Overview

Dashboards give at-a-glance health for AI services — request rate, failure rate, P95 latency, dependency health. In Azure you typically pin charts from Azure Monitor metrics, Log Analytics / App Insights queries, or workbook tiles onto a shared dashboard.

Exam tips

Dashboards are for ops visibility; workbooks are for interactive investigation
Pin from Metrics, Logs (render), Application Insights overview blades
Share dashboards with the team; use consistent time ranges
Track AI-relevant signals: error rate, P95 latency, dependency failures, queue depth (via metrics/KQL)

When to use dashboards vs workbooks

Dashboards	Workbooks
Always-on status for a service	Guided troubleshooting / deep analysis
Shared team wall / ops overview	Parameters, drill-downs, multi-step stories
Few key tiles	Many linked queries and visuals

Dashboard

An Azure dashboard is a customizable portal page that pins tiles from metrics, logs, resource blades, and markdown. Use it as the “front door” for service health.

Typical AI app tiles:

Tile	Source idea
Request rate	App Insights / metrics or KQL `summarize count() by bin(...)`
Failure rate	`countif(Success == false) / count()`
P95 latency	`percentile(DurationMs, 95)`
Top failing dependencies	`AppDependencies` summarize by `Target`
Exception count	`AppExceptions` count over time

Build from a Log Analytics query

Open Application Insights → Logs
Run a query ending with | render timechart (or barchart)
Pin to dashboard (Pin to dashboard)
Resize/arrange tiles; set auto-refresh if available

Kql

AppRequests
| where TimeGenerated > ago(24h)
| summarize
    Requests = count(),
    Failures = countif(Success == false)
    by bin(TimeGenerated, 1h)
| extend FailureRate = round(100.0 * Failures / Requests, 2)
| render timechart

Metrics vs logs tiles

Metrics explorer	Log-based tiles
Near-real-time platform metrics	Rich App Insights tables & custom dimensions
Simple charts, less flexible	Full KQL power
Good for CPU, request rate (standard metrics)	Good for AI custom attributes, joins, percentiles

Design tips

One dashboard per service or pipeline stage (ingest, embed, serve)
Put failures and latency above vanity traffic charts
Label tiles with the service.name / resource they cover
Avoid overcrowding — 6–10 clear tiles beat 30 noisy ones

Build dashboards for app telemetry