Skip to main content

AI Copilot

info

Enterprise Plan — This module is available on the Enterprise plan.

AI Copilot lets you embed AI assistance directly into your CRM workflows — drafting replies, summarising records, suggesting next actions, and executing guarded operations. You control exactly what the AI can do through prompt templates, action policies, and an approval queue. The built-in Evaluation Studio lets you test and compare AI outputs before they reach your users.

What's Included​

  • Copilot templates — reusable prompt templates tied to specific capabilities, with grounding configuration
  • Evaluation Studio — structured test suites with baseline comparison and ad-hoc evaluation runs
  • Action policies — role-based and permission-based controls per AI action key
  • Policy settings — global key-value configuration for AI behaviour
  • Approval workflow — human-in-the-loop review queue for sensitive AI-initiated actions
  • Governance dashboard — unified view of evaluation runs, pending approvals, and active policies

Copilot Templates​

Templates define the prompt that drives each AI capability. You create one template per use case — for example, "Draft Support Reply", "Summarise Contract", or "Suggest Next Action".

  1. Navigate to AI Copilot > Templates and click New Template.
  2. Enter a Name and an optional Capability key (for example, draft_reply, summarise, next_action). The capability key is what your code references when invoking the copilot.
  3. Write the Prompt Template. Use placeholder variables (for example, {{customer_name}} or {{ticket_body}}) that the system fills in at runtime.
  4. Optionally add a Grounding Config as JSON to attach data sources, retrieval settings, or model parameters.
  5. Toggle the template Active and save.

To edit a template, click its name in the list, update the prompt, and save. Changes take effect on the next copilot invocation. To remove a template that is no longer in use, choose Delete from the template list.

Evaluation Studio​

The Evaluation Studio lets you test how well your templates perform before enabling them in production. You create evaluation suites — collections of test cases — and run them to get scored results.

Creating an Evaluation Suite​

  1. Go to AI Copilot > Studio and click New Suite.
  2. Give the suite a unique Suite Key (for example, draft_reply_v2) and link it to a Capability.
  3. Add Cases as a JSON array. Each case contains an input, an expected output, and an optional scoring rubric.
  4. Set a Target Score (0–100) to define what a passing result looks like.
  5. Optionally attach a Baseline Run — a previous evaluation result to compare against.
  6. Save and click Run to execute the suite immediately. To delete a suite, choose Delete from the suite list. This does not remove historical evaluation run records.

Running Ad-Hoc Evaluations​

For quick, one-off tests without saving a suite:

  1. Go to AI Copilot > Governance and click Run Evaluation.
  2. Enter a suite key, paste the test cases as JSON, and submit. Results appear in the governance log.

Action Policies​

Action policies control which roles and permissions can trigger each AI action, and whether those actions require human approval before execution.

  1. Go to AI Copilot > Governance > Action Policies and click Add Policy.
  2. Enter the Action Key that identifies the operation (for example, send_email, update_contact).
  3. Specify Allowed Roles and Allowed Permissions as comma-separated lists.
  4. Enable Approval Required if this action must be reviewed by a human before it runs.
  5. Optionally add Redaction Rules (JSON) to strip sensitive fields from AI inputs or outputs.
  6. Save. The policy is evaluated every time the action is attempted.

Policy Settings​

Policy settings are global key-value pairs that control AI behaviour across your tenant — for example, default model parameters, rate limits, or feature flags.

  1. Go to AI Copilot > Governance > Policy Settings and click Add Setting.
  2. Enter a Policy Key (for example, max_tokens, temperature, enable_rag).
  3. Enter the Value as JSON (for example, 512, 0.7, or true).
  4. Save. The setting takes effect immediately.

Approval Workflow​

When an action policy requires approval, the AI Copilot queues an approval request instead of executing the action. Your team reviews and decides before anything is committed.

  1. Go to AI Copilot > Governance to see all pending approval requests.
  2. Open a request to view the proposed action, the requesting context, and any relevant data.
  3. Click Approve to allow the action to proceed, or Reject and provide a reason.

All decisions are logged with the reviewer's identity and timestamp for audit purposes.

For programmatic access to templates, evaluation runs, policies, and approval requests, see the AI Copilot API Reference.