AI Unit Costs
Detect abnormal AI spend before it becomes a billing surprise.
AI Cost Anomalies continuously monitors your AI spending and automatically detects unusual cost patterns across providers, models, projects, API keys, and teams. Every anomaly includes root cause analysis to help you understand what changed and where to investigate.
Why use AI Cost Anomalies?
AI costs can increase unexpectedly due to:
Traffic spikes
Incorrect model routing
Prompt changes
Agent loops
Misconfigured applications
New deployments
Shadow AI usage
Manually monitoring dashboards isn't scalable. AI Cost Anomalies proactively identifies unusual spending so your team can respond before costs escalate.
How It Works
OneLens uses statistical models to learn your historical spending patterns and establish dynamic baselines for every monitored dimension.
When spend deviates significantly from expected behavior, an anomaly is generated automatically.
No manual threshold configuration is required.
Detection Dimensions
Detect anomalies across multiple business and technical dimensions.
Supported dimensions include:
Provider
Region
Service
Model
Caller (User or API Key)
This helps quickly isolate whether a spike is caused by infrastructure, an application, or a specific workload.
Root Cause Analysis
Every anomaly includes an automatically generated root cause summary.
Understand:
What changed
Which dimension contributed most
Estimated cost impact
Timeline of the anomaly
Recommended investigation path
Instead of simply notifying you that costs increased, OneLens explains where the increase originated.
Token Contribution Analysis
Understand what actually drove the additional cost.
Depending on the provider, OneLens breaks down cost contribution across:
Input Tokens
Output Tokens
Cached Tokens
Reasoning Tokens
This helps determine whether increased costs were caused by larger prompts, longer responses, reduced cache efficiency, or reasoning-heavy workloads.
Anomaly Lifecycle
Track the complete lifecycle of every anomaly.
Available states include:
Open
Acknowledged
Investigating
Resolved
This allows teams to collaborate, avoid duplicate investigations, and maintain an audit trail of incident resolution.
Alerting & Notifications
Receive anomaly alerts through your existing communication channels.
Supported notification channels include:
Email
Slack
Microsoft Teams
ServiceNow
Notifications include a summary of the anomaly along with a direct link to investigate further.
Investigate Faster
Each anomaly provides enough context to begin troubleshooting immediately.
Quickly answer questions like:
Which model caused the spike?
Which API key generated the spend?
Which team owns the workload?
Is this affecting one provider or multiple?
Is the increase temporary or ongoing?
From the anomaly page, you can drill directly into the AI Cost & Usage Analyzer for deeper analysis.
Example Scenarios
Unexpected Model Upgrade
An application starts routing requests from GPT-4o Mini to GPT-4. OneLens detects the sudden increase in cost and highlights the affected model.
Agent Loop
An autonomous workflow repeatedly calls an LLM, generating thousands of unnecessary requests. The abnormal spending pattern is detected within the reporting cycle.
Cache Miss Spike
A prompt update reduces cache effectiveness, causing a significant increase in input token costs. Token contribution analysis highlights the drop in cache efficiency.
Shadow AI Usage
A newly created API key begins generating significant spend outside approved projects. The anomaly identifies the caller responsible for the increase.
Best Practices
Review anomalies daily to identify issues before invoices arrive.
Route alerts to the teams responsible for the affected workloads.
Investigate recurring anomalies to identify long-term optimization opportunities.
Use anomaly insights alongside AI Budgets to proactively manage spending.
Track resolution status to ensure every anomaly is investigated and closed.
Last updated