# ConveRact Intelligence workspace

Open **Intelligence** from the workspace navigation. Tenant administrators configure monitors and request AI investigations or procedure drafts. Staff can read investigations and review monitor issues. All records are scoped to the authenticated workspace. AI access uses the platform-managed configuration and existing usage allowance.

## Conversation monitors

Create a named monitor, choose a channel or all channels, and select one condition:

- Unanswered: the latest recorded public message is from the customer and is older than the configured number of minutes.
- Waiting for staff: the conversation currently awaits a human and the latest customer message is older than the threshold. This is customer-message age, not measured handoff duration.
- Negative sentiment: the existing conversation sentiment label is negative. This is not a customer satisfaction survey or a new AI quality judgement.
- Phrase: the latest customer message contains a configured case-insensitive literal phrase. Commas separate phrases; regular expressions are not supported.

Monitors start paused. Enable a monitor after reviewing its rule. Enabled monitors run every minute while the workspace service is active. **Scan enabled monitors now** runs the same checks on demand. The first match creates one review issue and one in-app notification for that monitor/conversation pair. Repeated scans do not create duplicate issues. New evidence updates the issue and can reopen a resolved issue. Customers receive no message from monitoring. There is no automatic email, Slack notification, or customer handoff.

Assign an active workspace member and acknowledge the issue, or resolve it with a review note. Open the linked conversation to perform actual customer assistance. If a rule stops matching, the issue is resolved automatically with a recorded reason. Pausing a monitor preserves its history and stops future scans. At most 20 monitors can be configured; the queue displays the latest 500 issues with total counts.

## Evidence-linked investigations

Ask a question, select 1–90 days and a channel. Follow up on a saved report to preserve the previous question as context without modifying it. The system calculates exact counts for conversations with recorded public messages in that period, then selects up to 48 relevant/recent messages for AI analysis. Counts of resolved, waiting and negative conversations describe their current stored state, not historical rates.

Reports show the sample size, period, channel, findings and source-message links. Email addresses, long numbers, links and obvious credentials are redacted before sending excerpts to ConveRact AI; this does not guarantee removal of every possible personal detail. A valid citation means the referenced message exists in the selected sample, not that an AI interpretation is necessarily correct. Review full context before acting. Insufficient evidence and invalid model output do not create invented findings. Reports never change customer records, send messages or publish knowledge.

The latest 200 reports are retained; the workspace displays the latest 50. A one-minute interval between report requests limits accidental repeated usage. Removing or anonymizing a source conversation removes reports containing its evidence.

## Procedure builder

Provide a name and a written operating procedure covering open-cart assistance, quote follow-up or subscription billing assistance. ConveRact AI proposes supported journey settings and lists assumptions or unsupported requirements. This is a typed recovery builder, not arbitrary code execution or a general integration generator.

Review the source, message, increasing delays, cooldown, contact cap, quiet hours and every caveat. Edit the draft and confirm review. Applying it always saves a **paused** journey with **approval required**, even if a request attempts to enable it. Reapplying the same draft returns the existing journey. Configure customer identity, permission and the live channel in Revenue workspace before separately enabling it. The existing source, conversion, opt-out, pricing and delivery checks still apply. No new payment, discount or customer message is created by applying a procedure.

## Session API

These endpoints use the existing sign-in session cookie and `X-CSRF-Token` for mutations. They do not accept an external workspace API key. Administrators must not embed store credentials or session tokens into public client scripts. Required roles are enforced on the server, and a tenant query parameter cannot switch the workspace.

| Method and path under `/api/v1` | Request and access |
| --- | --- |
| `GET /intelligence` | Admin/staff; configuration, latest issues/reports and counts |
| `POST /intelligence/monitors` | Admin; `name`, `kind` (UNANSWERED/HANDOFF/NEGATIVE/KEYWORD), optional `enabled` false, `minutes` 1–10080 default 30, `keywords` array up to 20 phrases of 2–80 characters, `channel` default ALL |
| `PATCH /intelligence/monitors/{id}` | Admin; `enabled` boolean |
| `POST /intelligence/monitors/scan` | Admin; empty object |
| `PATCH /intelligence/issues/{id}` | Admin/staff; `status` ACKNOWLEDGED or RESOLVED, optional active member `ownerId`; a `resolution` of 10–1000 characters is required for RESOLVED |
| `POST /intelligence/investigations` | Admin; `question` 10–1000 characters, `days` 1–90 default 30, `channel` default ALL, optional same-workspace `parentId` |
| `POST /intelligence/procedures` | Admin; `name` 3–120 characters, `procedure` 40–12000 characters |
| `POST /intelligence/procedures/{id}/apply` | Admin; `reviewed: true` and the reviewed `journey` object following the revenue API journey schema |

Channel filters: ALL, WEBSITE, WHATSAPP, INSTAGRAM, MESSENGER, TELEGRAM, EMAIL, SMS, VOICE. Success responses use `{ "success": true, "data": ... }`. Validation errors return 400; invalid sessions 401; insufficient roles 403; paused service 402; unavailable state or evidence 409; request cooldown 429; unusable AI output 502. These are review workflows, so an HTTP success never means a customer outcome has occurred.

Integration has no ConveRact installation or one-time fee. Recurring plan, usage and external provider charges continue to apply.