Plain-language summary: AI in AppReclaim suggests, summarizes, and scores — it never executes. Every consequential action (reclaim, downgrade, disable, reassign) requires explicit human approval. We do not train generalized AI models on your data.
Overview: This document explains where AI is used in AppReclaim and the guardrails in place. You always approve actions before execution.
1. AI Use Cases
AppReclaim uses AI to enhance analysis and provide actionable insights. AI features include:
- Usage Summaries: AI-generated plain-language explanations of license utilization patterns and trends
- Idle Seat Detection: Deterministic detection of 30/60/90-day inactivity is performed by our connectors; AI assists with prioritization and explanation
- Downgrade / Reclaim Suggestions: Recommendations for license optimization based on usage analysis and cost savings potential
- Health Scoring (MSP Partners): Automated scoring of client tenant health based on multiple factors including license efficiency, integration health, MFA posture, and savings realization
- Renewal Talking Points: AI-generated negotiation insights and data points for vendor contract discussions
- Smart Recommendations: Proactive suggestions for cost savings and optimization opportunities
- Report Insights: AI-powered executive summaries included in PDF reports
- In-Product Assistant: A standardized AI chatbot that answers questions about your tenant and surfaces optimization paths; streams responses and respects the same approval gate before any action
2. What AI Does Not Do
AI in AppReclaim is advisory only. It does not:
- Execute any tenant changes (reclaim, downgrade, disable) without explicit human approval
- Override your organization's security policies or RBAC settings
- Access data beyond what you've authorized through OAuth integrations
- Make autonomous decisions about license actions
- Share your data with other customers
- Use Customer Data to train generalized AI models
3. Execution Model
All actions in AppReclaim follow a deterministic execution model:
- AI Recommends: AI analyzes data and surfaces recommendations
- Human Reviews: Your admins review recommendations in the dashboard
- Human Approves: Explicit approval required before any action executes
- System Executes: Deterministic connectors execute approved actions via official APIs
- System Logs: All actions are recorded in audit logs with actor, timestamp, and result
4. Data Handling
- AI prompts are routed through the Lovable AI Gateway, which currently fronts Google Gemini and OpenAI models. The Gateway and upstream providers are contractually bound not to train on Customer Data.
- Inputs are minimized: typically aggregated usage metrics, license counts, and SKU metadata. Personally identifiable information is excluded where not strictly necessary.
- AI insight regeneration on org-wide scans is gated by a 6-hour staleness check to avoid redundant processing.
- Data processed in accordance with the Privacy Policy and DPA.
4A. EU AI Act Positioning
AppReclaim's AI features are decision-support tools: they generate recommendations, summaries, and scores that an authorized human must review and approve before any action is taken. They are not used to make solely-automated decisions that produce legal or similarly significant effects, and we do not consider them "high-risk" AI systems under the EU AI Act Annex III. We monitor evolving guidance and will update this disclosure if our classification changes.
5. Oversight & Auditability
- All AI recommendations logged with context and reasoning
- Approval workflows required for all execution actions
- Complete audit trail available in dashboard and exportable
- Restore paths available where supported by underlying platforms (30-day default rollback window; 60 days for Enterprise tier and inherited managed clients)
- Grace periods for soft-delete operations where applicable
6. Contact
Questions about AI usage? Contact legal@appreclaim.com.
7. Reporting AI Errors or Concerns
If an AI output appears incorrect, biased, or otherwise concerning, please report it:
We investigate all reports promptly and may withdraw or adjust AI features that do not meet our quality or safety standards.
8. Change Log
- May 12, 2026 — Named Lovable AI Gateway and current upstream providers; added EU AI Act positioning (§4A), 6-hour insight staleness gate, in-product assistant, tiered rollback windows, and AI error reporting path (§7).
- January 18, 2026 — Initial publication.
