# AI resolution and first-contact metrics

Evaluate conversation resolution and first-contact outcomes using AI, recent history and configured business rules.

*Release 3.3 · admin, supervisor*

**Settings › AI › Features**

## What it is

**Metrics** uses AI to assess conversation resolution and whether it was achieved on first contact. It records a status, confidence and explanation. Business rules and later events can change the recorded outcome.

## Who uses it and why

Users with **Settings** and permission to manage **AI**, or **Developer**, define the evaluation rules and review classifications. Supervisors use the recorded results to understand outcomes and identify conversations that need closer review.

Access to the feature's configuration and rule-adjustment tools requires **Settings** with permission to manage **AI**, or **Developer**.

## Where to find it

Open **Settings › AI › Features** and select a **Metrics** feature. Its result history lets you inspect conversations and use **Adjust Rules**.

Resolution and first-contact outcomes also feed the relevant AI performance reports.

## How it works

1. The evaluator reads the conversation and applies the configured **Metrics Instructions**.
2. The built-in evaluator includes eligible recent history within the lookback period, up to 30 conversations by default. Relevant earlier visitor-initiated conversations with recorded first-contact results can inform its decision.
3. It returns a resolution assessment, confidence and reason. First-contact resolution can be unknown when there is insufficient or inapplicable evidence.
4. INSIDE records the outcome and applies further business rules. A linked ticket forces the evaluated result to unresolved.
5. For autonomous conversations, later completed sales within the attribution window can mark the conversation resolved. A later satisfaction score of 4 or 5 out of 5 can mark it resolved; 1 or 2 can mark it unresolved.

The usual displayed outcomes include **Resolved**, **Unresolved** and **Undetermined**. Classification options can vary with the configured evaluation path.

When a repeat-contact result links earlier conversations, INSIDE can clear their previous first-contact resolution flags.

## Configure it

1. Create or edit a **Metrics** feature at **Settings › AI › Features**.
2. Select an **Account** and enable the feature.
3. Review **Metrics Instructions**, the lookback period and the sales attribution window.
4. Optionally enable **Add Resolution Status Thoughts on conversation** to add internal explanation notes.
5. Save the feature.

To refine rules from an example, open its result history and choose **Adjust Rules**. Select **Desired Resolution Status:**, explain the reason, then use **Adjust Metrics Rules**. Inspect the proposed changes and verification result before choosing **Apply Changes**. Applying changes creates a new feature version.

### Settings

| Setting | Where | Applies to | Default | Who can change it | What it does |
|---|---|---|---|---|---|
| Confidence Threshold for Resolution | Settings › AI › Features | website | 80% in the new-feature form; 0% if omitted | Settings role, Developer role | Sets the configured resolution confidence percentage; percentages below 100 do not enforce the displayed cut-off in this release. |
| Sales Attribution Window (Hours) | Settings › AI › Features | website | 2 hours in the new-feature form; website attribution period if zero or omitted | Settings role, Developer role | Sets the period from conversation start in which a completed sale can mark an autonomous conversation resolved. |
| First Contact Resolution Lookback Period (Hours) | Settings › AI › Features | website | 1 hour in the new-feature form; 24 hours if omitted | Settings role, Developer role | Sets how far back the built-in evaluator searches for earlier conversations relevant to first-contact resolution. |
| Add Resolution Status Thoughts on conversation | Settings › AI › Features | website | Off | Settings role, Developer role | Adds internal notes explaining recorded resolution decisions. |
| Metrics Instructions | Settings › AI › Features | website | Not set: uses the default resolution and first-contact evaluation instructions | Settings role, Developer role | Defines how AI should classify resolution and first-contact outcomes. |

### Fields

| Field | Where | Required | What it is for |
|---|---|---|---|
| Desired Resolution Status: | Settings › AI › Features | No | Selects the expected outcome for a conversation used to request a rule adjustment. |
| Explanation (why should it be classified differently): | Settings › AI › Features | No | Explains why the selected conversation should receive a different classification. |

### Permissions

| Permission | Lets you |
|---|---|
| Settings | Configure Metrics and inspect feature results when AI is allowed under Settings Permissions. |
| AI | Allows AI configuration within Settings Permissions. |
| Developer | Configure Metrics features and review rule adjustments. |

## Limits and good to know

- In 3.3, confidence percentages below 100 do not enforce the displayed percentage as a strict cut-off. Ask Powerfront to review results if you depend on this boundary.
- The new-feature form starts the lookback at one hour. Omitted configuration uses 24 hours. Check the saved value rather than assuming either default.
- The lookback is a retrospective history search. It does not prove that the customer will never contact you again.
- Default instructions allow **Resolved** when at least one of several customer intents is resolved. The result does not mean every need was fulfilled.
- A zero or omitted sale window uses the website's sales attribution period. It does not necessarily disable sale-based updates.
- Sale-based resolution and survey-based updates apply to autonomous conversations.
- A qualifying sale can mark a conversation resolved even when its service question was not answered.
- A missing evaluation is different from **Undetermined**. Unavailable AI, no matching feature or an execution failure can leave no result.
- Rule-adjustment verification tests one example. Applying it does not automatically reclassify all historical conversations.

## Depends on and works with

### This feature uses

- **Requires** [AI accounts and shared defaults](AIP-001.md): AI evaluation needs an available account and model.
- **Configured in** [Roles and permissions](ADM-008.md): Configuration access follows roles and Settings Permissions.
- **Configured in** [AI feature settings and change history](AI-001.md): Matching feature rules, filters and account select the Metrics configuration.
- **Configured in** [AI instructions and templates](AI-002.md): Metrics Instructions, shared defaults and prompt configuration define the evaluation rules.
- **Configured in** [Sales attribution to conversations and operators](COM-021.md): Website sales attribution supplies the fallback sale window.
- **Works better with** [Purchase tracking and sales totals](COM-020.md): Completed purchases can update an autonomous conversation to resolved.
- **Works better with** [Contact matching](CRM-003.md): Identified customer records can support looking up earlier conversations.
- **Works better with** [Custom scripts managed by Powerfront](EXT-015.md): A Powerfront-arranged website script can replace the built-in metrics evaluation.
- **Works better with** [Visitor survey questions and scores](VX-013.md): Later scored survey answers can change autonomous resolution outcomes.
- **Related** [Conversation history and search](CHAT-024.md): Earlier stored conversations provide evidence for reviewing first-contact outcomes.
- **Related** [Autonomous AI performance reports](REP-024.md): Autonomous AI Performance reports the recorded resolution outcomes.
- **Needs licence or module** Admin settings and feature switches (managed by Powerfront): Powerfront must enable AI before automatic metrics evaluation can run.

### Used by

- **Configures** [Autonomous AI performance reports](REP-024.md): AI metrics instructions and outcome rules determine the resolution classifications used in autonomous performance reports.

## FAQ and troubleshooting

### Why is a result Resolved when one question remains unanswered?

The default rules permit resolution when at least one intent is resolved. A later qualifying sale or satisfaction result may also have changed the outcome.

### Why did an earlier first-contact result change?

A later conversation can identify repeat contact and link the earlier conversation, clearing its first-contact resolution flag.

### Why is no result recorded?

Check that AI is enabled, a matching **Metrics** feature and account are available, and the evaluation completed. An absent result is not an unresolved verdict.

### Why did the saved confidence threshold not exclude this result?

Percentages below 100 do not enforce the displayed cut-off in this release. Later business-rule updates can also set a resolved outcome independently.

### Does Verification Passed prove the new rules are suitable for all conversations?

No. It confirms the proposed rules produced the requested classification for that example. Review other representative conversations before relying on the change.