DocCoreX

Usage limit best practices

Every DocCoreX plan —including the free one— includes a monthly pool of AI credits. Credits are the unit we use to measure the work AI does for you: reading and classifying a document, extracting its data, analyzing a case file or drafting a reply.

This guide explains what consumes credits, why consumption varies so much from one customer to another, and how to optimize it.

What AI credits are

Each plan includes a monthly allowance of credits that sets a fair-use limit. Under normal conditions, that allowance is more than enough to cover the plan’s usage: it exists to prevent anomalous spikes, not to limit your day-to-day operation.

  • The allowance renews every month.
  • If you exceed it, per-unit overage applies or you can upgrade. See the detail on the pricing page.
  • Every plan includes credits: consumption isn’t a hidden cost, it’s proportional to your usage.

What consumes credits

Every AI operation consumes credits, and a single document usually goes through several:

  • Reading a document (turning a PDF or image into text the AI can understand).
  • Classifying the document or email by your taxonomy.
  • Extracting its structured data.
  • Summarizing and validating.
  • Analyzing case files (claims, payment capacity) or generating replies and communications.

Each operation is one pass of the model. The more documents and emails you process —and the more operations each one requires— the more credits you consume.

Native vs. scanned PDF

This is the factor that weighs the most in your consumption. The difference between analyzing one and the other is huge:

  • Native PDF (with text). Generated by a system or an editor, so its text is already “readable” by software. DocCoreX reads it without consuming AI: reading is practically free and you only consume when classifying, extracting and summarizing.
  • Scanned PDF (image). A photo or scan of paper: it has no real text, just pixels. The AI has to “read” it page by page, in batches. A 30-page scanned document can cost more than 10 operations just to read it, before even classifying or extracting its data.

Rule of thumb: whenever you can, upload the native PDF (the digital original) instead of a scan. It’s the biggest credit-saving lever.

Complexity: certificate vs. contract

Reviewing a certificate is not the same as reviewing a contract, and the difference isn’t the “number of fields” but pages, length and whether it’s scanned:

  • A native 1-page certificate ≈ 3 light operations (classify, extract and summarize).
  • A 30-page scanned contract ≈ 10+ reading operations plus the analysis ones.

That’s why two customers on the same plan can consume very differently: a company that classifies complaint emails consumes far less than one reviewing scanned contracts or policies.

Other consumption factors

  • Number of pages and length: more pages and more text = more operations and more consumption per operation.
  • Full analysis: a document usually goes through several chained operations (read + classify + extract + summarize).
  • Long instructions and free text add consumption on every operation.
  • Repeating the same work: there are no automatic shortcuts; processing the same document again consumes again.

Reprocessing and categories

Keeping your categories well defined and up to date is the best way to control your consumption.

When a classification is wrong and an item has to be reclassified, DocCoreX re-runs the entire analysis process on that item and consumes credits again: what was already spent is spent again. Reprocessing and reclassification are the main source of avoidable consumption.

To minimize them:

  • Correct the type manually when you already know it: manual correction costs 0 credits, unlike reclassifying with AI. When you correct it, leave clear notes: from them the system itself will recommend changes and adjustments to your categories so classification becomes more and more effective.
  • Define clear, non-overlapping categories. If two categories are easily confused, errors and reclassifications go up.
  • Keep the taxonomy up to date. When your operation changes —new document types, new flows— update the categories before errors start.
  • Start with few categories and add only those that are justified, instead of starting with twenty that blur together.
  • Review the queue of ambiguous cases. Fixing a mis-classified pattern early prevents hundreds of reprocesses later.

How to optimize your consumption

  1. Upload native PDFs (with text), not scanned ones. By far the biggest saving lever.
  2. Pick the type on upload when you already know it: a manually typed document skips AI classification.
  3. Correct the type manually instead of reclassifying with AI (0 credits), and leave clear notes: the system uses them to suggest adjustments that improve classification.
  4. Turn on AI analysis only when you need it.
  5. Split very long documents and avoid merging several into one without need.
  6. Upload documents grouped and moderate auto-analysis when loading many at once.
  7. Watch the size of free-text instructions: very long instructions add consumption on every operation.

Renders are not AI credits

If you generate PDFs with Studio, those renders are counted separately (one render equals 2 pages) and don’t consume AI credits. See the detail in the Studio guide.

In short

  • Every plan includes AI credits with a fair-use limit.
  • Consumption depends mostly on whether your documents are native or scanned, their size, and the complexity of your processes.
  • Reclassifying reprocesses the whole analysis and consumes again; that’s why keeping your categories well tuned —and correcting manually when you already know the type— is the best way to control consumption.

Questions about your consumption or your plan? Write to [email protected] or check the pricing page.