Amplify Audit: Creating Audit Prompts

For Middle Office customers, Amplify Audit replaces the manual, reactive process of catching Billing and Payroll errors with automated, AI-powered checks that run across Placements, Billable Charges, and Payable Charges at scale. Instead of discovering a Rate Card mistake or compliance gaps when an Invoice gets rejected, teams can surface issues immediately — with plain-English pass or fail results — freeing Middle Office staff from report-running and putting their focus back on resolution. Learn more about the power of Amplify Audit below.

Amplify Audit is loaded with prompts supporting common use cases by default. Learn how to create your own prompts, and more about the default prompts below.

Creating Your Own Audit Prompts

The prompt is the heart of every Audit. It outlines exactly what to check, what data to look at, and how to format the results. Because this is plain-language AI, you have a lot of flexibility.

Steps for Creating a New Audit Prompt

Follow the steps below to create a new audit and enter your prompt.

Amplify Labs must be enabled ( Menu > Amplify Admin > Labs) to complete the steps below.

1. In Amplify Admin settings, go to the Labs tab.

2. Select the slider icon.
   

3. Click Add.

4. Fill out the following fields:
   

   • Name: Give the audit a clear, descriptive name (e.g. “Placement Auditor”, “Billable Charge Audit”). This name appears in Action menus.

   • Entity Type: Select the entity this audit will be run from. This controls which list view and record actions the audit appears on; not which entities can be accessed from the prompt.

   • Color: Optional. Used to visually distinguish audits in the results list.

   • Prompt: The audit instructions. This is where you define every check the audit will run, and how results should be formatted.

5. Click Save.

General Prompt Principles

• Be specific. Vague instructions produce inconsistent results. Instead of "check the rate card looks right", write "pay rate must be less than bill rate".
• Use Bullhorn field names. Refer to fields by the names used in Bullhorn (for example, "placement status" not "rec status", "Job" not "requisition"). The audit uses the same names that appear in Bullhorn.
• Test incrementally. Start with two or three checks in global Amplify Chat (for example, "For placement 12345, check the following...") before adding them to an audit prompt. This lets you validate results quickly without running a full audit job.
• Define context explicitly. If a check depends on a relationship (for example, "corresponding payable charge"), define that relationship at the top of the prompt before referencing it in checks.
• Control output in the prompt. Specify whether you want a table, whether to show passes and fails or only fails, how verbose the reasoning should be, and so on.

If a check is returning incorrect or inconsistent results, try these approaches.

• Add more context: Define what the check means explicitly rather than leaving it ambiguous. Describe exactly what a correctly configured rule looks like rather than asking the audit to infer it.
• Specify the field: If the audit is pulling data from the wrong place (for example, bill rate on the job rather than the rate card), name the exact field or object you want it to reference.
• Test in chat first: Run the check against a specific record in Amplify Chat to see how the audit interprets it, then refine the wording.
• Break it down: If a complex check is unreliable, split it into two simpler checks.

Prompt Structure

A well-structured prompt typically contains three parts.

• Definitions: Any relationships or terms the audit needs to understand before running checks. For example: "The corresponding payable charge is defined as the payable charge with the same placement ID and week ending date."
• Checks: A numbered or bulleted list of everything to validate. Each check should be a single, clear condition that can return pass or fail.
• Output instructions: How to format the results. Specify the output format (for example, a markdown table with columns #, Check, Result, Reasoning), whether to include passes, and what to include in the structured output.

Example Checks You Can Build

The following table illustrates the types of checks you can configure. Any check you can describe in plain language can be added to a prompt.

Default Prompts

The following example prompt is provided as a starting point for a Placement Audit. You can use it as-is, modify the checks to match your business rules, or use it as a template for building prompts for other entity types.

Placement Audit

This prompt runs a full audit on a given placement. It evaluates Rate Card configuration, client comparisons, Time & Labor Rules, Customer Required Fields (CRFs), and general placement settings.

## Placement Audit Instructions

Run a full audit on the given placement. Each check produces a pass (✅) or fail (❌) result with reasoning.

### Audit Checks

**Rate Card**
1. Pay < Bill — Pay rate must be less than bill rate.
2. Markup > 20% — Markup percentage must exceed 20%.
3. CA Meal Break Penalty — If the placement state is California, a Meal Break Penalty earn code must be present.
   If not in CA, mark as ✅ with note "N/A — not in CA."
4. Minimum Wage — Base pay rate must meet or exceed the minimum wage for the placement city/state.
   Use a web search to confirm the current local minimum wage.

**Rate Card — Client Comparison**
5. Matching Earn Codes — Rate card earn codes should match those on other active placements for this client.
6. Bill Rate & Margin Aligned — Bill rate and margin should be consistent with other active placements for this client.

**Time & Labor**
7. Company Rules Consistent — T&L rules should be consistent with other placements for the same company.
8. State Rules Consistent — T&L rules should be consistent with other placements in the same state.

**Customer Required Fields (CRFs)**
9. CRFs Present — If the company typically uses CRFs, the placement must have them. List all CRFs in use by the customer.

**General**
10. Time Capture Method Set — A time capture method must be configured.
11. Synced to BTE — Placement must be synced to BTE.

### Output Format
Return a single markdown table: #, Check, Result, Reasoning.
Use ✅ for pass and ❌ for fail.

| # | Check | Result | Reasoning |
|---|-------|--------|-----------|
| 1 | Pay < Bill | ✅ | Pay $25/hr < Bill $35/hr |
| 2 | Markup > 20% | ❌ | Markup is 18%, below 20% threshold |

### Structured Output
Set success to false if any check fails.
Set message to the rendered markdown table.
Only return "success": true if every check passes.