Get a demo

Extract email body & attachment data

Pull structured rows out of every inbound email and attachment. Watch the inbox, extract the fields you need with AI, and load the records into a clean table.

Get a demo

The prompt

I want to pull structured data out of incoming emails and their attachments — PDFs, Excel files — using AI. Can you build me a flow that monitors the inbox, extracts key fields from the email body and attachments with OpenAI, and loads the parsed records into a destination table?

Just copy and paste the prompt into a new Parabola flow to get started.

What Parabola builds

A workflow with six steps you can edit:

1. Watch the inbox. A shared inbox, a forwarding address, or a label-filtered Gmail view. The flow picks up every new email that matches the rule.

2. Split the body and the attachments. Read the message text. Pull every PDF, Excel, and CSV attached.

3. Extract the fields with AI. Pull the fields you care about from the body. Pull line-item rows from each PDF and spreadsheet. The AI step handles the varied formats senders use.

4. Validate the records. Check required fields, normalize dates, standardize codes, match against your master tables.

5. Land the records. Push the clean rows into a Parabola Table, a Google Sheet, your ERP, or wherever the downstream process picks up.

6. Route the exceptions. Anything the flow could not parse cleanly lands in a review queue with the source email attached.

Why teams stop doing this manually

The work nobody puts on their job description is the work that floods the shared inbox. Carrier confirmations, supplier price updates, customer order forms, retailer remittances, inbound shipping notices. They land in unstructured email and PDF, and somebody has to turn each one into a row before the ops team can act on it.

The manual version is a person in the inbox at the start of every shift. Read the email, open the PDF, copy the fields into the working sheet, file the email, move to the next one. It is fine work when the volume is twenty a day. It collapses at two hundred. The team starts batching. Batching turns into a Friday backlog. The Friday backlog turns into the reason the ops report is always running a day behind.

The judgment in the work is whether a record is real, complete, and properly coded. That part belongs to a person. The reading and typing belong to a flow. AI extraction has finally gotten reliable enough to do the typing without making the kind of mistakes that break the downstream report. Keep the human on the review, automate the rest.

How it works

Step 1. Paste the prompt.

Open Parabola, paste the prompt in section 2, and let it ask follow-up questions about which inbox to watch, which fields to extract, and where the clean records should land.

Step 2. Connect your data.

The inbox, the AI extraction step, your master tables for validation, and the destination table or system.

Step 3. Run it on every email.

New message arrives, the flow extracts, validates, lands. Exceptions land in the review queue with the source attached.

Related prompts

FAQ

Does this work for emails with multiple attachments?

Yes. The flow handles every attachment on the message. Each one runs through extraction. The clean rows land together with the email reference attached.

How accurate is the AI extraction?

Cleaner inputs get cleaner output, but the flow handles varied formats well in production. The validation step catches low-confidence rows and routes them to the review queue with the source attached.

What if the same sender uses a different layout each time?

The AI step is layout-flexible. It looks for the fields, not for a fixed template. New layouts handle without retraining. The exception queue catches anything that did not parse cleanly.

Can the flow write directly to my ERP or system of record?

Yes. The output can push records into NetSuite, your AP system, a Parabola Table, or a Google Sheet. Some teams prefer to land in a staging table and let a human approve the push.

How is this different from a Gmail filter and a manual paste?

A Gmail filter sorts the emails. The flow reads them, extracts the fields, validates against your master data, and lands clean rows. The work it removes is the reading and typing, not the sorting.
Stop reading the inbox. Read the table.
Paste the prompt, point it at your inbox, and let the extraction run on its own.
Start for free