---
title: "Budgets"
description: "Set daily and monthly spending caps for AI Agents in CRM > Workforce > Settings > AI Budgets, and estimate a realistic budget from token costs and conversation volume."
doc_type: explanation
audience: merchant
sidebar_position: 1
# Provenance — do not edit by hand; used by re-run diffing and redirects.
confluence_page_id: "4159766533"
confluence_url: "https://ultracart.atlassian.net/wiki/spaces/ucdoc/pages/4159766533"
last_updated_in_confluence: "2026-05-05T19:51:53.575Z"
---
AI Agent budgets let you control the operational costs of your AI Agents by setting daily and monthly spending caps. Once a cap is reached, agents stop picking up new conversations until the cap resets or you increase it.

## Overview

AI Agent costs are based on token consumption. Every message in a conversation -- from customer questions to agent responses to internal data lookups -- consumes tokens. UltraCart tracks this usage and charges per thousand tokens at three rates:

| Token type | Cost per 1,000 tokens | Description |
| --- | --- | --- |
| Input | $0.0004 | Customer messages and data the agent processes |
| Cached input | $0.0001 | Previously processed context reused within a conversation |
| Output | $0.001 | Agent responses sent to the customer |

Budget caps apply collectively across all AI Agents on your account, not per agent.

## What affects token consumption

Several factors influence how many tokens a conversation uses:

-   **Customer message length** -- longer, more detailed questions consume more input tokens
-   **Agent response length** -- comprehensive responses use more output tokens
-   **Data lookups** -- when the agent retrieves order status, subscription details, or product data, the retrieved information counts as input tokens
-   **Agent instructions and personality** -- more detailed instructions can lead to longer responses

## Example cost calculation

Here's a sample order status conversation to illustrate typical costs:

| Speaker | Message | Estimated tokens | Type |
| --- | --- | --- | --- |
| Customer | "I'd like to check on the status of my order." | ~12 | Input |
| Agent | "Can you give me the order number?" | ~11 | Output |
| Customer | "Sure, it's 1001234" | ~10 | Input |
| Agent | _(retrieves order data)_ | ~50 | Input |
| Agent | "Your order is currently processing and is scheduled to ship on May 27th. It should arrive within 2-3 business days of shipping. Is there anything else I can help with?" | ~41 | Output |
| Customer | "Nope, that's good. Thanks!" | ~5 | Input |

**Totals:**

-   Input tokens: 12 + 10 + 50 + 5 = **77 tokens**
-   Output tokens: 11 + 41 = **52 tokens**

**Cost:**

-   Input: 77 / 1,000 x $0.0004 = $0.0000308
-   Output: 52 / 1,000 x $0.001 = $0.000052
-   **Total conversation cost: ~$0.000083**

A typical short conversation costs a fraction of a cent.

## Setting your budget

To configure budget caps:

1.  Navigate to **CRM > Workforce > Settings**.
2.  Select **AI Budgets**.
3.  Set your **Monthly usage cap** and **Daily usage cap**.
4.  Select **Update Budget** to save.

:::tip

Start with a moderate budget and monitor actual usage during your first few weeks. Real-world data is the best guide for refining your estimates.

:::

## Estimating a realistic budget

Consider these factors when planning your budget:

-   **Conversation volume** -- how many customer interactions you expect daily and monthly. Review your current support ticket or webchat volume for a baseline.
-   **Average conversation length** -- simple order status checks use fewer tokens than complex subscription modifications.
-   **Peak periods** -- seasonal promotions or sales events may significantly increase conversation volume.

### Budget guidelines by volume

| Volume | Monthly interactions | Suggested starting budget |
| --- | --- | --- |
| Low | Under 100 | $1 - $5 |
| Medium | 100 - 500 | $5 - $15 |
| High | Over 500 | $15 - $50+ |

These are conservative starting points. Adjust based on your actual usage patterns.

## Monitoring and adjusting

After your agents are live, monitor usage regularly and adjust as needed:

-   Review token consumption in the AI Agent reporting tools
-   Optimize agent instructions to be concise and efficient, reducing unnecessary output tokens
-   Design conversation flows that resolve queries in as few turns as possible
-   Increase caps ahead of anticipated traffic spikes (promotions, product launches)

## Related pages

-   [AI Agents overview](./index.md)
-   [AI Agent capabilities](./ai-agent-capabilities.md)
-   [FAQ](./faq.md)
-   [Personality and instructions](./personality-and-instruction-examples/index.md)