Manager Guide
How plant managers and operations directors use Capped AI analytics: dashboard overview, handover quality scores, anomaly alerting, incident report review, aggregate metrics, and data exports.
This guide is for plant managers and operations directors who use Capped AI's analytics layer to monitor shift quality, review anomaly alerts, investigate incidents, and export data for compliance. For operator-facing guidance, see the Operator Training Guide.
Dashboard overview
After logging in as a plant manager, the Capped AI dashboard shows a real-time view of shift health across all process units. The three headline metrics are:
| Metric | What it measures | Target |
|---|---|---|
| Shift Quality Score | Composite score (0–100) for each completed handover. Weighted from completeness, DCS verification, and sign-off time. | ≥ 80 per shift |
| Handover Completion Rate | Percentage of shift transitions that have a signed Capped AI handover (vs. missed or unsigned). | ≥ 98% over 7-day rolling window |
| Anomaly Alert Feed | Real-time stream of DCS deviations that operators did not acknowledge in their handovers. | Zero unreviewed critical alerts |
The dashboard refreshes every 60 seconds. Historical trends are available via the Analytics tab (weekly and monthly views). Click any metric to drill into the individual shifts or alerts behind it.
Reading handover quality scores
The Shift Quality Score is a weighted composite. Understanding the components helps you give operators targeted feedback rather than a raw number.
| Component | Weight | What drives it down |
|---|---|---|
| Completeness | 40% | Missing sections (pending actions, equipment status, alarm summary). Capped AI flags the specific missing fields. |
| DCS-verified data | 35% | Operators who ignore or override DCS-sourced alarms without a written explanation reduce this score. |
| Sign-off time | 25% | Handovers signed more than 30 minutes after shift end are penalised. Late sign-offs suggest rushed reviews. |
Spotting low-quality shifts
From the dashboard, click Handovers and sort by Quality Score ascending. The lowest-scoring handovers show the component breakdown in the detail view — look for patterns:
- Consistently low completeness on the same unit often indicates an operator skipping the voice note step. Show them the Operator Training Guide section on voice notes.
- Low DCS-verification scores with no explanations may indicate operators dismissing alarms they do not understand — worth a safety conversation.
- Chronically late sign-offs on the night shift often indicate tablet access issues, not operator negligence — check with IT.
Giving feedback to operators
Open any handover → click Add Manager Note. Notes are visible to the operator on their next login and are attached to the handover record in the audit log. Notes cannot be deleted after submission (they are part of the tamper-evident record).
Anomaly alerting
Capped AI continuously compares live DCS readings against the operator's handover narrative. When a significant deviation occurs that the operator did not mention or acknowledge, Capped AI raises an anomaly alert visible on the manager dashboard.
How anomaly detection works
- Capped AI subscribes to OPC-UA tag values at 1-minute granularity during each shift.
- At handover submission, the AI cross-references the narrative against the tag history. Tags that exceeded their configured threshold without an operator explanation are flagged.
- Critical alerts (threshold exceeded by >20%) are shown in red on the dashboard immediately. Non-critical alerts appear as amber.
- Alerts are automatically resolved when the operator adds an addendum to the handover explaining the deviation.
Configuring alert thresholds per tag
Navigate to Settings → Anomaly Thresholds. Tags are grouped by process unit. For each tag you can set:
| Field | Description |
|---|---|
| Baseline band (±%) | The acceptable operating range expressed as a percentage deviation from the historian's rolling 30-day average. Default: ±10%. |
| Alert level | Critical (immediate red alert) or Warning (amber). Default: Warning for all tags. |
| Minimum duration (min) | How many minutes the tag must be outside the band before an alert fires. Prevents noise from momentary spikes. Default: 5 minutes. |
| Enabled | Toggle alerting on this tag on or off. Tags disabled here will not appear in the anomaly feed. |
Changes take effect on the next shift. Threshold history is logged — you can see what the setting was at the time of any historical alert.
Suppressing false positives
If a tag regularly fires alerts during normal operations (e.g., planned maintenance windows, seasonal variation), you can suppress it without disabling alerting entirely:
- Scheduled suppression: Settings → Anomaly Thresholds → select tag → Add Suppression Window. Set start/end time and repeat schedule (daily, weekly, or one-time). Alerts during the window are logged as “suppressed” but not shown in the active feed.
- Per-alert dismissal: Open any alert → Dismiss → select reason (Planned maintenance / Expected variation / Confirmed safe / Other). Dismissed alerts remain in the audit log with your dismissal reason and timestamp.
- Threshold widening: If the false positive rate on a tag is persistently high, increase the baseline band for that tag in Settings.
Incident report review
When a sequence of anomaly alerts and operator notes forms a coherent incident pattern, Capped AI automatically generates an AI incident summary. These summaries appear in the Incidents tab on the dashboard.
Viewing AI-generated incident summaries
Each incident summary includes:
- Timeline of relevant DCS tag deviations
- Operator handover excerpts that mention the affected equipment
- Anomaly alerts raised during the incident window
- AI narrative explaining the probable sequence of events
- Confidence score (how many independent data sources corroborate the narrative)
Summaries are AI-generated starting points. Always validate the narrative against raw DCS historian data before including it in a formal incident report.
Linking handovers to a root cause chain
From the Incident detail view, click Link Handovers. You can:
- Search for and attach any handover from any date range to the incident record.
- Mark a handover as Root Cause, Contributing Factor, or Follow-up Action.
- Add a manager note to any linked handover explaining the connection.
- The root cause chain is preserved in the incident record for the full 7-year WORM retention period.
Exporting PDF incident reports
From the Incident detail view, click Export PDF. The PDF includes:
- Cover page with incident ID, date range, affected units, and your plant name
- AI narrative summary (clearly watermarked as AI-generated)
- Full text of all linked handovers (with operator signatures and timestamps)
- DCS tag trend charts for the incident window
- Anomaly alert log with dismissal notes
- Manager notes and root cause chain
- SHA-256 integrity hash of the export (for chain-of-custody verification)
PDF exports are also logged in the audit trail. The export record includes the exporting user, timestamp, and a hash of the exported content.
Aggregate shift metrics
The Analytics tab provides weekly and monthly trend charts across all process units. Available metrics:
| Metric | Description | Why it matters |
|---|---|---|
| Average handover time | Time from session start to operator sign-off, averaged across all shifts in the period. | Consistently long handover times may indicate operators need more training or the AI draft quality is low. |
| Alarm acknowledgement rate | Percentage of DCS alarms during a shift that appear in the handover (mentioned or explicitly dismissed). | Low rates correlate with higher anomaly alert volume and incident frequency. |
| Permit compliance rate | Percentage of permit-to-work items listed in the handover that are confirmed closed or transferred to the next shift. | Regulatory requirement in many jurisdictions. Trend below 95% warrants immediate investigation. |
| Shift quality score trend | Rolling average of the composite quality score over the selected period. | Use to track the effect of operator feedback and training interventions. |
| Unsigned handover rate | Percentage of shift transitions where no signed handover was completed within 4 hours of shift end. | Persistent unsigned handovers are a compliance and safety risk. |
Charts can be filtered by process unit, shift type (Day / Evening / Night), or operator. Click any data point to see the individual handovers behind it. Use the date-range picker to compare week-over-week or month-over-month.
Exporting data
CSV export of shift logs
Navigate to Analytics → Export → Shift Logs (CSV). Configure the export:
| Field | Options |
|---|---|
| Date range | Any range within the appliance retention period (default: 7 years) |
| Process units | All units, or select specific units |
| Columns | Handover ID, unit, shift type, operator, start/end time, quality score, anomaly count, sign-off time, pending actions count |
| Format | UTF-8 CSV, ISO 8601 timestamps, semicolon or comma delimiter |
The exported CSV is suitable for import into SAP PM, Maximo, or any ERP/CMMS that accepts tabular shift data. The handover ID column can be used as a foreign key to link back to the full handover record in Capped AI.
Audit log download for compliance
Navigate to Settings → Audit Log → Export. The audit log export covers:
- All user authentication events (login, logout, failed attempts)
- All handover submissions, edits, addendums, and manager notes
- All anomaly alert events (raised, dismissed, suppressed)
- All exports and PDF downloads (with recipient user)
- All configuration changes (threshold updates, user role changes, suppression windows)
- All software updates applied to the appliance
Audit log exports are available in JSON (structured, machine-readable) or PDF (human-readable, suitable for regulatory submission). Each export is itself logged and hashed. The raw audit log on the appliance is WORM-protected and cannot be altered or deleted without physical access and dual-person authorization.
For scheduled compliance exports (monthly, quarterly), contact hello@capped.ai to configure automated exports delivered to a secure destination of your choice.