Zendesk
Microsoft Fabric
2026
7 min read
Zendesk to Microsoft Fabric Migration: Real-Time Support Analytics Case Study
Industry
SaaS / Customer Support
Data Source
Zendesk REST API
Region
Global
Platform
Microsoft Fabric · Power BI
Numlytics delivered a complete Zendesk to Microsoft Fabric migration for a growing global SaaS business - building a fully automated incremental Fabric pipeline from the Zendesk REST API through a Bronze-Silver-Gold Medallion Lakehouse on OneLake, and serving real-time support analytics to Power BI via DirectLake. Within 60 days of go-live: average resolution time down 40%, two critical escalation bottlenecks identified and fixed, and £180K+ in annual SLA penalty and churn costs structurally avoided.
The Challenge: Thousands of Tickets, Zero Unified Analytics
The client ran all customer support operations on Zendesk, a mature, well-configured deployment with thousands of tickets, structured SLA tiers, CSAT surveys, and agent routing logic. On paper, the data existed. In practice, it was completely inaccessible for analytics. Teams relied on static Zendesk reports, reactive manual exports, and custom API queries that took hours to produce and were outdated by the time anyone read them.
-
Fragmented support analytics: Tickets, agents, SLA metrics, and CSAT scores existed in separate Zendesk API endpoints with no unified Lakehouse model - no single view of end-to-end support performance existed across the organisation
-
Stale visibility during incidents: Without automated incremental pipelines, leadership saw ticket volume trends hours after the fact, making real-time response to SLA breaches impossible during high-severity incidents
-
Reactive SLA management: Breached deadlines were discovered retrospectively in weekly review meetings, not proactively via live monitoring, the window to intervene had already closed
-
Disconnected customer context: Support ticket data could not be joined with customer segments, product usage, or revenue data due to siloed API endpoints - meaning high-value customer escalations received no differential handling
-
High analyst overhead: Two analysts spent roughly 60% of their time writing custom API queries and manually stitching data exports - a direct drag on the team's capacity to produce actual insights
Business cost of inaction: Without an automated Zendesk data pipeline and real-time analytics layer, the client estimated £180K+ in annual losses from SLA penalty payouts, agent inefficiency caused by invisible bottlenecks, and preventable churn from delayed issue resolution - problems a Zendesk to Microsoft Fabric migration was built to eliminate structurally.
The Numlytics Solution:
Numlytics designed a complete Zendesk to Microsoft Fabric migration - starting with structured data profiling of the Zendesk API before any pipeline work, building an automated incremental Fabric pipeline with watermark logic, and delivering four real-time Power BI dashboards via DirectLake on the Gold Lakehouse tier.
-
Zendesk Data Landscape Audit and API Profiling
Numlytics began with a structured data audit - mapping all Zendesk API endpoints in scope (tickets, users, organisations, SLA policies, satisfaction ratings, ticket events, and custom fields) and running structured profiling before any pipeline work began. This step caught three critical data quality issues: tickets re-opened multiple times had ambiguous resolution timestamps; SLA thresholds had changed twice across the three-year history meaning older tickets were measured against different breach criteria; and custom fields added over time were missing entirely on early records. Profiling before pipeline build is non-negotiable on any Zendesk data migration — without it, the reporting numbers in Power BI would have been wrong from day one.
-
Microsoft Fabric Pipeline - Incremental Watermark Architecture
Numlytics built the automated Microsoft Fabric pipeline using four activity types: Copy Data (pulling from Zendesk REST API), Lookup (reading and writing the watermark timestamp), ForEach (looping across ticket batches), and Notebook (running PySpark transformation logic). The key architectural decision was keeping all transformation logic inside the Notebook activity within the ForEach loop - rather than a separate external layer - making the entire journey from Zendesk API to clean Delta table a single, auditable unit. The watermark approach reads the last successful updated_at timestamp and fetches only what changed since. Incremental runs completed in under 8 minutes versus 47 minutes for a full historical load, a gap that widens every week as ticket data grows. This is the architecture that makes a Zendesk Fabric pipeline maintainable at production scale.
-
OneLake Medallion Lakehouse - Bronze, Silver, Gold
The Fabric Lakehouse was structured in three tiers on OneLake, all in Delta Parquet format for ACID compliance and time-travel capability. Bronze holds raw Zendesk API output exactly as received. Silver applies the data quality rules identified during profiling: deduplication, resolution timestamp standardisation, SLA policy version mapping across the three-year history, and custom field backfill for early records. Gold contains reporting-ready aggregates for support analytics, Microsoft Fabric Power BI: ticket performance by agent and team, SLA compliance by priority tier, CSAT score trends by ticket type, and first-contact resolution rates.
-
DirectLake Power BI Dashboards - Four Stakeholder Views
Rather than import mode, which introduces minutes of lag and stale data between scheduled refresh cycles, Numlytics implemented Zendesk Power BI DirectLake connectivity. DirectLake reads directly from Delta table files on OneLake, so dashboards reflect the latest pipeline output the moment they open. Four role-separated dashboards were built: Agent (own tickets, SLA status, CSAT), Team Lead (team performance, volume trends, escalation flags), Executive (KPI summary, cost per ticket, resolution benchmarks), and SLA Operations (live breach alerting). Role-level security was configured once at the semantic model and applied across all four dashboards without per-report configuration.
-
Historical Backfill and Incremental Sync Go-Live
The full three-year historical backfill completed in 47 minutes - loading all tickets, events, and satisfaction data through the Medallion pipeline. Incremental syncs went live immediately after, running every 15 minutes using Zendesk's incremental export endpoints with watermark tracking. The pipeline ran unattended for 30 consecutive days before handover, logging zero failures. The support team had live SLA monitoring, agent dashboards, and CSAT trend data within the same business day as go-live.
The Results
40%
Resolution Time ↓
Average ticket resolution reduced within 60 days of Zendesk Fabric migration go-live
15 min
Pipeline Refresh
Incremental Zendesk Fabric pipeline syncs every 15 minutes via watermark tracking
£180K+
Cost Avoided
Annual SLA penalty payouts and churn costs avoided through real-time SLA monitoring
Real-time
SLA Monitoring
Proactive DirectLake Power BI alerts replaced retrospective weekly SLA review meetings
"We knew the data was in Zendesk. We just had no way to get to it in any form useful for decision-making. Numlytics built the pipeline, profiled three years of API data before touching a single dashboard, and had us live in the same week. The DirectLake dashboards are instant - no waiting for a refresh, no stale numbers. Within two months we identified two escalation bottlenecks we genuinely did not know existed. One of them had been costing us roughly £40K a quarter in SLA penalties. That finding alone justified the entire engagement."
— Head of Customer Operations, Global SaaS Company
Key Engineering Insights from This Zendesk Fabric Migration
Insight 01
Building Inside the Pipeline Was the Right Decision
The most important architectural decision was keeping transformation logic inside the Fabric pipeline - within a Notebook activity triggered by the ForEach loop, rather than a separate external layer. This kept the entire journey from Zendesk API to clean Delta table as a single, auditable unit. Every run produces a clear activity log showing Copy Data, Lookup, ForEach, and Notebook status in one place. No multiple tools, no multiple schedules, no multiple systems to monitor. One pipeline. One place to look. One place to fix.
Insight 02
Watermark Logic Is Not Optional at Scale
The client's previous process pulled the full ticket history on every export, same 100K+ records, every time. Watermark logic reads the last successful updated_at timestamp and fetches only what changed since. Incremental runs landed under 8 minutes; the full historical load took 47. That gap widens as data grows. Skipping watermark logic produces a Zendesk Fabric pipeline that gets slower and messier every week it runs; it is not optional at the production scale.
Insight 03
Zendesk API Data Needs Profiling Before Pipeline Build
The Zendesk interface looks clean. The API data is not. Tickets reopened multiple times had ambiguous resolution timestamps. SLA thresholds had changed twice across three years. Custom fields were missing entirely on early records. Structured profiling before any pipeline work caught all of it. Without that phase, the support analytics Microsoft Fabric dashboard numbers would have been wrong from day one, and wrong in ways that are difficult to detect retrospectively.
Insight 04
DirectLake Removed the Refresh Ceiling
Import mode copies data into the Power BI dataset on every scheduled refresh - minutes of lag, stale data between cycles, no path to near-real-time. Zendesk Power BI DirectLake reads directly from Delta table files on OneLake, so dashboards reflect the latest pipeline output instantly. This was the enabling change that made proactive SLA breach alerting possible, structurally impossible with import-mode datasets.
Insight 05
The Metric That Matters Is What Changed After
Pipeline reliability and dashboard accuracy are table stakes. The real measure of a successful Zendesk to Microsoft Fabric migration is what the organisation could do that it could not before. Within 60 days: two escalation bottlenecks identified and fixed, shift patterns adjusted on actual volume data, average resolution time down 40%. Support leadership had been making staffing decisions on instinct - the data existed in Zendesk but had no path to analysis until Numlytics built one.
Technology Stack
Frequently Asked Questions
Zendesk · Microsoft Fabric · Support Analytics
Ready to migrate your Zendesk data to Microsoft Fabric?
Numlytics builds automated Zendesk to Microsoft Fabric pipelines - incremental watermark sync, Bronze-Silver-Gold Medallion Lakehouse, and DirectLake Power BI dashboards that give your support team live visibility.