Short answer
Two things, and only one of them is a price. The hours somebody spends reading and marking the documents, and the pages of the bundle you release. This guide says how to count both before you commit to either.
What a page means on an invoice
Answering an access request costs two different kinds of thing, and they are worth separating before anything is added up. One is human time, which is the part that has always existed. The other is whatever the tooling charges, which for this product is the pages of the bundle you release, at the rates the published plans state.
A billable page is one page of the output bundle Veil generates for you. Pages are counted from the generated output at the moment you release a case, and never from the file you uploaded. A document that you do not release is never billed, and a document the verifier withholds is never billed. An email attachment, and a file inside an archive, is its own document with its own pages. Every document counts as at least two pages. A spreadsheet counts as at most fifty pages, however many rows it holds.
That paragraph is the whole of the meter. It is one string in this repository, rendered on the pricing page and carried verbatim in the terms, so the definition a contract holds you to and the definition a marketing page states are the same sentence rather than two descriptions that agree today.
The three things that move the number most
Most of an estimate is ordinary arithmetic. Three rules account for nearly all of the difference between what people expect and what a case actually counts, and none of them is hidden.
An upload is not a document
An email with three attachments is four documents, and an archive is every file inside it, each with its own pages. This is not a billing trick: an attachment has to be extracted, detected over, rendered and verified exactly as its parent does, and it is the one the third party's name is usually in. Count what expansion produces, not what you dragged in.
Every document counts as at least two pages
A one line covering note and a forty page report are not the same work, but they are not proportional either: the fixed cost of a document is real and the floor is where it is charged. The consequence for an estimate is worth stating plainly. A case made mostly of short letters and one page emails bills close to two pages for every page the bundle actually draws, and a case of long reports bills about one for one.
A spreadsheet counts as at most 50 pages
A workbook laid out as readable rows becomes a great many pages, and those pages are an artifact of printing a grid as prose rather than something you asked for. So a spreadsheet stops adding to the bill at 50 pages however many rows it holds. The bundle still contains every page that was drawn; the meter simply stops counting.
None of the three is a judgement call, so all three can be applied to a case that has not been uploaded yet. That is what the next section is.
How to estimate a case before you open it
Five steps on a case sitting in a folder. It needs no account and no upload, and it gives the same number the meter will.
Count documents, not uploads
Expand every archive and open every email. A message with attachments is the message plus one document per attachment, recursively. This is usually where an estimate made by counting files in a folder goes wrong, and it goes wrong in the direction of being too low.
Take each document's page count
Use the pages the document actually has. Do not try to guess what the regenerated version will run to: for ordinary prose the two are close, and the sections below say where they are not.
Round anything under two pages up to two
This is the floor, applied per document rather than across the case. On a bundle of correspondence it is the single biggest term in the estimate, which is why it is a step of its own rather than a footnote.
Cap each spreadsheet at 50
Per workbook, not per case, and it only applies to spreadsheets. A large workbook that would print as hundreds of pages counts as the cap and nothing more.
Add it up, then price it twice
The total is the number the plans charge for. Put the same case into the cost model with your own reading speed and your own hourly rate beside it, and you have both halves of the answer rather than one of them.
The arithmetic is worth doing by hand once, because it is short enough to check. After that, the cost model on the Desk does the same steps over your own inputs, and it reads its rates from the same module the pricing page renders, so it cannot quote a price the price list does not.
What the same case costs done by hand
This site publishes no market rate for manual review, and the omission is deliberate rather than coy. There is no honest single figure: an hour of a person qualified to decide whose personal data stays in a disclosure costs one thing at a law firm and another inside a public body, and a number invented to make a comparison look good is the kind of figure a marketing page is not allowed to print.
What can be stated is the shape. The manual cost is pages, multiplied by the minutes one page takes to read, mark up and check, at whatever an hour of that person costs you, plus however many of those pages need a second read. Three of those four are yours, which is why the cost model takes them as inputs and why this guide does not assume them.
The comparison worth making is not really about money in any case. The two methods, question by question sets out what else changes: how many times one person gets decided about, who checks the result, and what happens at the point of doubt. Cost follows from those rather than the other way round.
What an estimate made now cannot tell you
The arithmetic above is exact and the inputs are yours, and the gap between those two sentences is where an estimate goes wrong. Four things it does not cover, in the order they are likely to matter.
- The bundle is generated, not edited. Every document you receive is laid out afresh from the text that was extracted, so its page count is a property of the output and not of the file you uploaded. For ordinary prose the two are close. For a dense spreadsheet, a slide deck or a scan they are not, and the direction differs per format.
- There is no typical case published here, because there is not an honest one to publish. The corpus this product is measured against is built of short documents, which is the right shape for testing how much a case can handle and the wrong shape for predicting how many pages one draws. Rather than average a figure out of it and print that, this guide gives you the arithmetic and leaves the distribution to your own folder.
- A document you do not release is never billed, and neither is one the independent check withholds rather than clears. Both are counted in the quality report, so a case where the number came in under the estimate says why.
- Nothing here is a storage charge or a seat charge. The material is encrypted under a key belonging to that case alone, it is purged on the schedule you set, and neither the holding nor the purging appears on an invoice.
One more thing worth separating out, because it is the cost people forget. Deciding what comes out at all is a legal judgement rather than an arithmetic one, and it is made before any of this applies. What a disclosure keeps and what comes out is the guide for that part, and it is the one to read first if the request is still open on your desk.
Questions
Am I billed for documents I decide not to release?
No. Pages are counted from the bundle at the moment you release a case, so a document you withhold and a document the independent check refuses to clear are both outside the count. The quality report in the bundle records both, which is what lets you reconcile a bill that came in lower than the estimate.
Does a scanned page cost more than a typed one?
Not on the meter. A scan is read before it can be redacted, and that reading is part of what a page costs us rather than a line on your invoice. What a scan does change is the output: the text is reproduced and the original page imagery is not, with a note in the bundle saying so.
The same attachment appears in twenty emails. Is it billed twenty times?
It is extracted once, because identical content is recognised as identical, and it is then rendered and counted as a document wherever it appears. That is deliberate: each copy is a document in the disclosure with its own place in the thread, and dropping nineteen of them would hand back a bundle that does not match what was held.
Why is the bundle longer than the files I sent?
Because every document is regenerated rather than marked up, so the layout is the output's rather than the input's, and because short documents count at the floor. A folder of one page emails is the case where the difference is largest, and it is also the case the floor exists for.
Are there charges per user?
No. The product is priced on pages released, so adding a colleague who reviews a case, or an auditor who only reads the log, costs nothing. The plans differ on volume and on features rather than on how many people you put on them.
If you would rather see the output before costing it, the sample bundle is a complete disclosure, manifest and quality report included, produced by a real run on invented people. Counting its pages is the cheapest way to check every rule on this page.