How do I build line items with the AI Assistant?
The AI Assistant — also known as the Line Item Builder — helps you generate, refine and test automated line items without writing prompts from scratch. Describe what you want to measure in plain language, upload an existing framework, or start from a line item you already have. The assistant proposes a full definition (name, description, theme, prompt and scoring template), lets you test it against real conversations, and saves the result as a draft or an active version.
Required Feature Flags
The following feature flag and permission are required to use this feature:
Feature Flag | Technical Name | Description |
Line Item Builder |
| Enables the AI Assistant for building, refining and testing automated line items |
Required Permissions:
Line Item Builder (
reporting.line-item-builder.access) — to open the AI Assistant and build, optimise and test automated line items
Opening the AI Assistant
Go to Analytics > Manage topics. If your contract has the Line Item Builder feature and your role has the Line Item Builder permission, an AI Assistant button (with a BETA badge) appears in the action bar next to New topic.
Click AI Assistant. The Line Item Builder chat panel slides in from the right of the screen.
Choosing how to start
The assistant opens with four starting options:
Option | When to use it |
Generate automated line items | You want the AI to propose a full set of line items based on your business context. Best when you're starting from scratch and want a broad initial pass. |
Upload file | You already have a quality framework, compliance checklist or scorecard document. Upload it and the AI converts the criteria into draft line items. |
Create a new line item | You know the specific line item you want to build. Describe it in the chat and the AI drafts a single line item, ready to review and test. |
Improve an existing line item | You want to iterate on a line item you've already built. Opens the picker so you can select one of your Draft or Active line items to refine. |
You can also type freely into the message box at any time — the assistant handles natural-language instructions like "add a Pass example for authenticated callers" or "the Fail explanation should mention refunds explicitly".
Supported upload formats
The upload option accepts:
Spreadsheets:
.csv,.xlsx,.xlsDocuments:
.docx,.doc,.pdfPlain text:
.txt
The AI reads the file, identifies scorable criteria and generates matching line items. You can then edit them further in the chat.
Reviewing generated line items
For each line item the assistant proposes, you'll see:
The name and description
The theme (used to group the line item on reports)
The prompt the AI will use to score conversations
The scoring template it recommends
The automation level — either Automated (the AI scores it) or Manual (a human evaluator scores it) — along with the assistant's reasoning for the recommendation
Click any line item card to expand it and review the full definition. You can edit the fields directly from the chat by typing follow-up instructions.
Testing before you go live
When you're ready to check accuracy, ask the assistant to test the line items. The testing console opens with two selections:
Select line items to test — up to 3 line items at once. Only automated line items can be tested; manual line items are scored by evaluators, so testing doesn't apply.
Select conversations — up to 25 sample conversations drawn from your contract's imported contacts.
Run the test and the console shows the AI's outcome for each conversation, its reasoning, and extraction cards that highlight the matching utterance in the transcript. Use Agree and Disagree to feed your judgement back into the model — these signals inform future prompt tuning across the platform.
Iterate on the definitions in the chat and re-run the test as many times as you need. Line items that have been tested show a Tested badge on their card.
Saving your line items
When you're happy with the results, open the confirm panel:
Save as Draft — the line items are stored as Draft versions under Manage topics. Existing scorecards keep scoring against the current Active versions. Use this when you want to review the change with your team or run more validation.
Save & Activate — the line items become Active immediately. Scorecards using them pick them up on the next scoring pass. The previous Active versions are retained in the version history, so you can reactivate them later if you need to roll back.
Every version records who saved it, when, and a short commit message describing the change. Open the topic in Manage topics to see the full version history.
Tips for a smooth session
Set your business context on the contract before you start — the AI uses it to tailor generated line items to your operation. Without it, the AI produces more generic output.
For the fastest way to see a good result, ask the assistant to draft a line item for a common, easy-to-score behaviour, for example a call opening: "I'd like to create a line item that assesses how well the agent welcomed the customer during the interaction — the agent must state the organisation name, their own name and then ask how they can help the customer."
Test with a mixed sample of conversations — pick some you expect to pass and some you expect to fail. This surfaces prompt weaknesses faster than testing on a single outcome.
If the AI keeps drifting from what you want, start a new chat (top-right of the panel) rather than fighting the current one. A clean context often produces a better second pass.
Related guides
How do I refine an existing automated line item — Iterate on a Draft or Active line item and save a new version
How do I build a Gen AI-based Line Item — Build a GenAI line item manually via New topic
How do I write effective prompts for GenAI line items — A reliable shape for prompts and outcome explanations that score consistently