AI Margin Stack Try CSV preview

Free preview -> paid AI margin diagnostic

Find out if one AI workflow is eating your margin.

Upload a sanitized usage CSV, preview the first margin signal, then request a written diagnostic when the file shows a real workflow, customer, retry, or failed-run risk.

Try CSV preview See sample report

Free sample included · local CSV preview · early diagnostic from $149 when the file has signal

Sample margin report
Evidence Observed Cost Signal
Useful answers 18 delivered $29.90 value
Retried calls 11 observed $11.30 27% spend
Failed answers 6 observed included no value
Finding PDF Q&A uses 84% of Customer A plan revenue.

27% of its spend comes from retries. Add a soft cap and instrument failed outcomes before scaling.

Free preview Check one export before paying for a written diagnostic
OpenAI / Anthropic Find margin leaks before changing pricing or caps
Metadata first No prompt bodies, secrets, or production credentials

What the report reveals

$49/month can look profitable until one workflow costs $41.20.

Stripe tells you what the customer pays. Provider bills show aggregate spend. The report connects those two through workflow, customer, usable outcomes, retries, failed runs, and plan assumptions.

Generic observability explains what happened technically. The margin report asks where money is leaking and which decision should change next.

See the sample margin report
Customer A PDF Q&A workflow
Plan revenue
$49.00
Workflow cost
$41.20
Retry spend
27%
Margin left
$7.80

Sample AI margin output. Actual findings depend on the CSV columns and pricing context available in your product.

Preview first, pay after signal

Know what the diagnostic will answer.

Start with sample data or a local CSV preview. Request the deeper written diagnostic only when the file has enough signal.

Method Connect spend to outcomes.

Review provider cost, retries, failures, customer/workflow attribution, and pricing context.

Access CSV first. No production credentials.

Start with exported logs or metadata. Prompt bodies, response bodies, secrets, and customer data are not needed for preview.

Deliverable A decision report, not another dashboard.

Receive the evidence table, health finding, data gaps, and recommended cap, pricing, reliability, or model action.

Honest result No material issue means no integration recommendation.

If the file does not show meaningful exposure, the report should say so instead of inventing work.

Who built it Transparent founder-built product.

The first version is built directly by Duman Torbayev. Verify on GitHub.

Qualified use case

Best for paid AI products that already have usage data.

The diagnostic is intentionally narrow: one workflow, one export, one economics question.

Best first fits RAG, PDF Q&A, support agents, report generators, and agency AI apps.

Use it when fixed plans or client retainers need a workflow-level cost check before caps or pricing change.

Bring this An existing usage export with customer, workflow, cost, status, and pricing context.

Partial data is okay. The preview will show whether the CSV has enough signal or which fields are missing.

Not for you if No paying users or almost no model spend.

If monthly provider spend is under roughly $100, the report may honestly say there is no material margin issue yet.

Not for you if You need real-time observability today.

Use Langfuse, Helicone, or a gateway if your immediate need is traces, latency, evals, or live production monitoring.

First workflow diagnostic

AI Margin Diagnostic

$149

Async written decision report for one AI workflow or usage export.

  • Free CSV preview before payment
  • Workflow, customer, retry, and failed-run margin signals
  • Evidence table, data gaps, and recommended next action
  • Delivered in 2-3 business days after scope is agreed
Try CSV preview
Messy or larger export

Deep Workflow Audit

$399+

For teams whose export needs column mapping, normalization, or deeper segment review.

  • One larger workflow or multiple related exports
  • Column mapping and data-shape review
  • Customer/workflow concentration analysis
  • No subscription or live monitoring included
Request diagnostic

CSV fields

Your CSV only needs business metadata.

The preview works best with fields that connect provider spend to a workflow, customer, outcome, and pricing snapshot. Do not upload prompts, response bodies, secrets, API keys, or private customer data.

useful columns CSV preview
customer_idwho paid
workflow_idwhich AI feature
statuscompleted, failed, or blocked
cost_usdprovider cost
retry_countrepeated work
price_usdrevenue snapshot
Early diagnostic · async report · preview first

Request an early AI margin diagnostic when the preview shows a real signal.

Start in the app with sample data or a local CSV. Use this form when one workflow has enough signal for a written review of margin leaks, data gaps, and the next pricing, cap, reliability, or instrumentation decision.