Operations Insight Dashboard
Metrics and filters as the primary source of truth, with an AI narrative whose every observation links back to the data.
The problem
Operations managers need faster insight into workflow performance and exceptions. Charts alone require interpretation, while an unsupported AI summary is not trustworthy — so the prototype must connect interpretation back to the data.
Target user
Operations manager, with team leads and business stakeholders as secondary viewers. Key task: identify where work is delayed and what to investigate.
Product goal
Combine metrics, filters, drill-down, and evidence-linked insight so a manager can move from a performance signal to a specific investigation.
User flow
- 1Select a time range, region, and workflow.
- 2Review service volume, completion time, exception rate, and backlog metric cards.
- 3Read the completion-time and exception-rate trend charts.
- 4Drill into the sorted exception table.
- 5Generate an AI narrative summary of observed changes.
- 6Follow each observation's evidence link back to the underlying metric.
- 7Export a management summary.
Design decisions
- Metrics are the source of truth — filters and cards come first; the AI narrative is secondary and clearly labelled.
- Every observation links to a metric — the insight panel shows the metric, region, and date behind each claim, so it can be verified.
- The AI narrative is framed as interpretation — "Suggested interpretation" and a generation timestamp prevent it being read as observed fact.
- Drill-down is built in — the exception table is sorted by exception rate so the biggest problem is the first thing a manager sees.
AI interaction design
Input is InsightService.generateInsight({ timeRange, region, workflow }). Output is an OperationalInsight with a headline, observations (each carrying a metricId), suggested investigations, and evidence metric IDs. The UI resolves each metricId to the displayed metric and renders an inline evidence reference. The human judgement remains with the manager: the insight suggests where to look, not what to conclude.
Technical approach
Next.js App Router with a React client component for the dashboard. TypeScript types for OperationalMetric and OperationalInsight. A seeded, deterministic metric fixture generates 14 dates across 3 workflows and 4 regions. Recharts renders the trend charts. Atomic component structure with MetricCard, FilterBar, and TrendChart. The insight generator computes observations directly from the filtered metrics, guaranteeing the evidence link is real.
Tradeoffs
- Data is a deterministic fixture rather than a live warehouse, so the demo is fast and stable.
- The insight generator is rule-based over the displayed metrics rather than model-generated, which guarantees evidence linkage.
- Filters apply to the visible window only; there is no cross-period comparison engine.
Success measures
- Time to identify a bottleneck
- Action-item completion
- Trend detection accuracy
- User confidence in the insight
Next steps
Connect real data sources, add cross-period comparison and anomaly detection, persist action items, and evaluate whether the evidence-linked insight improves decision speed and confidence.