DocxIntel is licensed by deployment footprint — how many environments, how many GPU workers, what level of redundancy and support. Page volume is not an input. Process ten thousand pages or ten million; the licence is the same, and reprocessing your archive costs nothing extra.
The licence figure itself is quoted per deployment after a sizing call. Everything that determines that figure is listed on this page, and the Proof of Value fee and conversion price are both fixed in writing before any work begins.
We would rather explain the pricing model honestly than publish a table that would be wrong for most readers. A licence covers a specific deployment, and deployments are not interchangeable — so the number is quoted after we know what we are deploying, and it is fixed in writing before you commit anything.
If a published number matters more to you than a sized one, the arithmetic further down this page will tell you roughly where metered alternatives land at your volume.
There is no capability tier, no page bundle and no feature that unlocks at a higher spend. Everything below is included in every DocxIntel licence.
Ask, Analyse, Identify, Classify and Compare, plus drafting, comparison, workflows, agents, approval and audit, are one platform, not a dozen SKUs. You do not buy extraction and then discover workflow is an add-on.
No page allowance, no credit balance, no overage rate. Reprocessing a two-million-page archive after a model update or a schema change costs nothing beyond the compute you already own.
Layout, OCR, table-structure, classification and language models ship inside the deployment bundle and run on your GPUs. The models are ours; the hardware, the data and the keys are yours, and nothing loads from an external registry at runtime.
Air-gapped, private cloud, managed single-tenant and evaluation sandbox are all licensable footprints. Air-gapped operation is not reserved for a top tier — it is the default assumption.
Security patches, bug-fix releases, model bundle refreshes and upgrade runbooks are published throughout the support window, packaged for offline installation as well as connected ones.
Human-in-the-loop review is part of the product, so reviewer seats are not metered either. Add the whole operations team to the review queue without a seat conversation.
Every input to a DocxIntel quotation is a property of the deployment. None of them is a property of your document volume.
Sizing is done against your document mix, not a page count. The sizing and throughput reference explains how GPU workers translate into pages per hour for scanned, photographed and native inputs.
Metered parsing APIs and per-page IDP platforms are good software with honest published rates. The problem is not the rate — it is that the rate multiplies. Below is the list-rate arithmetic at two realistic enterprise volumes.
How you are charged
Published rate used below
100,000 pages a month (1.2M a year)
1,000,000 pages a month (12M a year)
Reprocessing a 2M-page archive once
Largest publicly priced plan allowance
Setup and onboarding fees
Self-hosted or air-gapped option
If page volume doubles
Illustrative list-rate arithmetic, not vendor quotations. Metered figures use published 2026 credit pricing of $1.25 per 1,000 credits at 10 credits per page ($0.0125) and 45 credits per page ($0.05625); per-page IDP figures use a published growth-plan rate of around $0.30 per page. Volume commitments, negotiated discounts, annual-payment discounts, setup fees and taxes are excluded. Verify against current vendor documentation before you decide — we would rather you checked the numbers than took ours on faith.
The per-page rate is the visible cost. These are the costs that show up in the project plan, the procurement file and the automation business case instead.
Per-page platforms commonly charge implementation or setup fees on top of the subscription, so the first-year cost is not the rate multiplied by the volume. Ask for the first-year total, not the unit price.
When re-reading the archive costs the same as reading it the first time, teams stop doing it. Schema changes get deferred, model upgrades are not applied retrospectively, and the historical corpus stays as badly extracted as the day it was ingested.
Failed jobs, tuning runs, prompt or schema iteration and A/B comparisons all consume pages. Development and calibration traffic is real traffic, and on a meter it lands on the same invoice as production.
Third-party risk assessment, a data protection impact assessment, legal review of the processing agreement and a residency argument for your regulator are all real project costs. One metered vendor caches parsed data for 48 hours by default, which is a control you then have to review, disable and evidence.
Where self-hosting exists on a metered platform it is generally an enterprise-plan entitlement, and it runs inside your cloud tenant — data residency rather than a genuinely air-gapped environment. You buy the most expensive tier to get the deployment model you needed from the start.
Metered spend rises with every process you automate, so the business case for the next workload is worse than the one before it. A fixed licence inverts that: the eleventh use case costs nothing more than the tenth, which is how automation programmes actually spread.
Four steps, each with a written output. You know the Proof of Value fee and the licence conversion price before the Proof of Value begins.
We go through your document types, the volumes you handle at peak, your target environment, your regulator and the systems the output has to land in. You leave the call knowing whether DocxIntel fits; we leave it knowing what we would have to deploy.
We size GPU workers, storage and resilience against your real document mix, then quote the annual licence for that footprint together with the fixed Proof of Value fee. The quotation states what each input costs, so you can see what changes if you drop the disaster-recovery environment or add a second site.
Deployed inside your infrastructure, measured on your documents, against an accuracy threshold agreed in writing beforehand. If it misses the threshold you keep the measured report and owe nothing further.
The conversion price was fixed before the Proof of Value started, so there is no renegotiation on the strength of a good result. The deployment you evaluated becomes the deployment you license, and the footprint recorded in the order is what renews annually.
The commercial objections that come up in every evaluation, answered without a discovery call.
Because the licence is sized by what we deploy, not by what you process, and we cannot size a deployment we have not seen. A single-node evaluation footprint and a dual-site high-availability install with a disaster-recovery environment are different pieces of engineering, and quoting a number before the sizing call would either overcharge the first or underquote the second. What we will commit to in writing, before you spend anything, is the fixed Proof of Value fee and the price the licence converts to if the accuracy threshold is met.
An hour with an engineer produces a written quotation: the annual licence for your footprint, the fixed Proof of Value fee, and the price the licence converts to if the accuracy threshold is met.
