# How to fix a bad AI phone call with Norm

By Samuel Oram-Jones

To fix a bad AI phone call on Bland, flag the call as a Triage issue and hit **Norm Fix**. Norm, Bland's AI assistant, pulls in the failing call and its agent, reproduces the failure, applies a fix, and verifies it against the original call.

You describe what you want in plain English, and Norm gathers evidence from your call logs and transcripts first. A change reaches live calls only when you publish and promote it, and [Bland's Norm guide](https://docs.bland.ai/tutorials/norm) sets out each stage.

## What do you need before you start?

You need an agent that has taken at least one conversation, and a call that went wrong. A Triage issue is a record that captures a problem found on calls, bundles the evidence, and hands the work to Norm to investigate and fix.

Live calls run the version pinned to **production**, and development is the one environment where you and Norm make edits. Publishing and promoting a version takes an owner, admin, or operator role.

## How do you fix a bad AI phone call with Norm?

You fix a bad AI phone call on Bland in six steps, from finding the call to promoting the version that fixes it.

1. Find the call. Open Conversations under Monitor in the sidebar of Bland's dashboard and click a row, and a panel opens on the right. Search matches the contact's number, the contact's name, the pathway name, and the conversation ID on rows already loaded, so press **Load more** if the call is missing.
2. Flag it. Every voice row has a **Flag** button at the far right, and the same button is in the panel header. Flagging opens a new issue, which takes a name, a severity, a description, and a note, or adds the call to one that already exists, which still takes a note. The submit button stays disabled until you write one.
3. Hit **Norm Fix**. Norm reproduces the failure, applies a fix, and verifies it against the original call.
4. Review the change. Every change Norm makes is reviewable on the issue, so you can see exactly what it touched before the fix reaches production.
5. Let Norm save a version. Norm makes its change in your agent's development environment, and once the change passes validation and any tests you run, Norm saves it as a new version. Saving a version does not change live traffic.
6. Publish and promote it. In your agent, open Environments under Configure, press **Publish** and choose **Publish to staging**, then press **Promote** in the Environments page header to point production at that version. If evaluations are configured on staging, the required ones must pass first, once the Evals feature is enabled for your organization.

<!-- screen: a Triage issue in Bland's dashboard with Norm Fix and the reviewable change, captured and dated for this page -->

You can also start without flagging anything. [Norm's chat](/norm/how-it-works) takes a call ID: copy it with **Copy Conversation ID** from the copy icon in the panel header, then ask Norm "Why did call `abc-123` fail to transfer?", and Norm pulls the full call and reproduces the failure before suggesting a fix.

To read the plan before any edit, "Walk me through what you'd change before doing it" is a valid prompt.

## How do you know the fix worked?

On Bland, Norm verifies its fix against the original call, and [Testbed](https://docs.bland.ai/agents/testbed) lets you check the exact turn that went wrong yourself. Open it from the test-tube icon on a conversation, or hover a turn in the transcript, then the icon that appears, then pick what to test.

Testbed re-runs one turn of a real conversation with your changes, without redialing anyone, as many times as you set. Identical answers are grouped and badged `count` / `total`, then marked **Matches original** or **Different from original**. Groups are marked against what the real call answered, so on a bad call, read **Different from original**.

That result is a stability reading rather than a pass or a fail, since five identical answers show consistency, not correctness. Testbed stops after one turn, so try the fix against whole conversations with [evaluations before launch](/no-code/how-to/test-an-ai-phone-agent-before-launch). After you promote, read the next calls in Conversations to see what production did.

## What can go wrong when you fix a call with Norm?

On Bland, most fixes stall at one of three points: a call missing from Conversations, a Testbed control that is hard to find or greyed out, or production still pinned to an older version.

- **A conversation I know happened is not in the list.** Simulations are out while Call source is on Live calls, a web chat appears only inside its agent, and the list loads in pages, so change Call source, switch the selector to the agent, or press **Load more**.
- **I cannot find a button that says Open in Testbed.** On a conversation it is a test-tube icon with no label.
- **The re-run answer is different and I changed nothing.** That is a real result, so run it several times to see how wide the spread is.
- **Push change is greyed out.** The agent has changed since that conversation, or you opened Testbed from a builder test chat, so make the change in the agent builder.
- **The agent answers with old behavior.** Production is pinned to an older version, because publishing to staging does not move production. Check what production is pinned to, then **Promote**.
- **Publish or Promote is disabled.** Publishing and promoting need an owner, admin, or operator role, and hovering the button shows which roles can use it.

## What comes next after the fix?

On Bland, a bad call can also become a test case that runs before you promote, once the Evals feature is enabled for your organization. [How Norm works](/norm/how-it-works) goes through each stage, and [building an agent with Norm](/norm/no-code-agent-builder) covers the build side. The [no-code AI call center](/no-code-ai-call-center) guide and the [no-code AI phone agents](/no-code) hub cover the rest.

Norm is also reachable from code, through the Norm plugin for Claude Code and Bland's Triage API, which creates issues, adds comments, and starts Norm sessions. Through that API, a comment that includes @norm dispatches a Norm investigation.

Idaho Housing and Finance Association saves $750,000 a year by replacing its IVR with Bland, which handles 4,000 inbound calls a day and cut average call time by 20%.

> "We save thousands of dollars a day."

Greg Blake, Chief Information Officer, Idaho Housing and Finance Association

## Frequently asked questions

### Does Bland have a no-code option?

Yes. Norm, Bland's AI assistant, builds voice agents using natural language descriptions, and Bland's dashboard includes a visual builder for non-technical users.

### Can Norm fix a Bland agent without an engineer?

Yes. On Bland, you flag a bad call and hit **Norm Fix**, and Norm reproduces the failure, applies a fix, and verifies it against the original call. The change still reaches live calls only when its saved version is published and promoted.

### Does Norm change live calls on Bland?

Only through a version you publish and promote. On Bland, Norm never edits what live calls run: it works in your agent's development environment, and approval to save a version is not approval to move live traffic.

### What is a Triage issue on Bland?

A Triage issue on Bland captures a problem found on calls, bundles the evidence, and hands the work to Norm to investigate and fix. Flagging a call opens a new issue or adds the call to one that already exists.

### Can I ask Norm about a call without flagging it?

Yes. On Bland, you can paste a call ID into Norm chat and ask what went wrong, and Norm pulls the full call and reproduces the failure before suggesting a fix.

### Does Testbed replay a whole Bland call?

No. Bland's Testbed re-runs one turn of a real conversation, so it shows what that one answer becomes when you change what led to it, not how the call would have gone.
