Standard CS workflows
At a glance
Six phases to every CS engagement: Discovery → Solution design → Configuration → Eval/UAT → Deployment → Monitoring/Expansion.
Each phase has inputs, outputs, and handoff criteria. Knowing the workflow prevents dropped balls between phases and gives customers predictable timelines.
The end-to-end workflow
Phase 1: Discovery
Goal: Understand the customer’s pain points, current process, and what “success” looks like.
| Input | Activity | Output |
|---|---|---|
| Customer intro, existing contracts | Stakeholder interviews, process mapping | Discovery doc: pain points, volume data, current workflow |
| Identify automation candidates | Prioritized use case list with estimated ROI | |
| Technical assessment | Integration requirements, data availability |
Key questions to answer:
- What percentage of CX volume is WISMO / tracking queries? (If >30%, WISMO agent is the obvious first deploy.)
- What systems do they use? (TMS, WMS, ERP — determines integration scope.)
- What are their data residency requirements? (Drives deployment mode choice.)
- What channels do their customers use? (Voice, WhatsApp, email, chat.)
Phase 2: Solution design
Goal: Map use cases to agents and define the technical architecture.
| Input | Activity | Output |
|---|---|---|
| Prioritized use case list | Map use cases to agent templates | Solution architecture doc |
| Integration requirements | Design integration layer (APIs, webhooks) | Integration spec |
| Data residency needs | Choose deployment mode (cloud/hybrid/on-prem) | Deployment plan |
| Define success metrics (SLA, accuracy, cost) | KPI framework |
Tools: Use the AgentFleet builder to configure agents. Use AgentFlow Copilot for workflow automation.
Phase 3: Configuration
Goal: Build and configure the agents, tools, and integrations.
| Input | Activity | Output |
|---|---|---|
| Solution architecture | Configure agents in Dashboard or via API | Agent definitions (workflow DAGs) |
| Integration spec | Set up tool connections (ProjectX, LIA, carrier APIs) | Working integrations |
| Write/customize system prompts | Prompt library | |
| Configure policies (HITL thresholds, retry, follow-up) | Policy configs | |
| Set up guardrails (PII masking, content safety) | Guardrail configs |
Phase 4: Eval / UAT
Goal: Verify agents work correctly before going live.
| Input | Activity | Output |
|---|---|---|
| Configured agents | Build eval scenarios (happy path, edge cases, adversarial) | Eval dataset |
| Run evals — check accuracy, tool-call correctness, latency | Eval results | |
| Customer UAT with real (or realistic) data | UAT sign-off | |
| Fix issues from UAT | Updated agent configs |
Eval categories: Happy path, edge cases, adversarial inputs, timeout handling, language switching, escalation triggers. See Eval framework for details.
Phase 5: Deployment
Goal: Go live with monitoring in place.
| Input | Activity | Output |
|---|---|---|
| UAT-approved agents | Deploy to production (cloud/hybrid/on-prem) | Live agents |
| Configure observability (New Relic, Langfuse, Elasticsearch) | Monitoring dashboards | |
| Set up alerting (error rate, latency, cost thresholds) | Alert rules | |
| Gradual rollout (% of traffic, then full) | Rollout plan | |
| Customer training on Dashboard | Trained customer admins |
See Deployment modes for cloud/hybrid/on-prem specifics.
Phase 6: Monitoring & expansion
Goal: Maintain performance and grow the deployment.
| Input | Activity | Output |
|---|---|---|
| Live deployment | Monitor accuracy, latency, cost in Dashboard + Langfuse | Weekly performance reports |
| Review escalation patterns — what’s the agent missing? | Agent improvement backlog | |
| Run periodic evals to detect drift | Drift detection alerts | |
| Identify new use cases from monitoring data | Expansion proposal | |
| Upsell additional agents | Expanded deployment |
Sources
- CS Org — Roles & Ways of Working
- See Templates for reusable documents at each phase
- See Eval framework for evaluation methodology
Changelog
- 26 May 2026: Full content with six-phase workflow, activities, and handoff criteria.