The 33% nobody mentions
You have a 24 MB PDF. Gmail says 25 MB. You attach it, you send it, and it fails, with a message that either says nothing useful or nothing at all. Then you compress it to 20 MB and it fails again.
The reason is a design decision from 1992. Email was built to carry plain text, and it still is. Anything that is not text, a PDF, a photograph, a spreadsheet, has to be translated into text before it can travel, using an encoding called base64. That encoding represents every three bytes of your file as four characters.
Four divided by three is 1.33. The encoded text is then broken into lines of 76 characters, each with a line break, and the message headers come on top. Every attachment you send is well over a third bigger than the file on your disk, and the limit is applied to the bigger number. Since services rarely say whether their MB is 1,000,000 or 1,048,576 bytes, the figures below take the smaller.
| Service | Stated limit | Largest file that actually fits |
|---|---|---|
| Gmail | 25 MB | ≈ 17.4 MB |
| Yahoo Mail | 25 MB | ≈ 17.4 MB |
| Outlook.com | 20 MB | ≈ 13.9 MB |
| iCloud Mail | 20 MB | ≈ 13.9 MB |
| Outlook / Exchange (common default) | 10 MB | ≈ 6.96 MB |
This is why "compress it to just under the limit" keeps failing. Just under the limit is still over the limit once encoded. The target you actually need is the limit, less the headers, divided by about 1.37, and that is the number the tool above compresses to.
Compress or split?
Both work. Which one is right depends on what is making the file large, and it is worth thinking about for ten seconds rather than reaching for compression by reflex.
Compress when the size comes from images. A scanned document, a report full of photographs, a presentation exported to PDF, these contain genuine redundancy and often shrink by 70% or more with no visible difference. Your recipient gets one file.
Split when quality is the point. A signed contract, a technical drawing, a design portfolio, a medical scan a consultant will actually examine. Splitting copies the pages across untouched, so nothing is degraded at all. The cost is several emails and a recipient who has to keep track of them.
There is a third option worth naming: a text-heavy PDF that is enormous is often enormous for a fixable reason, an embedded font subset gone wrong, or a scan of text that should have been text. The PDF compressor reports which of those it found.
Why splitting by page count does not work
The obvious approach, 100 pages, four parts, 25 pages each, produces parts of wildly different sizes, and often at least one that is still too big.
PDFs do not store pages independently. A font is embedded once and referenced by every page that uses it. A logo appearing on all 100 pages is stored once. A single scanned photograph on page 60 might be larger than the other 99 pages combined. File size and page count are only loosely related, and for scanned documents they are barely related at all.
So this tool does not guess. It adds a page, rebuilds the document, and measures it, repeating until adding the next page would break the limit, at which point it closes that part and starts the next. Slower, and correct: every part it produces is genuinely under the limit, and it tells you the encoded size of each one so you can see that for yourself.
The alternatives, honestly
If the file is very large, a video, a big scanned bundle, a design archive, attachments are the wrong mechanism and no amount of compression will change that.
Gmail and Outlook both offer to upload to Drive or OneDrive and send a link instead. That works well, with two caveats worth knowing: the recipient needs permission to open it, which is a common failure with external addresses, and the link keeps working after the email is deleted, which for a confidential document may not be what you want. A file-transfer service has the same trade-offs plus a third party holding your document.
For anything sensitive, splitting is often the least bad option: the document goes only to the recipient, leaves no copy anywhere, and needs no permissions. That is the case this tool is built for.
Before you send
Two things worth doing to the file first. If it is going to someone outside your organisation, the metadata remover will strip the author name, the software that produced it and the creation timestamps, none of which are visible on the page, and all of which travel with the file. And if you are sending a draft, the watermark tool will mark it as one, which is considerably harder to argue with later than a line in the covering email.