Know what your AI completed, and whether it succeeded.

Define the work unit and what success means. We measure both from the full journey of user actions, workflow events, and AI traces, and your billing tool charges for it.

Talk to us

Moving from seat-based SaaS to AI work-based pricing?

Your agents are completing work that used to require people. But seats, tokens, and API calls don't tell you how many customer workflows actually finished. DataFramer helps you define, measure, and verify completed work so you can introduce usage- and outcome-based pricing.

Measure usage

How much completed work did each customer receive?

Understand value

What quality, efficiency, and business outcomes came with it?

Support pricing

Use completed work units for usage- or outcome-based pricing.

Explain the bill

Trace a measured unit back to the workflow that produced it.

Define the unit your customer understands.

A work unit is a completed piece of work your customer can count, understand, and verify. Define it differently for each workflow: a completed case, processed application, accepted questionnaire, resolved request, or another meaningful outcome.

Workflow-specific

Each workflow can have its own definition of completed work.

Journey-level

One unit can span multiple agents, traces, tools, systems, and user actions.

Value-aligned

Count the completed work delivered, not every low-level event or model call.

One customer outcome. Multiple agents, humans, and systems. One verified record.

A single questionnaire can pass through an agent, a reviewer, and an approval service before it is done. DataFramer stitches those steps into one journey and counts the finished work once, with the evidence attached.

Count the work. Then decide if it succeeded.

Three states, kept apart. Attempted is every run that started. Completed is the subset that reached a saved result, and only these enter the billable count. Succeeded is decided separately, from what happened next: user acceptance, edits, a workflow state, a quality check, or another signal from the journey. Price on completed work, on success, or both.

Turn completed work into a usage record you can explain.

Measure work units by customer, workflow, and time period. Completed is what you bill; attempted is kept beside it so the gap is visible. Trace every unit back to the journey, events, and AI activity that produced it.

CustomerWork unitAttemptedCompletedSucceededWaitingSuccess rate
Acme Completed cases 141 128 112 6 92%
Acme Questionnaire drafts 49 43 36 3 90%
Acme Processed applications 334 317 291 9 94%

Usage numbers your Finance team and customers can trust.

A missed outcome is lost revenue. A duplicate is an overcharge. Both are answerable from the record, because every counted unit keeps the evidence that produced it.

No double-counting

Each completed unit carries a work-unit ID and counts once. A retry keeps the same ID; a genuinely new piece of work gets a new one.

Abandoned & rejected work

Only a saved completion event enters the count. Attempts that are abandoned, corrected, or rejected stay visible as attempts, outside the billable total.

Late success

Success can be decided days later. The unit is assigned to the period its completion event names, not the period you happened to receive it.

Disputed charges

Open any counted unit and see the result, the acceptance record, and the supporting user, system, and AI activity behind it.

Reconciliation reports against your source systems are on the roadmap.

AI-assisted vs human

The journey records which steps an agent performed and which a person did, so you can separate AI-assisted work from work done entirely by hand.

Completion records are kept independently of raw traces, so historical usage stays verifiable after trace retention expires.

Works with your billing tool.

DataFramer decides what was completed and what succeeded. Your billing tool sets the price and sends the invoice.

Stripe

Send usage and success totals to Stripe meters for each customer.

Metronome

Send the same totals as usage events to Metronome.

Your system

Export by customer and period through CSV or the API.

Raise the success rate and delivered value.

Every unit that did not succeed is a lead. Find the pattern, fix the cause, and show the outcome improved.

Find

See which work units fail and how often.

Explain

Trace failures to the step, system, or AI behavior behind them.

Fix

Turn expert corrections into tested changes.

Prove

Compare the success rate before and after the change.

Stop pricing AI on proxies.

Price on the outcomes you define, then improve them.

Talk to us