Auto-save invoice PDFs to Google Drive from email and apps

Email is the easy half. The SaaS tools are the honest half

The brief said "various SaaS tools", and this is where I have to be straight with you. There are three kinds, and only two can be automated.

The first kind emails you the invoice as a PDF attachment every month. Those run through the same four tests. Nothing to build.

The second kind has a billing page and either an API or an export. For those the workflow runs a scheduled pull: sign in with an API key, list invoices since the last run, download each PDF, hand it to the same rename-and-upload path. Once a day is plenty. Some tools email a link instead of a file. If the link is a direct PDF, the workflow fetches it. If it needs a login, it drops into the third kind.

The third kind only shows a billing page behind a login, with no API and no export. Those stay a monthly manual download. I say so up front. The alternative is a scraper that logs in and clicks around, which breaks the first time the vendor redesigns the page and often breaks their terms of use. Instead I put a one-page list in the same Drive folder: tool, billing URL, the day of the month the invoice appears. Five minutes on the first of the month, with a checklist, instead of a hunt.

If you're choosing which automations to build first, this post walks through the order.

How the upload works, and how the file gets its name

Two facts about the Drive API shape the whole build.

First, a file goes into a folder by setting the folder's ID in the parents field. Leave parents empty and the file lands in the root of My Drive, which is how invoices get lost. And a file can only have one parent folder: specifying multiple parents isn't supported. So there's one invoices folder, and if you want a subfolder per year, the year folder is the one parent.

Second, Google offers three upload types. A simple upload sends a file of 5 MB or less with no metadata. A multipart upload sends a file of 5 MB or less together with its metadata in one request. A resumable upload is for files larger than 5 MB, or for a poor connection. Invoices are small and need a name and a parent, so this build uses multipart, with resumable as the fallback for the rare oversized scan.

The name is the part the bookkeeper cares about. The convention is date, vendor, amount, in that order, so the folder sorts itself by date. For example: 2026-08-31_slack_1240.00-chf.pdf. The date is the invoice date, not the day the email arrived. The vendor is lowercase, one word where possible. The amount carries its currency. Named like this, "where's the August Slack invoice" stops being a question anyone asks.

Duplicates happen more than you'd think. Vendors send the invoice, then a reminder with the same PDF. The daily SaaS pull overlaps with last week's. So the workflow keeps a small ledger: a hash of every PDF it uploads, plus the vendor and invoice number. Same hash again, skip it. Same vendor and invoice number but a different hash, which is what a corrected invoice looks like, and it files the new one with -v2 on the end. It never overwrites.

It files. It never pays and never deletes

This is the rule I'd refuse to build without.

The workflow has no access to a bank, a card, or a payments tool. Not because that would be hard to add. Because fake-invoice phishing exists: an email that looks exactly like your vendor's monthly invoice, with the IBAN changed. If the workflow could pay, the phishing email would only need to pass four tests to get paid. So it can't pay. It files, and a person approves.

It never deletes either. It doesn't archive the email, doesn't move files out of the folder, and its Drive access is scoped to that one folder. The worst thing it can do is file a PDF that isn't an invoice, which a person notices and fixes in ten seconds.

Here's what AI cannot do in this build. It can't tell you whether an invoice is legitimate. It can't tell that the bank details changed since last month. It can't tell whether you actually ordered the thing, or whether you already paid it. A model reading a PDF is a reader, not an auditor. Approval stays with a person.

Polling on a schedule, or Gmail push

There are two ways for the workflow to notice a new email.

Push means Gmail tells you: you call the API's watch method on the mailbox, notifications arrive through Cloud Pub/Sub with a history ID, and you fetch what changed. It's fast, and it has to be renewed. Google's guide says you must call watch at least once every 7 days or you stop receiving updates, and recommends calling it daily.

Polling means the workflow wakes up on a schedule, say every 30 minutes, searches the inbox for messages with PDF attachments since the last run, and processes what it finds. Slower. Also boring, which is what you want here. The same guide says that for user-owned devices the poll-based sync approach is still the recommended one.

For an invoice folder I poll. Nobody needs an invoice filed within a second, and a daily watch renewal is one more thing that can silently lapse, after which nobody notices for a fortnight. I made the same call for the twice-daily digest.

What it's worth, and what to do this week

Here's the estimate, and it stays an estimate until it has been measured over 30 days in use. Say you get 40 invoices a month. Say each one costs 3 minutes to find, download, rename and upload. 40 × 3 is 120 minutes, which is 2 hours a month, or about half an hour a week for one person. The month-end hunt adds more, but I'll only claim what I can count.

Half an hour a week isn't the reason to build it. The reason is that the folder is complete, named consistently and shared, so nobody hunts and the bookkeeper stops asking.

Here's what to do this week, before anyone builds anything:

1. Make the one folder in Drive and share it with your bookkeeper. One folder, because the API only allows one parent anyway.
2. Open last month's invoices and write down every sender address. That's your allowlist on day one.
3. Sort your SaaS tools into the three piles: emails a PDF, has an API or export, login only. The third pile is your monthly checklist.
4. Pick the naming convention and rename last month's files by hand, once, so the folder is consistent from the first day.

Then decide whether you build it yourself or I do. With me it's a fixed-price build plus a monthly retainer to run and maintain it, and what that looks like is on the implementation page. If you'd rather talk first, book a growth call. It's free, and I'll tell you if it isn't worth building. The Chimpy version is being built now, September 2026, and after a month in use I'll replace the estimate above with measured numbers.