Data valuation estimate
What your records could be worth to AI buyers
This calculator gives an estimate, shown as a wide range. It is built from the workflows your records hold: the chains of decisions that AI buyers have paid for.
This is an estimate. It is not an offer, a quote or an appraisal. The range is wide on purpose, and every rate behind it is an assumption. A real figure needs a sample of your records. The calculation runs in your browser, and nothing you enter is sent or stored.
What a real valuation takes
- A sample of the records. Without one, any figure is a rough guess, and a buyer who has not seen a sample may change the price later.
- Knowing the use. Records bought for post-training are priced on workflows. Records bought for pre-training are priced on volume.
- The history of the work: orders, transactions and every job from start to finish.
What would improve this estimate
- Check the starting numbers against a small, representative sample of your records and against completed deals.
Limits of this calculator
- The range is wide on purpose. Every rate behind it is an assumption, not a measured rate.
- The workflow rules are our assumptions, drawn from the pattern in published deals. Those deals were made by other firms, and Pewter did not advise on them.
- The estimate prices training tasks from counted workflows. It gives no value for text volume or for evaluation data.
- The model counts text records only. It does not value audio, images, CAD drawings, video or software code.
- Deal readiness is a rule of thumb, not a probability.
- The calculator does not check who owns the records or whether they can be licensed. Rights and privacy need an independent review.
- What a buyer already holds changes what it will pay. This estimate cannot see that.
- A real figure depends on what the records hold and what a buyer accepts after review. Some firms receive no offer.
How the estimate is worked out
The range
- The range is wide on purpose. Every rate here is an assumption.
- The range is built from the counted workflows, at $150 to $450 a task for a workflow of 500 accepted tasks.
- Value grows more slowly than volume. The ten-thousandth example of a workflow teaches a buyer less than the hundredth.
- A workflow of 500 accepted tasks is priced at the full rate. A larger one earns less per task and a smaller one more. Ten times the tasks is priced as about four times.
- Below 25 tasks the per-task rate stops rising.
- Because each workflow is priced on its own, three different workflows are worth more than one workflow with the same number of tasks. Variety counts.
- That rate is set so that firms like those in published deals land near what those deals paid. Those firms had 60 to 100 people and about six years of records. The rate is in line with what buyers and partners have told us.
- The bottom of the range is lower in industries with few or no published deals.
- Where records are regulated or privileged, the bottom and the top are both lower. A compliance review removes records.
- Known licensing restrictions lower the bottom, because some records may have to be left out.
- A firm that has closed is shown 3 to 10 times less, as published reports show.
- No range goes above $2.5 million. Larger deals have been reported, but we have too little evidence to estimate that high.
- Amounts are rounded to two figures. The estimate is not more exact than that.
- Text volume is shown but not priced. We know of no deal priced on it.
- Rates as of November 2026. Prices move as buyers fill their needs.
What counts
- These rules are our assumptions. They come from the pattern in published deals. They are not measured rates.
- Buyers in those deals licensed chains of skilled decisions, so the calculator counts workflows. Each industry has three. Each takes a fixed share of the monthly activities.
- A workflow is counted only when its system of record and the email and chat about it are both ticked. Records without the conversation around them are listed but not counted.
- The first six years of records count in full. Years after the sixth count half. Nothing past fifteen years is counted.
- "Not sure" gives a range of tasks. The low end uses the worst answer. The high end uses the middle answer.
- Law, healthcare, and financial services and insurance are regulated industries. They score lower on readiness, because no published deal involved regulated or privileged records.
The sums
- Usable months = counted years × 12 × the share of history that is usable.
- Monthly activities = the number you enter. If blank or zero, employees × stand-in activities per employee per month.
- Usable tokens = (monthly activities × tokens per activity + employees × extra tokens per employee per month, if Email and chat or Shared documents and files is ticked) × usable months × the share of text that is usable.
- Complete examples of a workflow = monthly activities × the workflow's share × usable months × workflow coverage × a linking factor. The factor is 1 for consistent IDs, 0.7 for some and 0.35 for mostly separate.
- Accepted tasks for a workflow = complete examples × the share that are not repeats × the share accepted after quality checks × an outcomes factor. The factor is 1 for reliably tracked, 0.65 for some and 0.25 for few or none.
- Priced-as tasks for a workflow = 500 × (accepted tasks ÷ 500) to the power 0.6. Below 25 tasks, accepted tasks × 3.31.
- The bottom and the top of the range use the priced-as tasks, added up across the counted workflows. They do not use the plain count.
- Bottom of the range = the low priced-as tasks × the low rate. It is halved where an industry has few or no published deals, and quartered where records are regulated or privileged. Known licensing restrictions quarter it again.
- Top of the range = the high priced-as tasks × the high rate. It is halved where records are regulated or privileged.
- For a firm that has closed, the bottom is divided by 10 and the top by 3.
- Deal readiness, out of 100 = 30 × the share of workflows counted + 10 for four or more years of records (less for fewer) + 15 × the linking factor + 20 × the outcomes factor + 15 if rights are reviewed with evidence (5 if you believe you have them) + 10 if the industry is not regulated. "Not sure" scores as the worst answer.
Method version 0.6. Rates as of November 2026. A modeling tool, not a formal appraisal.