Know the role, the sources and the tools. Decide where a person takes over. Give your AI employee a way of working your team can review.
Control layers
Start from approved knowledge sources, define what each AI employee can answer, and keep rollout scope clear.
Connect tools intentionally with OAuth, API keys, bearer tokens, custom headers, or scoped MCP servers.
Design the moments where Aivah should collect context, summarize, and route the conversation to your team.
Review available records in Insights → Conversations, Leads, Quizzes, and Calls alongside Usage. Check the date range, employee filter, and sample size.
Use the enterprise conversation to review data handling, retention, access, procurement, and compliance expectations.
Launch one contained workflow first, prove the operating model, then expand into more roles and channels.
Review checklist
Review the employee’s Knowledge sources and instructions, then test answers against the source material.
Scope each channel, integration, tool, credential, and action to the workflow it is meant to support.
Define escalation moments, required context, routing rules, and the teammate responsible for the next step.
Choose the conversations, calls, usage, and feedback your team will inspect. Distinguish Insights usage metrics from its sampled review signals.
Choose one repeat function with clear demand, source material, owners, and a measurable business outcome.
Define trusted documents, URLs, policies, scripts, media, and escalation rules before launch.
Deploy a focused role to one or two channels so teams can validate quality, safety, and handoff behavior.
Add MCP tools, phone, Slack, WhatsApp, CRM, calendar, or custom workflows once the operating boundary is clear.
Review conversation quality, available Insights records, usage, and team feedback before expanding the workflow.
Define the operating expectations alongside the job. Confirm deployment-specific commitments with your team.
Identify approved material, who maintains it and what the employee should do when information is missing or outdated.
Read the data guide