Skip to content
AurionAI

Usage patterns

3 min readBy Emmanuel Yazbeck

The Usage Patterns tab shows when callers reach your AI voice agent and how much the agent uses its AI language models (LLMs) during calls.

Availability: AI Voice and AI Voice + Helpdesk profiles. Every dashboard role can open it.

Prerequisites

  • Your organization uses the AI Voice or the AI Voice + Helpdesk profile.
  • Your voice agent has taken calls.

Open usage patterns

  1. In the sidebar, select Analytics.
  2. Select the Usage Patterns tab.
  3. Choose the period in the filter bar. Read Choose the period.

The Usage Patterns tab with the LLM figures and the call distribution charts.

Read the LLM figures

AurionAI counts the model requests made during the calls that started in the chosen period. Only voice calls count. AI work on conversations is not included.

  • Avg. Tokens/Call: the average number of tokens per call. Calls with no model request count as zero.
  • Total LLM Requests: the number of model requests, for all providers together.
  • Total Tokens: the input tokens and the output tokens together.
  • Avg. Latency: the average time a model took to answer, in milliseconds. A provider with more requests weighs more.
  • LLM Provider Usage: for each provider, the requests, the input and output tokens and the average response time.
  • LLM Error Rates: for each provider, the share of requests that failed, with the number of errors and requests.

A token is a small piece of text that a model reads or writes.

Read the call distribution

  • Hourly Distribution: the number of calls in each hour of the day.
  • Weekly Distribution: the number of calls on each day of the week.
  • Monthly Trends: the number of calls in each calendar month of the period.

Hours, days and months use the time zone of your browser.

Result

The tab shows the figures for the chosen period. The figures are computed from your current records each time the page asks for them, about every minute. Select Refresh to ask at once.

If something goes wrong

  • Error loading usage patterns The figures could not load. The message gives the reason when the server sends one. Select the retry button if it is shown, or change the period.
  • No data available No call started in the period, or no call used a model.
  • A provider has a high error rate. Some model requests failed during calls. If the rate stays high, contact support. Read Get help.
  • The tab is missing. Your organization uses the Helpdesk profile.

Still stuck? Submit a request or .