Analyze and Optimize AI Credit Consumption
Monitor AI credit usage by user, agent, model, and project in workspace settings to identify consumption patterns and optimize spending.
If individual users or AI agents consume particularly large numbers of credits, it's worth taking a closer look at their usage. High consumption doesn't automatically mean something is wrong. What matters is which models, agents, and features are being used, and which processing steps are required.
This article is for awork admins who want to analyze and optimize AI credit consumption. You'll learn how to spot unusual usage patterns, narrow down possible causes, and identify suitable optimization measures.
What causes high AI credit consumption?
Credit consumption depends, among other things, on the model used, the amount and complexity of the information being processed, and the tools and processing steps involved. High consumption is therefore not automatically a sign of inefficient usage.
A failed or incomplete AI action can also consume credits if AI resources have already been used.
👉 To learn which factors affect consumption and see typical ranges for different agent types, read the article AI Credits and Consumption.
Where can you analyze credit consumption?
To find out where the most AI credits are being used, you can review AI usage in awork. Go to Einstellungen → AI → Nutzung & Aktivität and select the time period you want to investigate. In the overview, you can analyze the credits consumed and the number of AI threads. You can filter, sort, and group the table by users, agents, models, projects, and how a thread was started, among other criteria. This lets you narrow down unusual usage patterns and, if needed, export the data as an Excel file.
👉 For detailed instructions on the available reports and filtering options, see the article Review AI Usage and AI Activity.
How should you interpret unusually high consumption?
High AI credit consumption doesn't automatically mean a user is doing something wrong. To narrow down possible causes, always consider consumption in relation to the number of threads, the agents and models used, and how the threads were started.
The following overview can help you identify unusual patterns and decide what to check.
| Unusual pattern | Possible explanation | What should you check? |
|---|---|---|
| High overall consumption across many threads | Intensive, possibly normal usage | Credits per thread and overall usage |
| High consumption across a few threads | Complex requests, extensive context, or resource-intensive processing | Model used, agent, and credits per thread |
| One agent accounts for a large share of consumption | Particularly demanding tasks or room to optimize the agent configuration | Agent configuration, tools used, and processing steps |
| Significant consumption spikes on individual days | Unusually intensive usage or more automations being triggered | Timeline and what triggered the threads |
| One user consumes more than others on comparable tasks | Differences in working methods or processing | Comparable tasks, the same model, and the same agent |
| High consumption from automated threads | Regular process execution or frequent triggers | How often the automations run and what triggers them |
How should you approach the analysis?
1. Narrow down the consumption and time period
Select the user with the unusual usage and an appropriate time period. First, check how many threads were started and how high the total consumption is in relation to the number of threads.
2. Compare usage patterns
Look at which agents and models were used and what started the threads. Where possible, compare the user with others who work on similar tasks. A comparison is most meaningful when the model, agent, and type of task are similar.
3. Investigate unusual areas more closely
Is consumption concentrated on a particular agent, model, or automated threads? Use this information to narrow down possible causes. If consumption is concentrated on one agent, it may be worth checking its configuration and the tools it uses. For automated threads, look into how often they run and what triggers them.
4. Decide on an appropriate measure
Depending on what you find, it may make sense to use a different model, review the agent configuration, or adjust automations. If consumption is caused by demanding and necessary tasks, it may simply be a normal consequence of that usage. The key is whether consumption can be reasonably explained by the tasks being handled.
Important: High consumption doesn't automatically mean inefficient usage.
The report helps you identify patterns and narrow down possible causes. However, it doesn't by itself show that a user is doing something wrong. When assessing usage, always take into account the scope of the task, the model used, and the type of processing involved.
Also note that consumption for individual agent runs within a thread isn't shown separately, and failed AI actions can still consume credits if AI resources have already been used.
What are the typical ranges for credit consumption?
Credit consumption depends, among other things, on the model used, the amount of information processed, and the tools involved.
👉 For typical ranges and more information about the factors that affect consumption, see the article AI Credits and Consumption.
How can you optimize credit consumption?
1. Review the model being used
awork offers the Fast, Balanced, and Deep Thinking model categories. Check whether the model being used is a good fit for the task. Also take the relative credit consumption shown into account.
👉 You can choose which models are available in your workspace in the AI model settings.
2. Review agents and their configuration
If high consumption is concentrated on specific agents, review their configuration and the tools and connections they use.
Consider what tasks the agents handle and how many processing steps those tasks require.
3. Review automations
If consumption mainly comes from automatically started threads, investigate the related activities and how often they occur.
This helps you understand which automations contribute to the observed consumption and whether adjustments would make sense.
4. Take model changes into account
Changing models during an active thread can increase credit consumption because technical prompt caching can't be reused across different models.
So when investigating differences in consumption, also take into account any model changes within threads.
What are the limitations of the report?
The usage data shows you how many credits were consumed and which users, agents, models, or start types were involved. However, it doesn't automatically tell you whether a specific request was necessary or efficient.
Also keep the following limitations in mind:
-
Consumption for individual agent runs within a thread isn't shown separately.
-
Private threads belonging to other users may not be accessible to admins if they aren't linked to a project. Certain metadata, including the agent, provider, model, date, and credit consumption, remains visible.
-
Failed or incomplete AI actions can also consume credits.
The data can therefore support your analysis, but it isn't by itself proof of incorrect or inefficient usage.
What happens if there aren't enough credits in the allowance?
Each user has their own monthly credit allowance. Credits can't be transferred between users or pooled across the workspace.
Once an allowance has been used up, AI usage is limited. More expensive models and more complex agent features are no longer fully available. No additional charges are applied automatically.
As an admin, you can upgrade a user to a higher AI plan during the current billing period. The upgrade generally takes effect immediately, and credits already used count toward the new allowance.
Summary
- High credit consumption isn't automatically a problem.
- First, analyze consumption by user, agent, model, time period, and start type. Compare similar usage, then check whether it makes sense to adjust models, agent configurations, or automations.
This helps you distinguish between intensive usage that's necessary for the tasks at hand and potential opportunities to optimize.
👉 Want to learn how to make effective use of AI credits? Find out more here: Use AI Credits Efficiently.
FAQ
How can I find out which user is consuming the most AI credits?
Go to Einstellungen → AI → Nutzung & Aktivität. There, you can review consumption for a specific time period and filter, sort, or group it by users, agents, models, and other criteria.
Which factors affect AI credit consumption?
Consumption depends, among other things, on the model used, the amount and complexity of the information being processed, the inputs and outputs, and the number of processing steps. Tools, connections, and an agent's configuration can also affect consumption.
Can I track credit consumption for individual agent runs?
No. The report shows consumption at the thread level. Individual agent runs within a thread aren't shown separately as consumption units.
Do failed AI actions also consume credits?
Yes. A failed or incomplete AI action can consume credits if AI resources have already been used.
Can admins view the contents of private AI threads?
Not automatically. Admins can see certain metadata for private threads, such as the agent, provider, model, date, and credit consumption. This doesn't automatically give them access to the full contents of private threads.
Powered by happysupport.ai