one, model your access boundaries as workspaces and roles, then handle finer-grained data scoping at the datasource or query layer.
A Practical Migration Plan
A phased plan keeps the lights on while you transition. The following five phases work well for most teams.
Phase 1 — Inventory and prioritize
Catalog what you have in Power BI today:
- Data sources and the semantic models built on them
- Reports and dashboards, ranked by usage and business value
- RLS rules and the roles they serve
- Subscriptions and scheduled deliveries
Rank by value and usage. High-value, high-traffic content migrates first; long-tail reports can wait or be retired.
Phase 2 — Stand up the platform
Install AI Insights 365 on Windows inside your environment. Configure the essentials:
- SSO via AD and Entra ID so sign-in matches your existing identity
- Workspaces that reflect your team and content structure
- RBAC roles aligned to who authors versus who only views
- Direct, in-network connections to any SQL Server or PostgreSQL sources, reaching remote subnets over private networking or VPN and using least-privilege read-only accounts
- The AI backend — Azure AI Foundry or air-gapped Ollama — per your data-residency posture
Phase 3 — Connect data
Recreate your connections using the connector catalog. Where you have invested in Power BI semantic models, connect to them directly rather than rebuilding. For relational systems behind the firewall, connect directly in-network — over private networking or VPN if the database lives in a remote subnet — using a least-privilege, read-only account so credentials never leave the network. Validate each connection and introspect its schema before moving on.
Phase 4 — Rebuild and validate priority content
Recreate the highest-value reports and dashboards first. Lean on AI auto-reports to accelerate first drafts, then refine with the chart library and no-code transforms. Validate numbers against the Power BI originals — matching totals on a few key measures builds confidence quickly. Configure scheduled email reports to replace existing subscriptions.
During this phase, Power BI export is your safety net: stakeholders can keep receiving familiar outputs while authoring shifts over.
Phase 5 — Cut over and govern
Once priority content is validated and adopted, move remaining users over, retire redundant Power BI subscriptions on your own timeline, and lean on audit logs and backup/restore for ongoing governance. Keep a documented rollback path until the team is fully settled.
What's Better, and What's Different
It is worth being clear-eyed about both.
What's genuinely better for the right team:
- AI-native interaction. Natural-language-to-SQL and AI auto-reports lower the barrier for occasional users who never learned to author in a traditional BI tool.
- Full data residency. Self-hosted on Windows, connecting directly to your in-network databases and with an optional air-gapped Ollama model, means data and credentials can stay entirely within your boundary.
- Cost structure. Shifting away from strict per-seat licensing can favor organizations with large viewer populations.
- Breadth of connectors. 24 connectors with a uniform read-only, sandboxed, AES-256-encrypted model.
What's different and should be planned for:
- You operate the platform. Self-hosting means you own installation, updates, and the host on Windows. That is the trade for control.
- Governance is modeled differently. RLS thinking has to be re-expressed as RBAC plus workspaces.
- Rebuild effort is real. Semantic-model reuse and Power BI export soften it, but visuals and scheduled deliveries still need recreating and validating.
Common Migration Questions
A few questions come up in nearly every migration planning session. Addressing them early keeps the project on track.
Do we have to rebuild every report at once?
No. The combination of connecting to existing Power BI semantic models and Power BI export interop is specifically what makes a phased approach viable. You migrate the highest-value content first, keep exporting to Power BI for stakeholders who are not yet moved over, and retire the long tail on your own schedule. Treat the migration as a rolling program, not a single cutover date.
How do we keep the numbers trustworthy during the transition?
Validation is the discipline that earns trust. For each rebuilt report, pick a handful of headline measures — totals, counts, period-over-period figures — and reconcile them against the Power BI original before you publish. When connecting directly to an existing semantic model, the underlying calculations come along with the model, which removes a major source of drift. Where you model natively against a datasource, lean on schema introspection so the AI generates queries against the correct columns.
What happens to our scheduled subscriptions?
Power BI subscriptions map to scheduled email reports. As you validate each migrated report, recreate its delivery schedule and recipient list, then disable the corresponding Power BI subscription only once the new delivery has run successfully at least once. Running both briefly in parallel is a cheap insurance policy.
How does sign-in change for end users?
For most users, it should not change noticeably. Because AI Insights 365 supports AD and Entra ID SSO, people sign in with the same corporate identity they already use. The visible change is the destination, not the credentials.
Where does the AI processing happen?
That is your choice and a key part of the data-residency story. Azure AI Foundry keeps processing within your Azure tenant, while a local Ollama model keeps it fully air-gapped inside your network. Decide this during Phase 2 so the rest of the rollout aligns with the posture you have committed to.
A Note on Governance After Cutover
Migrating the content is only half the job; sustaining it is the other half. Once you are live, three platform capabilities carry the ongoing load. RBAC and workspaces keep access aligned to roles as teams change. Audit logs give you a defensible record of who accessed and changed what — valuable both for security reviews and for diagnosing unexpected report changes. Backup and restore protects your configuration and content, and should be wired into your existing operational runbooks from day one rather than bolted on later. Establishing these habits early means the platform stays as governed a year from now as it is on launch day.
The Bottom Line
Power BI remains a strong product, and not every team should migrate. But if data residency is becoming a hard requirement, if per-user costs are outgrowing the value, or if you want analytics your business users can simply ask questions of, AI Insights 365 On-Prem is built for exactly that scenario. The path is incremental: connect to the semantic models you already have, export into Power BI while you transition, and rebuild your highest-value content with AI assistance. Done in phases, the migration is a controlled project — not a leap of faith — and it ends with your data, your credentials, and your intelligence layer all running inside your own walls.