> _ P402 OPTIMIZE

Prepare AI spend for
measured savings.

P402 Optimize prepares AI spend for measured savings across models, cache, retries, and context. Recommendations are gated until baseline and outcome data prove the recommendation.

Status: readiness checks live. Recommendations gated.

See optimization readiness Read docs

Recommendations are gated. No savings claim ships before measured baseline and outcome data are available.

The problem

Savings claims without proof are noise.

Generic savings claims without measured baseline and outcome data are unverifiable. Optimize prepares the data plane that makes a savings claim verifiable: model substitution opportunities, cache hit ratios, retry waste, context waste, and accepted output rate per dollar.

What Optimize checks

Model substitution

Check for workflows where a cheaper model would meet the accepted outcome rate.

Cache hit ratio

Check for cache hit ratio per workflow and the share of cost served from cache.

Retry waste

Check for retried calls and the cost spent on attempts that did not produce an accepted output.

Context waste

Check for input tokens that did not contribute to the accepted output.

Prompt redundancy

Check for repeated prompt structure across events that could share a cached prefix.

Provider drift

Check for cost and outcome drift between providers and models over time.

Cost per accepted output

Check for cost per accepted output, segmented by workflow, model, and owner.

Outcome drift

Check for outcome rate change against the recorded baseline per workflow.

Readiness score

Check for whether each underlying check has enough baseline and outcome data to back a recommendation.

How it works

01
Connect P402 metering.
Route AI calls through P402 or send meter-only events. Optimize reads the same recorded ledger Meter writes.
02
Optimize collects baseline and outcome data over the historical ledger.
Baseline cost, outcome rate, cache hit ratio, retry rate, and context use are gathered per workflow, model, and owner.
03
Readiness checks run continuously over the data plane.
Each check tracks its own readiness state against the baseline and outcome data the ledger holds.
04
Recommendations remain gated until each check has sufficient proof.
A recommendation surfaces only when its readiness check has measured baseline and outcome data sufficient to back the claim.

Privacy

Check economics, not content.

Optimize reads the same metadata-only events Meter records. Readiness checks run on owner, workflow, model, vendor, tokens, cost, and outcome metadata. Prompt and response storage stay off by default.

metadata_onlyOwner, cost, tokens, budget, policy, outcome, and evidence status. No prompts. No responses. Default.
fingerprint_onlyAdds a hash fingerprint of prompt and response for dedup and replay protection. Content stays out.
redacted_tracePrompt and response retained after a redaction pass for fields the tenant policy allows.
private_gatewayYour environment hosts the inference, P402 records the economic event over a signed channel.
full_traceOpt-in. Prompts and responses retained verbatim with the event. Requires explicit tenant policy.

Read the trust posture

Proof

Same metering layer, four shipped workflows.

Each vertical demo is a working surface on the same metering layer. Use them to see the event detail, ownership attribution, policy result, and evidence status against a concrete workflow.

Healthcare

Medicaid prior authorization with HIPAA-aligned demo mode and human review gate.

Legal

M&A contract due diligence across an 8-document synthetic data room.

Real estate

Tenant application screening with fraud scoring and HUD fair-housing audit trail.

Enterprise

Department, employee, model, and workflow attribution across one organization.

For developers

Find the workflow that quietly eats your margin.

Optimize segments cost per accepted output by workflow, model, and owner. Readiness checks surface retry waste, context waste, and cache misses against the recorded ledger. Your code keeps calling the same endpoint Meter already records.

Start free

For enterprise

Build the proof finance needs before changing model choice.

Give finance, procurement, and engineering one place to see baseline cost, outcome rate, and readiness state per workflow. A model change or routing decision ships with measured baseline and outcome data, not a generic savings number.

Run AI Spend Audit

FAQ

Are recommendations live?+

No. Recommendations are gated. Readiness checks are live and run continuously over the historical ledger. A recommendation ships only after the underlying check has measured baseline and outcome data sufficient to back the claim.

What does readiness mean?+

Readiness means a check has gathered baseline and outcome data over enough events to back a savings claim. Each check tracks its own readiness state. The page surfaces those states without inventing a savings number.

Does Optimize change my code?+

No. Preparing the data plane does not touch the runtime path. Your code keeps calling the same endpoint Meter already records. The readiness checks run on the recorded ledger.

When do recommendations ship?+

A recommendation ships when its readiness check has measured baseline and outcome data sufficient to support the claim. Until then, the page reports check state, not a savings figure.

Can Optimize see prompts?+

No. Optimize reads the same metadata-only events Meter records by default. Readiness checks use owner, workflow, model, vendor, tokens, cost, and outcome metadata, not prompt content.

What is the first step?+

Create a P402 key, send metered events, then open the optimization readiness page. The readiness checks run over the events the ledger already holds.

Get started

Prepare the proof. Then change the model.

Connect P402 metering, let the readiness checks gather baseline and outcome data, and review readiness state per check. Recommendations ship when the proof is ready.