Optimizations

Open Optimizations to measure AI spend savings from techniques such as cache reads and context compression. Compare would-have-paid costs with actual spending and see which models, providers, and users benefit most.

Use this page when you need to:

• Prove ROI after enabling compression or other optimization features.
• Compare projected versus realized savings for a reporting period.
• Find models, providers, or teams with low optimization coverage.
• Investigate why savings changed after access or model updates.

From this page you can learn projected and actual savings, breakdowns by technique, and which principals contribute most to saved spend. Savings appear only when optimization is enabled for the relevant users, teams, or agents—configure that in AI Access (or on Agents).

Next: pick a date range, review optimization metrics, then drill into Savings breakdown or Model and Team views. Use Usage and Cost for raw volume and spend, Traces for per-request evidence, or Dashboards for ongoing monitoring.

Who benefits

The Optimizations page is intended for administrators and managers who need to measure the impact of AI optimization features on cost and efficiency.

• Platform administrators and FinOps managers — Monitor organization-wide savings, evaluate ROI, and compare projected and actual cost savings over time.
• Organization managers — Verify that context compression and other features deliver expected benefits across teams.
• Team leads — Analyze savings for teams or user groups and improve configuration.
• Engineering managers — Identify models and providers that benefit most from cache reads, prompt compression, and related techniques.

Review optimization metrics

Summary cards at the top of the page provide an overview of organization-wide optimization metrics for the selected time period:

• Projected annual savings — Estimated annual cost savings extrapolated from optimization results in the selected period.
• Saved (net) — Net cost savings after subtracting retrieval cost from gross savings.
• Actual / Would-have-paid — Actual AI spend after optimization versus the estimated cost if the same requests had run without optimization.
• Saved (gross) — Total cost savings from cache read, prompt compression, and memory retrieval before retrieval cost.
• Retrieval cost — Spend attributed to memory and context retrieval used by optimization features.
• Retrieval hit rate — Share of retrieval attempts that successfully reused stored context.
• Prompt-compressed tokens — Tokens removed through prompt and context compression.
• Cached-read tokens — Tokens served from cache instead of being processed by the model.

Use these cards to compare projected versus realized savings, weigh retrieval overhead against gross savings, and confirm that token reductions are driving the cost improvements you see in Savings breakdown.

Analyze the Savings breakdown

The Savings breakdown chart displays daily AI cost savings as a stacked bar chart, showing how each optimization technique contributes to the total savings over the selected time period:

• Cache read — Savings achieved by serving repeated or cacheable context from the cache.
• Prompt compression — Savings achieved by reducing prompt size before sending requests to the AI provider.
• Memory retrieval — Savings achieved through memory-based context retrieval.

Use this chart to:

• Compare the contribution of each optimization technique.
• Monitor savings trends over time.
• Identify changes after configuration updates.
• Determine whether additional configuration, such as enabling context compression for more users, teams, or agents, could further reduce AI costs.

Compare these results with the Usage and Cost page to understand how optimization affects overall AI spending.

Analyze optimization by Model and Team

Use the Model and Team views to identify where AI optimization delivers the greatest cost savings and to determine which models and teams benefit most from optimization features.

The Model view breaks down savings and token reductions by model. Use it to identify high-cost models, compare optimization effectiveness across providers, and determine which models benefit most from cache reads, prompt compression, and memory retrieval.

The Team view groups savings and token reductions by team. Use it to identify teams that generate the greatest savings, compare optimization effectiveness across workloads, and determine where additional optimization—such as enabling context compression—could further reduce AI costs.

For each model or team, review metrics such as:

• Would-have-paid cost and Actual cost.
• Saved amount and Savings %.
• Token reductions from cache reads and prompt compression.
• The primary optimization technique contributing to savings.

Use these views together with the Usage and Cost page to compare overall AI consumption with the savings achieved through optimization.

Workflow and best practices

Use the following workflow to enable AI optimization, monitor its effectiveness, and continuously improve AI efficiency across your organization.

1. Enable context compression. Configure context compression for the required users, teams, or agents. For configuration instructions, see AI Access — Enable context compression.

2. Generate optimization data. Allow users to continue using OptScale AI Chat. The platform automatically applies cache reads, prompt compression, and memory retrieval where applicable, generating optimization data for analysis.

3. Monitor optimization performance. Use Dashboards for a high-level view of optimization health, cost savings, AI usage, and operational trends. Dashboards help you quickly identify changes in optimization efficiency or AI spending.

4. Perform deep analysis. Use Optimizations, Usage and Cost, and Traces to investigate optimization effectiveness, identify high-cost models or teams, compare optimization strategies, and analyze individual requests.

5. Refine your configuration. Based on your findings, adjust context compression settings, select more appropriate models, or update provider configurations. See AI Access and Providers for configuration details.

6. Repeat the workflow. Continue monitoring optimization metrics and periodically review detailed analytics to validate configuration changes, identify new optimization opportunities, and improve AI cost efficiency over time.

Fast monitoring

Use Dashboards to monitor AI optimization at a glance. Dashboards provide continuous visibility into cost savings, optimization efficiency, request volume, and operational health.

Create an Optimization dashboard. For instructions, see Create a dashboard.

At a minimum, include these panels:

• Cache efficiency
• Cost breakdown
• Total spend
• Total input tokens
• Models breakdown

For broader operational visibility, also include:

• Total output tokens
• Teams breakdown
• Total requests
• Success rate
• Average latency

These panels help answer questions such as:

• Are optimization features reducing AI costs?
• Has optimization efficiency changed recently?
• Did a configuration change affect latency or request success?
• Which models or teams account for most AI usage?

Deep analysis

When dashboards indicate unusual spending or optimization trends, use the following pages for detailed investigation:

Goal	Page
Identify which models achieve the highest savings and which require further optimization.	Optimizations
Compare optimization savings with overall AI consumption and spending.	Usage and Cost
Investigate optimization behavior for individual requests, including cache usage and compression.	Traces

Use these insights to:

• identify high-cost models and teams;
• evaluate the effectiveness of cache reads, prompt compression, and memory retrieval;
• detect optimization opportunities;
• validate the impact of configuration changes;
• optimize provider and model selection for cost and performance.