AI Post-Deployment Monitoring
AI post-deployment monitoring turns existing AI runtime evidence into control-plane records that governance, audit, work, and posture subscribers can use. It does not copy prompts, provider responses, source text, extracted values, vectors, credentials, or private storage paths.
Use it after AI systems, providers, profiles, context sources, Document Intelligence, watchers, or controlled agent runs have been active long enough to produce operational evidence.
What It Materializes
The Materialize post-deployment evidence action on the AI Governance operations tab creates sanitized records for:
| Record | Evidence included |
|---|---|
| Portfolio metrics | Run counts, failure counts, review queue size, action applications, token estimates, latency, policy blocks, budget blocks, and human review burden. |
| Trust posture signals | Provider health, provider-not-configured posture, profile version drift, retrieval/index freshness, quality degradation, policy or suspicious-usage counters, and agent execution health. |
| Data-flow entries | AI run and watcher execution lineage with safe source, target, status, and evidence references. |
The records are generated from AI-owned ledgers such as runs, suggestions, provider connections, profiles, context sources, index runs, embeddings, watcher signals, and agent sessions/plans/steps/tool calls.
AI Governance also keeps customer-managed consumption and subscription evidence: usage policies, usage ledger entries, subscription or service records, seat allocations, token-wallet posture, external spend imports, cost allocations, and operational targets. Creating those records adds sanitized Activity and audit evidence so reviewers can see who recorded the control and what period, provider, model, member, or service it applies to. Imported spend remains evidence and reporting posture; it is not treated as real-time provider enforcement unless a separate approved integration enforces it.
The Usage & spend tab in AI Governance lets admins record, edit, and review these consumption controls and subscription records from the same workspace: policy, seat, wallet, import, allocation, and target posture. Edits can adjust limits, scope, provider, model, surface, member, period, owner, or status where the record type supports those fields; each edit requires a governed reason and leaves an Activity and audit trail.
When an external spend import includes sanitized allocation lines, admins can reconcile it into normal cost-allocation records. Use this for provider bills, internal model clusters, or customer-managed subscriptions that are paid outside Novantra. Keep imported evidence summarized: amounts, currency, period, provider/model refs, member refs, seat counts, token counts, and source refs are appropriate; raw invoices, provider billing payloads, credentials, prompts, source documents, and contract documents are not.
The same tab summarizes utilization by member, period, and AI surface from usage ledger and cost-allocation evidence. Use it to see observed runs, token estimates, warning or block posture, and whether cost-allocation evidence exists for the same period. Where allocation evidence includes amounts and currency, the utilization table also shows allocated spend for the member and period. Organization-level rows appear when consumption is not member-specific.
Decision Receipt Handoffs
AI-owned decisions also hand off sanitized decision receipts to Governance Trace. Receipts are recorded for Document Intelligence run execution, suggestion review, suggestion application, reviewed decision dossiers, watcher signal conversions, and controlled agent session/plan/step/tool-call action states.
Those receipts bind the AI-owned record to safe source, subject, authority, evidence, lineage, and integrity refs. They do not copy prompts, model responses, extracted values, source text, vector payloads, provider payloads, credentials, or storage paths into Governance Trace.
Operator Workflow
- Open AI Governance.
- Select Operations.
- Choose Materialize post-deployment evidence.
- Enter the governed reason for creating the records.
- Review the resulting portfolio metrics, data-flow entries, and trust posture signals.
- Use subscriber workflows, work items, posture dashboards, audit packages, or SIEM exports only through approved descriptors and routes.
The materialized records are evidence and posture. They are not final legal, regulatory, accuracy, SLA, or certification claims.
Safety Boundary
Post-deployment records contain counts, states, safe identifiers, and evidence references. They intentionally avoid raw AI payloads and raw source content.
If a reviewer needs source details, use the owning source module, approved evidence display, governed retrieval, or the original AI run context. The control-plane record should remain a sanitized handoff record.
Cloud And On-Prem Parity
Cloud and on-prem deployments expose the same product contract for post-deployment monitoring. Deployment configuration decides which providers, OCR engines, vector backends, and runtime processors are available, but the evidence model, routes, permissions, Activity trail, docs, and subscriber descriptors stay aligned.
No-AI deployment profiles do not include the AI feature bundle.