P402
P402.io·Meter
DocsPricingDashboard
> _ AI COGS DASHBOARD

AI COGS
dashboard.

P402 Monitor turns the metering ledger into an AI cost-of-goods-sold dashboard with cost per feature, per customer, per workflow, and per accepted output.

See dashboardRun AI Spend Audit

Built on the P402 Monitor surface. Parent product page: /monitor.

The problem

AI cost is a lump sum, not a unit cost.

Provider invoices show a total. Embedded AI teams need cost per unit of product delivered, per customer served, per workflow executed. Without that attribution at event time, AI COGS stays a guess.

What this dashboard shows
Per feature
AI cost grouped under each product feature the call served.
Per customer
AI cost grouped under each paying customer the call served.
Per workflow
AI cost grouped under the task or process the call ran inside.
Per accepted output
Total AI cost divided by count of events with outcome accepted.
Retry waste
Tokens spent on retried calls that produced no new outcome.
Context waste
Tokens spent on context that did not change the outcome of the call.
Model and vendor
Cost split by model and vendor for the same feature, with concentration flagged.
Budget
Budget the event drew from and remaining balance per customer and feature.
Evidence
Receipt and evidence bundle status per event for audit and finance review.
How it works
  1. 01
    Create a P402 key.
    Sign in, generate a key. The key is returned once. P402 stores only a hash.
  2. 02
    Route AI calls through P402 or report meter-only events.
    Use the OpenAI-compatible endpoint, or post metadata-only events from your backend. Both land in the same ledger.
  3. 03
    Attach customer, feature, and workflow on each call.
    Events are indexed under every owner you supply. The same event rolls up under customer, feature, and workflow at once.
  4. 04
    Read AI COGS.
    The Monitor surface groups AI cost per feature, per customer, per workflow, with accepted-output rate and waste flagged.
Privacy

Meter economics, not content.

AI COGS numbers come from token counts and cost, not from prompts. Prompt and response storage are off by default. Sensitive customer workflows can run under private gateway mode.

  • 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 AI COGS per feature, per customer, and per accepted output 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

Read AI COGS per feature, not per invoice.

Wrap AI calls once. The dashboard groups cost per feature, customer, workflow, and accepted output. Retry waste and context waste are flagged so engineering can act on real numbers, not estimates.

Start free
For enterprise

AI COGS rolled up for finance.

AI COGS grouped by department, customer, vendor, and budget at event time, with outcome and evidence attached. Finance and engineering read from the same ledger.

Run AI Spend Audit
FAQ
What does AI COGS mean in this dashboard?+

AI cost of goods sold. The direct AI cost attributable to a feature, customer, or workflow. The dashboard rolls cost up under each of those owners at event time, alongside accepted-output rate and waste.

How is AI COGS different from infrastructure cost?+

Infrastructure cost is shared platform spend. AI COGS is the per-call cost tied to delivering a unit of product to a paying customer. The dashboard records that attribution at the moment of the AI call.

Does the dashboard require sharing prompts?+

No. Default privacy mode is metadata-only. Owner, model, tokens, cost, budget, policy result, outcome, and evidence status are recorded. Prompts and responses stay out unless tenant policy opts in.

How are retry waste and context waste flagged?+

Retry waste flags tokens spent on retried calls that produced no new outcome. Context waste flags tokens spent on context that did not change the outcome. Both roll up per feature, per customer, and per workflow.

What is the accepted-output rate?+

Share of events with outcome accepted out of all events for that feature, customer, or workflow. Read alongside cost per accepted output to find where AI COGS earns its keep and where it does not.

How does this relate to the AI Spend Audit?+

The AI Spend Audit is a one-time delivered report on top of the same metering layer. The AI COGS dashboard is the live surface. The audit gives finance a discrete deliverable, the dashboard gives engineering and finance an ongoing read.

Get started

Read AI cost as a unit cost, not a lump sum.

Start with the Monitor surface, or run the one-time AI Spend Audit to get a delivered report on top of the same metering layer.

See monitorRun AI Spend Audit