Outcome-Based Pricing for Intelligent Document Processing: Per-Outcome vs Per-Page vs Per-Seat

Ask five intelligent document processing vendors what their product costs and you’ll get five different answers — or no answer at all. Per page. Per seat. Per token. “Contact sales.” The IDP market prices its work in at least four different currencies, and most vendors prefer it that way: when you can’t compare, you can’t negotiate, and you can’t budget with confidence.

This guide is an attempt to fix that — at least the education part. It lays out the pricing models used across the document-automation industry, who each one favors, where each one hurts, and the questions that cut through the fog no matter which vendor you’re evaluating. Including us.

What is an outcome, exactly?

Before comparing models, the word at the center of this guide needs a definition — because “outcome” is doing real work here, not marketing work.

In MaxxLogix®, an outcome is a business result, not a step count. An outcome is a forms submission that is stored or used in a workflow — or any document that is part of a workflow process. The counting rule is simple: one completed business process, one outcome, no matter how many steps it took to get there.

Three examples:

• Example 1. Someone completes a form and stores the document. That’s one outcome.
• Example 2. A user submits a form that runs through a two-step workflow requiring a digital signature. That’s one outcome.
• Example 3. A document is imported from email or scanned in — it’s an invoice, and it enters the invoice approval process. Seven steps. That’s one outcome.

Or, the way we’d explain it to a customer:

In MaxxLogix, an outcome is a piece of work that reaches a useful result. For example, a customer submits an invoice. MaxxLogix reads it, routes it for approval, and sends the approved information to QuickBooks. “Invoice approved and entered into QuickBooks” is the outcome.

Which gives us the line that anchors everything below:

The invoice is the document; the outcome is what got done with it.

An outcome is a process that supports a business result. Keep that definition in mind — it’s the unit the rest of this guide is built on.

How it adds up

Two scenarios, same four pages:

• A 4-page invoice is imported, read by IDP, routed through approval, and posted to QuickBooks. That’s 4 IDP Pages + 1 Outcome — the pages for the intelligence, the outcome for the business result.
• A 4-page scanned file is read by IDP for index fields and stored in the repository. That’s 4 IDP Pages · no Outcome — the intelligence was used, but no business process reached a result.

Same pages. Different math, because different work got done.

How IDP vendors charge you: the market’s models

Every IDP pricing model on the market starts from one of three inputs. The differences matter because the model determines what you’re actually paying for — and what the vendor is incentivized to sell you.

1. Per page / per document. You pay for each page or document processed — an invoice, a claim form, a page of a mortgage file. This is the classic capture-industry model: it maps cleanly onto scanning-era economics, where each page had a real marginal cost. It’s easy to understand and easy to quote.

Who uses it: established capture vendors and document-AI providers with volume-based tiers. It’s the model most buyers encounter first.

Where it hurts: under per-token-style AI billing, not all pages are equal — a simple one-pager and a complex document with splitting, line-item extraction, and summaries can cost wildly different amounts, because you’re paying for the computation, not the page. And per-page pricing on its own says nothing about accuracy, exceptions, or what happens when the document is messy.

2. Per seat / per user. You pay per named user, per month. The model the software industry has used for decades, applied to document automation.

Who uses it: vendors with platform plays and broad workflow suites — a few publish per-seat pricing tiers publicly, which is still the exception in this space.

Where it hurts: seats don’t process documents. A company with 500 employees who touch documents occasionally pays for 500 licenses whether the system delivers 500 outcomes a day or five. Per-seat pricing measures access, not results — and it creates a quiet tax on adoption: every new user you add to get value out of the system raises the bill, so IT rations licenses, and the platform never reaches the people who’d benefit most.

3. Consumption / token-based. You pay for what the AI consumes — tokens in and out, API calls, operations. This is the newest model and increasingly the default for AI-powered features. The pattern is visible across the market’s recent AI launches: assistants shipping with a monthly token or operation allowance, overage billed on consumption, and IDP capabilities offered as consumption-based add-ons. Metered AI consumption is becoming the incumbent norm — even vendors that don’t publish pricing are building on token-metered infrastructure underneath.

Who uses it: any vendor whose AI features run on large language models. Some vendors’ own platform messaging now warns buyers about “skyrocketing token costs driven by frontier models” — an acknowledgment of the dynamic from inside the industry.

Where it hurts: this is the least predictable of the three. Your bill scales with usage you can’t fully control or forecast — a busy month in claims, a discovery surge, quarter-end close. Every spike in real business activity becomes a spike in AI spend. Worse, the model punishes thoroughness: the long, complex documents where AI delivers the most value — dense contracts, multi-hundred-page medical records, sprawling case files — are precisely the ones that cost the most to run. You’re taxed most on the work that matters most. Finance starts treating AI as a cost to contain rather than a capability to deploy.

The fifth model: outcome plans

Here’s where we’re honest about our own model — because it isn’t any one of the three above. It’s a hybrid, and we’d rather describe it accurately than pretend otherwise.

MaxxLogix® sells Outcome Plans: bundles of outcomes, IDP pages, MaxxOwl™ SI Units, import sources, and storage, sized to the business. (Each MaxxOwl™ SI task carries a unit value based on its complexity — simple questions cost little, complex analysis costs more — deducted from SI Capacity when the task runs.) The entry plan starts with the foundation — no IDP, no SI — and higher plans add both. When a customer needs more of anything, they add capacity packs: Outcome Capacity, IDP Capacity (more pages), and MaxxOwl™ SI Capacity.

There’s a deliberate line in how the two halves count. When the document and its extracted data are used in a workflow that reaches a business result, that work also counts as one outcome. When IDP only extracts data for indexing and the file is stored, it uses IDP Pages but does not count as an outcome. An outcome is counted once, when a process reaches its result — the number of steps, approvals, or documents involved does not change the count.

So yes — there is a page component. IDP is part of an outcome process, and plans start with a page allowance. But here’s the differentiator, stated precisely:

A page is a page. Whether it’s a simple one-page document or a complex one requiring splitting, line-item data extraction, and summaries — the page costs the same. Customers can add SI capacity for IDP, which means more pages. At the unit level it breaks down to a price per page — but the price doesn’t change with the document’s complexity.

Contrast that with per-token billing, where a complex document costs more than a simple one by design — more tokens in, more tokens out, bigger bill. Under our model, the document that takes the most AI work to process costs you the same as the simplest one. The vendor doesn’t profit from your documents being difficult.

And one more structural difference: every plan includes unlimited users. No per-seat math, no license rationing, no IT gatekeeper deciding who gets access. When adding a user doesn’t raise the bill, adoption stops being a budget decision — and a platform people actually use is a platform that delivers outcomes.

This is the thesis behind MaxxLogix SI™ (Strategic Intelligence) — our AI-powered intelligence layer, within the broader Intelligent Outcome Engines™:

SI should be measured by business value. Not tokens.

What changes when the model is built around outcomes:

• Predictable spending. Plans bundle what the business needs; capacity packs scale it. The cost of a business result is known up front, so finance can plan and IT can govern.
• Complexity doesn’t cost extra. Simple or complex, a page is a page. The documents where AI delivers the most value don’t carry a surcharge.
• Adoption isn’t rationed. Unlimited users on every plan. The right model is the one where heavier use is a success story, not a budget incident.
• Incentives point the right way. The vendor wins when the product delivers valuable results efficiently — not when you burn tokens, add seats, or process difficult documents.

BUSINESS SI. SIMPLIFIED. Simple to buy. Easy to use. Built for outcomes. Measure what matters. Deliver what counts.

Why input-based pricing misaligns incentives

Step back and look at the pattern across the market’s three models. Per-page, per-seat, and per-token all measure the input while ignoring the result:

• Per page — pay for volume, whether the processing helped or not.
• Per seat — pay for access, whether those people got value or not.
• Per token — pay for the raw material of computation, whether the answer mattered or not.

In every case, the vendor’s revenue grows when your consumption grows — regardless of whether your outcomes improve. That’s a misalignment of incentives, stated plainly. The vendor profits when you’re inefficient: longer prompts, bigger documents, more retries, more seats. A pricing model should reward the vendor for making you successful. Input-based pricing rewards them for making you busy.

We’ve written at length about the token version of this problem — the per-token trap, why it punishes thoroughness, and what it really costs at scale. If you want the full math, start here: Why Tokens Are the Wrong Way to Price AI

A simple comparison

Model	How you’re charged	Predictable?	Who it favors	Watch out for
Per page / per document	Each page or document processed	Medium — volume is forecastable, mix isn’t	High-volume, uniform documents	Says nothing about accuracy or exceptions; success raises the bill
Per seat / per user	Each named user, per month	High — until headcount changes	Small teams with heavy users	Access ≠ results; adoption gets rationed
Consumption / token	Tokens, API calls, operations consumed	Low — spikes with real business activity	Vendors (scales with your usage)	Punishes thoroughness; complex documents cost more; hardest to budget
Outcome plans (MaxxLogix)	Business results (outcomes) + IDP pages + SI Units, bundled; capacity packs to scale	High — plan cost known up front	Buyers who want cost tied to results	New mental model; “outcome” must be crisply defined per use case (see above)

The honest footnote on our own row: there is a page component in IDP — plans include a page allowance and you can add more. The difference is what a page costs you: the same, whether the document is simple or complex. No complexity surcharge, no token meter running faster on the documents that matter most.

Questions to ask before you sign

Whatever vendor you’re evaluating — us included — these cut through every pricing model:

1. What happens to my bill when volume doubles? If the answer requires a spreadsheet and a prayer, the model is the problem. You should be able to state the cost of twice the outcomes in one sentence.
2. Does a complex document cost more than a simple one? Under token billing, it does — by design. Ask whether the price follows the document’s difficulty or the business result.
3. Are overages a profit center or a rounding error? How a vendor handles the months you exceed plan tells you what business they’re really in. If surprise usage is where the margin lives, the meter is the product.
4. What am I actually buying — inputs or results? Ask the vendor to describe what a unit of their pricing does for your business, not what it measures technically. If they can’t, you’re buying meter readings.
5. Who gets punished when we use it more? Under per-seat pricing, IT rations licenses. Under per-token pricing, finance rations usage. The right model is the one where heavier use is a success story, not a budget incident.

Keep going: Why Tokens Are the Wrong Way to Price AI · the full business case for private & hybrid AI