Introduction:
Over the past few months, I have been actively working on building and stabilizing a Power BI ecosystem in one of our client's environment. One of the key deliverables in this journey was designing and implementing the Migration Service Dashboard, which brought significant visibility and value to stakeholders.
When I first started building Power BI dashboards in the client's environment, I assumed the hard part would be the data modelling. It turned out the data was the easy part. The real work — the frustrating, hair-pulling, occasionally exhilarating work — was everything around it: the gateway, the version control decisions, the publish pipelines, the app vs. workspace confusion, and eventually unlocking AI Copilot inside our dashboards.
This blog is a candid account of the challenges I faced, the mistakes I made, and the patterns I now follow. If you're building with Power BI at scale — especially connecting to on-premises Oracle databases and shipping dashboards to business users — I hope this saves you some hours.
The Big Picture — Architecture at a Glance
Before diving into specifics, here's the end-to-end architecture I settled on after several iterations. Everything flows from Oracle through the gateway, gets versioned in Azure DevOps, lands in a Power BI Workspace, and finally reaches end users via a Power BI App — with AI Copilot woven into the service layer.
Connecting Power BI to Oracle — Gateway Battles I Didn't Expect
I'll be honest: I underestimated the On-Premises Data Gateway. I thought it would be a 20-minute setup. It took me almost two full days to get a stable, performant connection between Power BI Service and our Oracle database — and I learned more about network topology in those two days than I had in months.
Our Oracle DB sat behind a corporate firewall on-premises. Power BI Desktop connected fine locally — but the moment I published to Power BI Service and tried to schedule a refresh, it timed out. Every. Single. Time.
What's actually happening under the hood
The On-Premises Gateway acts as a relay — Power BI Service in the cloud sends requests to the gateway software installed on a machine within your network, and that gateway reaches the Oracle DB directly. The gateway machine must be always-on, network-reachable, and have the right Oracle client drivers (ODP.NET or ODBC) installed.
The traps I fell into
My first mistake was installing the gateway on my local laptop. It worked until I closed the lid. The fix: dedicate a server VM for the gateway. My second mistake was missing the Oracle 64-bit client requirement — Power BI's gateway is 64-bit only, and I had a 32-bit Oracle client installed. The error messages were cryptic and sent me in the wrong direction for hours.
The third issue was credential mismatch. Power BI Service stores credentials for the data source centrally, and they must exactly match what the gateway expects. Switching from Windows auth to basic auth on the Oracle side required updating credentials in three places.
Dedicated gateway VM (Windows Server 2019), 64-bit Oracle ODP.NET driver, static credentials configured in Power BI Service data source settings, and gateway clustering for redundancy.
Enable gateway logs and watch them in real time during your first refresh. The log entries are far more informative than the UI error messages in Power BI Service.
Scheduled refresh has been rock-solid since.
Power BI Desktop vs. Power BI Web Portal — They Are Not the Same Tool
This confused me early on, and I've seen it confuse colleagues too. Power BI Desktop and the Power BI Web Portal (Power BI Service) serve very different purposes, and conflating them leads to workflow headaches.
I tried to build complex calculated columns directly in the Service's web editor, not realising how limited it is. Always build the model in Desktop, then publish. The Service is for distribution and governance, not authoring.
"Desktop is your studio. Service is your gallery. You create in the studio; people visit the gallery"
Why GitHub Didn't Work — and Why Azure DevOps Did
When I first thought about version control for our .pbix files, my instinct was GitHub. It's what our engineering teams use. It seemed natural. It didn't work.
Power BI's native Git integration in the Service is built exclusively for Azure DevOps Git repositories. There is no native GitHub connector for the Power BI Service workspace sync feature. I spent time trying to configure OAuth tokens and custom webhooks before I finally accepted that this just wasn't a supported path.
Why Azure DevOps Git is the right answer
Power BI Service has a built-in workspace-level Git integration that talks directly to Azure DevOps. You connect your workspace to a branch in an Azure DevOps Git repo, and changes made in the Service sync there. The .pbix is stored as a folder structure (the Power BI Project format, .pbip), which makes diffs actually meaningful in pull requests.
This gives us a full audit trail of every dashboard change — who changed what, when, and why. For compliance and governance, this alone justified the Azure DevOps setup over ad-hoc file sharing.
The critical behaviour to know: you must manually publish app updates
This one tripped up my whole team. When you update a report in the workspace — even if the dataset refreshes automatically — the Power BI App does not update automatically. You have to go into the App settings and click "Update app" for the changes to reach consumers. I missed this on our first production update and spent 30 minutes wondering why stakeholders were still seeing old data in the App while the workspace showed the right figures.
Every deployment checklist ends with: Update the App It sounds simple, but making it explicit in the runbook has eliminated that mistake entirely.
Integrating AI Copilot into Power BI Dashboards
This is the piece that genuinely changed how our leadership team interacts with data. Enabling AI Copilot inside Power BI turned our static dashboards into interactive conversations about the numbers.
What Copilot in Power BI actually does?
Power BI Copilot allows report consumers to ask natural-language questions against the underlying dataset — "What drove the spike in costs in Q3?" or "What are the key customer requirements and constraints for migrating from Windows Server 2016 to a newer platform?" — and Copilot generates a visual or a summary on the fly. It can also auto-generate entire report pages from a dataset description, which I used to prototype executive summary layouts in minutes rather than days.
What I had to set up
Copilot in Power BI requires a Fabric-enabled capacity (F64 or higher, or a Power BI Premium P1 equivalent). It also needs to be enabled at the tenant level by the Power BI admin in the Admin Portal, and enabled at the workspace level. If any of those three layers is off, Copilot simply doesn't appear — no helpful error, just absence.
AI Copilot shifts Power BI from a reporting tool to a decision-support system. When leaders can interrogate data in natural language during a meeting, data literacy barriers dissolve and data-driven culture accelerates.
Final Thought
I'm sharing this because I know others on the team are navigating similar terrain. If you're hitting walls with the gateway, struggling with deployment pipelines, or curious about how to unlock Copilot for your reports — reach out. These challenges are solvable, and the patterns are worth sharing.
Raj Kanango
Professional Services Delivery- Global Competency Centre
HPE Operations – Services Experience Delivery
I work at HPE
HPE Support Center offers support for your HPE services and products when and how you need it. Get started with HPE Support Center today.
[Any personal opinions expressed are mine, and not official statements on behalf of Hewlett Packard Enterprise]
Very Well Written!
[Any personal opinions expressed are mine, and not official statements on behalf of Hewlett Packard Enterprise]
Please follow the official documentation.
Power BI @ HPE - Documents - All About Gateway - All Documents
I work at HPE
HPE Support Center offers support for your HPE services and products when and how you need it. Get started with HPE Support Center today.
[Any personal opinions expressed are mine, and not official statements on behalf of Hewlett Packard Enterprise]