The bottom line
Microsoft Fabric in production needs proactive managed services — F-SKU capacity monitoring, semantic-model drift detection, Power BI usage analytics, workspace governance. The retainer model where one senior practitioner owns the estate works better than break-fix tickets.
In This Article
Introduction
Your Fabric rollout went live. The pipelines are pulling from SAP S/4HANA. The Power BI dashboards are published. The project closed on budget. And then, three months later, the morning ops review is running slow, two ADF pipelines failed overnight without an alert, and someone in finance is working off a semantic model that refreshed four hours late.
This is not a platform failure. It's what happens when a data estate is treated like a project — built, handed over, and left to run itself.
The production gap nobody budgets for
Most mid-market industrials allocate serious budget for the Discover and Prototype phases of a Microsoft Fabric deployment. Architecture design, data engineering, semantic modelling, Power BI development — the visible work gets funded.
What gets underestimated is the operational overhead that begins the day the estate goes live. Microsoft Fabric, OneLake, Azure Data Factory, and Direct Lake are not set-and-forget. They're living infrastructure. Capacity drifts. Storage costs compound. Refresh schedules conflict with each other during peak load windows. A schema change in SAP S/4HANA silently breaks a downstream lakehouse table, and nobody notices until the plant manager asks why OEE is showing null for three production lines.
The gap between "deployed" and "stable" is where most Fabric estates bleed.
What actually gets watched in a managed retainer
A fractional engineering retainer for a Fabric estate is not a helpdesk. There are no ticket queues, no SLA timers counting down to a junior analyst acknowledging your email. It's closer to embedded site reliability engineering — a small, senior team watching the estate so your internal IT doesn't have to.
Here is what actually gets monitored:
**ADF pipeline health.** Azure Data Factory runs the ingestion layer — pulling from SAP S/4HANA, SAP ByDesign, or Microsoft Dynamics 365, depending on your stack. Pipelines fail silently more often than they should. A retainer establishes alerting rules and a weekly pipeline audit so failures surface before they corrupt downstream models.
**OneLake storage costs.** OneLake charges on consumption. Without governance, storage costs grow 15–30% quarter-on-quarter as stale tables accumulate and Delta Lake compaction gets skipped. A retainer includes a monthly storage review with explicit housekeeping actions.
**Power BI capacity utilisation.** If your organisation is on F-SKU or P-SKU capacity, utilisation patterns matter. Direct Lake queries are fast — until they're not. Capacity saturation during the morning report peak (typically 07:30–09:00 local time, exactly when your operations team needs the data) causes query throttling that looks like a broken dashboard. Monitoring capacity burn and scheduling semantic model refreshes around peak usage is an engineering task, not a user task.
**Semantic model refresh latency.** A model that was refreshing in 8 minutes at go-live might be taking 35 minutes six months later because upstream tables have grown and partition strategies haven't been reviewed. Left unchecked, this creates a lag between the plant event and the dashboard that defeats the purpose of the investment.
**Direct Lake misconfiguration.** Direct Lake mode in Power BI is powerful — but it breaks in specific, non-obvious ways when Delta tables aren't correctly formatted or when fallback to import mode triggers without warning. Monitoring for fallback events is a non-trivial operational task that most IT teams don't have visibility on.
What this is not
This is not 24/7 incident response. If your ERP goes down at 2 a.m., that is an ERP support issue — not a data estate issue. A fractional retainer does not replace your Microsoft support agreement, your internal IT team, or your SAP basis team. It sits between the platform and the business, watching the data layer specifically.
It also does not cover new feature development. If the Operations Director wants a new OTIF breakdown by carrier lane added to the Supply Chain dashboard, that is a scoped change — budgeted and delivered separately from the retainer. The retainer keeps the existing estate healthy. Development is a distinct motion.
What this looks like in practice
A packaging manufacturer running Microsoft Fabric with three source systems — SAP ByD, a MES, and a WMS — had 14 ADF pipelines in production at go-live. Within six months, four pipelines were failing intermittently due to schema drift from a SAP ByD update, OneLake storage had grown 40% beyond the initial projection because failed pipeline runs were writing partial tables, and two Direct Lake semantic models had silently fallen back to import mode, adding 25 minutes to the morning refresh cycle.
A three-month retainer engagement identified and resolved all of these. The ongoing scope settled at: weekly pipeline health review, monthly OneLake storage audit, quarterly capacity optimisation review, and a named contact for ad-hoc escalations during business hours. Total retainer effort: approximately 12–16 hours per month.
That is the shape of the engagement. Not a war room — a watchful eye.
Where this approach doesn't fit
If you are still in the build phase — pipelines not yet in production, semantic models not yet published — a retainer is premature. The right motion is a structured Prototype engagement, not ongoing operations support.
If your Fabric estate has fewer than five active pipelines and a single Power BI workspace with under 10 reports, the overhead of a formal retainer may not be warranted. A quarterly health check is likely sufficient until the estate grows.
If your organisation has an internal data engineering team with Fabric expertise and genuine capacity to watch the estate, an external retainer duplicates effort. The retainer model is designed for the common mid-market situation: a capable IT team that built and deployed the estate but does not have bandwidth to operate it.
Six weeks to first value
A retainer engagement starts with a two-week Discover phase — a structured audit of the existing estate covering pipeline health, OneLake storage patterns, capacity utilisation logs, and semantic model refresh history. By week four, a prioritised list of operational risks is in front of the CIO with remediation effort estimates. By week six, the highest-risk items are resolved and baseline monitoring is in place.
That is the Prototype shape. What follows is a monthly operational rhythm, not a project with a close date.
Production Fabric is a living estate. The capacity bill grows with usage; the semantic models drift with team rotations; the workspace permissions sprawl. Managed services keeps the estate clean. The alternative is the 11pm Slack message about a throttled report and no one available to fix it.
Free Assessment
Where does your operation sit on the data maturity curve?
8 questions. 3 minutes. You get a scored breakdown across data infrastructure, analytics readiness, and automation potential — with a specific next step for your industry.