Prices you can put in a budget before the first call.
Every engagement starts from a published price and a written scope. Assessments and pilots are fixed fees, expert judgment is priced per item, and data work that keeps running after launch moves to a monthly retainer.
Five ways to work with us
Each model fits a different kind of risk. Fixed fees suit work whose scope we can see in advance; unit rates suit work whose volume you control.
Fixed-fee assessment
Three weeks for $6,800, one fee regardless of what we find. The usual first step when you have a list of AI ideas and need to know which ones your data can carry.
Fixed-scope pilots
One document family, one dataset, or one workflow, with a stated volume, a deliverable list, and a measured before-and-after. Prices run from $8,900 for a de-identification pilot to $26,000 for an agent environment build.
Per-judgment expert work
From $4.80 per judgment, billed on accepted judgments only. The rate depends on the credential the task needs, the length of what's being judged, and whether reviewers write rationales.
Pipeline builds
From $14,000 per pipeline: one distributed processing or embedding pipeline, built in your cloud and handed over with a cost model and a runbook. Your cloud provider bills the compute, not us.
Monthly managed operations
From $8,900 per month, a standing team that samples production traffic, labels failures, refreshes evaluation sets, and ships retraining data. Three-month minimum, then month to month.
Every starting price, in one place
Each price buys the scope written next to it. If your project fits that scope, that is the price. If it doesn't, the next section explains what changes it.
| Service | Engagement | Starts at | What that includes |
|---|---|---|---|
| AI Data Readiness Assessment | AI Data Readiness Assessment | $6,800 | Three weeks, fixed fee: up to 12 data sources inventoried and sampled, a readiness score for each candidate use case, and a 90-day plan. |
| Document AI & RAG Data Pipelines | Document AI pilot | $18,500 | One document family, up to 50,000 pages parsed, chunked, enriched and indexed, with a retrieval test set and an accuracy report. |
| Data Curation & Deduplication | Curation pass | $9,200 | Quality scoring, exact and near-duplicate removal, and evaluation decontamination for up to 20 million records, with a before-and-after report. |
| De-identification & PII Redaction | De-identification pilot | $8,900 | Up to 25,000 documents or 500 hours of transcripts redacted or pseudonymized, with recall measured on a hand-labeled sample. |
| Data Provenance & AI Compliance | Provenance package | $5,400 per dataset | A datasheet, Croissant metadata, a source and license register, a lineage manifest, and a draft public disclosure for one dataset. |
| Synthetic Data Generation & Validation | Synthetic data sprint | $12,800 | Generator design, up to 100,000 synthetic records, and a validation report on fidelity, downstream utility, and privacy leakage. |
| Fine-Tuning Dataset Engineering | Fine-tuning dataset sprint | $16,500 | Four weeks: a task specification, up to 5,000 reviewed training examples, a held-out evaluation set, and a baseline fine-tune comparison. |
| Expert Annotation & Preference Data | Expert judgment | $4.80 per judgment | Rubric scores and pairwise preferences from credentialed reviewers in your domain. Pilots start at 2,000 judgments. |
| Agent Trajectories & RL Environments | Agent environment build | $26,000 | One workflow rebuilt as a sandbox with tools, seeded data and an automatic verifier, plus 300 reviewed expert trajectories. |
| Evaluation, Benchmarks & Red-Teaming | Evaluation suite | $13,500 | A golden set of up to 400 cases, scoring rubrics, judge-model calibration against your experts, and a release gate in your pipeline. |
| Computational Data Processing | Pipeline build | $14,000 per pipeline | One distributed processing or embedding pipeline designed and built in your cloud, with a cost model, monitoring, and a runbook. |
| Managed Data Operations | Managed Data Operations | $8,900 per month | A standing team that samples production traffic, labels failures, refreshes evaluation sets, and ships retraining data every month. |
What moves a quote up or down
Quotes follow the work, and the work follows your data. These are the factors we size after the scoping call.
| Factor | Moves it up | Moves it down |
|---|---|---|
| Source format | Scanned, faxed, or handwritten pages; tables that span pages; call audio with crosstalk | Born-digital documents, structured exports, and clean transcripts |
| Sensitivity | Health or financial data that must be redacted, measured, and processed only on your premises | Records already de-identified, or operational data with no personal information |
| Expertise per item | Tasks where every item needs a physician, attorney, or licensed engineer | Tasks a trained annotation team can label, with experts reviewing a sample |
| Quality targets | Agreement or accuracy targets above our defaults, which add second and third reviews | Our standard gates: Krippendorff's alpha of 0.80 and a second review on 10% of each batch |
| Outcome labels | Records that never say what happened, so every label is made by hand | Records that already carry outcomes: paid, denied, reopened, readmitted |
| Timeline | Compressed dates that need parallel teams and on-demand compute | Flexible dates that let us batch expert work and schedule GPU jobs off-peak |
| Location | On-site work outside the Jackson area, or in more than one location | Remote access through your cloud tenant or a virtual desktop |
| Starting point | No guidelines, no rubric, no evaluation set | Existing guidelines, labeled examples, or a test set we can build from |
How a pilot becomes production
A pilot is a measurement, not a down payment. Its job is to tell you whether the production run is worth buying.
The pilot ends with a number
Every pilot closes with a before-and-after against an evaluation set your experts accepted. If the number didn't move, we say so and recommend stopping or changing course.
Unit rates come from actuals
The production quote is built from what the pilot actually took: cost per page parsed, per record redacted, per judgment, per trajectory. You see the arithmetic, not just the total.
Setup isn't paid for twice
Guidelines, rubrics, parsers, and the evaluation set built in the pilot carry straight into production. The production quote prices volume, not another round of setup.
Assessment fees are credited
If you start a build within 90 days of a readiness assessment, half the assessment fee is credited against it.
One-time work or a retainer
A fixed volume goes under a fixed-price statement of work. Data that keeps arriving, such as new claims, tickets, or call recordings, moves to Managed Data Operations at a monthly rate.
Payment terms
Invoices are due within 30 days by ACH, wire, or check. Travel outside the Jackson area is billed at cost and approved in advance. Rates in a signed statement of work hold for its full term.
| Engagement | When we invoice | Notes |
|---|---|---|
| Readiness assessment | 50% at kickoff, 50% at the readout | Half the fee is credited toward a build that starts within 90 days. |
| Fixed-scope pilots and sprints | 50% at kickoff, 50% on delivery | Delivery means the data, the report, and the provenance manifest are in your hands. |
| Expert judgment | Monthly in arrears, on accepted judgments | Judgments that fail our quality gates are redone at our cost and never billed. |
| Pipeline builds | 50% at kickoff, 50% at handover | Compute runs in your cloud account and is billed to you directly by your provider. |
| Managed Data Operations | Monthly in arrears | Three-month minimum term, then month to month with 30 days' written notice. |
Questions about pricing
Why publish prices at all?
Because the first question a budget owner asks is what it costs, and "it depends" wastes everyone's meeting. Starting prices tell you whether we're in range before you spend time on a call. The scope written next to each price is real: if your project fits it, that is what you pay.
Is the scoping call free?
Yes. The 30-minute scoping call and the written scope that follows cost nothing, and we don't need any of your data to write either one. See what happens after you contact us.
Do you mark up cloud or GPU costs?
No. Processing runs in your cloud account or on your hardware, so your provider bills you directly at your negotiated rates. Every pipeline quote includes a cost model that estimates those charges before anything runs.
What if the pilot shows the approach doesn't work?
Then the pilot did its job. You keep everything it produced, including the data, the guidelines, the evaluation set, and the report, and you owe nothing beyond the pilot fee. We would rather tell you to stop than sell you a production run that won't move the number.
Are expert judgments priced the same in every domain?
No. $4.80 per judgment is the starting rate for rubric scoring by credentialed reviewers on short items. Judgments that need a physician or an attorney, long documents, or written rationales cost more. We set the per-judgment rate after a paid calibration batch, so it reflects how long the task actually takes.
Will you sign our contract paper?
Usually, yes. We can work under your master services agreement, NDA, business associate agreement, or data processing addendum, or offer ours. Procurement moves faster when we see your templates early, so send them with your first message if you have them.
Keep reading
Bring us one messy data source and one question.
Tell us what the data is and what you want a model to do with it. A data engineer, not a salesperson, will reply within one business day with questions and a proposed first step.