Classification & Routing
Where it fits: intake queues, service requests, tickets, permits, RFPs, contracts, claims, support issues.
Example: Classify incoming requests and route them to the right team, queue, or workflow.
Watch-out: Requires clear categories, routing rules, and escalation paths.
Summarization
Where it fits: calls, tickets, documents, cases, work orders, account histories, meeting notes.
Example: Summarize long inputs so users can act faster.
Watch-out: Must preserve material facts and context.
Information Extraction
Where it fits: forms, invoices, contracts, permits, PDFs, emails, evidence packets.
Example: Pull key fields, dates, obligations, IDs, amounts, risks, or entities.
Watch-out: Needs validation for accuracy and edge cases.
Knowledge Retrieval
Where it fits: policies, SOPs, manuals, playbooks, knowledge bases, training content.
Example: Answer user questions using approved source material.
Watch-out: Requires source governance and freshness controls.
Drafting Assistance
Where it fits: responses, reports, summaries, emails, memos, explanations, closeout notes.
Example: Draft first-pass content for human review.
Watch-out: Should not replace accountable review.
Decision Support
Where it fits: exceptions, prioritization, triage, risk flags, root-cause suggestions, next-best actions.
Example: Surface recommendations or risk indicators to support human decisions.
Watch-out: Decision rights and override paths must be clear.
Quality Review
Where it fits: closeout documentation, required fields, compliance checks, evidence review, QA workflows.
Example: Check completeness, consistency, and policy alignment before submission.
Watch-out: Requires a clear standard for complete or acceptable.
Workflow Orchestration
Where it fits: multi-step workflows across queues, approvals, documents, and systems.
Example: Coordinate next steps while keeping humans in approval loops.
Watch-out: Integration and permissions must be controlled.
Agent-Assisted Execution
Where it fits: structured, repeatable work where an AI assistant can help move steps forward under supervision.
Example: Prepare updates, collect missing information, draft actions, and prompt human approval.
Watch-out: Requires strong boundaries, logging, and oversight.