Gmail to QuickBooks: automating inbound POs without Zapier

A wholesale plumbing distributor processes 40 purchase orders a day. They come in through Gmail. Customer sales reps copy the line items into QuickBooks Online estimates by hand. The team has tried Zapier, looked at Conexiom, and asked their ProAdvisor about a custom integration. None of it landed. Here's the workflow that actually ships in 2026.

Why the obvious tools don't fit this job

The PO-from-email problem looks like a straight Zapier use case for about ten seconds. Trigger: new email in Gmail with a specific label. Action: create estimate in QuickBooks Online. Done.

The reason it falls apart is the middle. The trigger is fine. The action is fine. The middle is a customer email with seven line items, three of which use the customer's part numbers instead of yours, one of which mentions a 1/2" elbow that maps to your BR-ELB-050-NPT SKU, and a PDF attachment that adds two more lines you'd miss if you only read the body. Zapier's data layer has no concept of any of that. You'd end up either creating empty estimates that someone has to fill in by hand (no savings) or building a 40-step Zap that breaks every time a customer changes their PO template.

Conexiom and Esker handle the parsing better but they price for enterprises and they need an implementation project. We did the comparison in detail. A 4-person distributor doing 600 POs a month doesn't have $800 a month or a 6-week implementation in them. They have a Tuesday morning.

What the end-to-end workflow looks like

Here's the deterministic version, the one that runs in the background without anyone touching it.

1. Gmail filter applies a purchase-orders label to every PO that lands. Filter is based on sender domains and a few keyword rules. Takes 5 minutes to set up; that's the only Gmail config you do.
2. SideQuest connector, running on your Mac or Windows computer, reads the labeled inbox every morning. The connector is local; your PO content never leaves your machine.
3. Parser pulls the buyer name, ship-to, PO reference, terms, need-by date, and line items from the email body and any PDF attachments. Five recognized line shapes — tabular tables, marker rows (3 × PART-001), bulleted lists, free-form narrative, and the industrial "qty UOM PN" shape. Try it free here with any PO you have on hand.
4. Matcher compares every parsed line against your QuickBooks Online item catalog. Exact part number wins. Customer-specific cross-reference comes next (your buyer's PN ACME-12X4 resolves to your BR-ELB-050-NPT). Description-fuzzy match is the last resort, and it's always flagged for human review.
5. Draft Estimate gets built locally with all the matched lines, prices pulled from your QuickBooks catalog, and any flagged-for-review items called out. The draft does not write to QuickBooks until you click submit.
6. You review in 15-45 seconds per PO. Anything clean clears in one click. Anything flagged shows you exactly why (unmatched part number, price variance over threshold, description-only match) so you can fix it before submitting.
7. Submit writes the Estimate to QuickBooks Online via the official Intuit API. The customer gets the same Estimate document they would have received if you typed it in by hand. They never see the seams.

The morning routine in practice

Eight AM Tuesday. Your overnight queue has 23 POs in it. You open Claude Desktop on your Mac and type one sentence: "process the overnight queue."

SideQuest runs through every email with the purchase-orders label. Each one becomes a draft in your local review queue. The 19 that match every line cleanly stack into one "clean" batch. The 4 with issues — one PDF that scanned poorly, two with customer part numbers we haven't seen before, one with a price 12% below your catalog — sit in a separate "needs-review" batch with the specific reason on each line.

You bulk-submit the 19 clean drafts with one command. That's 19 Estimates written to QuickBooks in about 30 seconds. The 4 needs-review ones you click through one at a time. Two are quick fixes (you confirm the cross-reference, SideQuest learns it for next time). One needs a quick reply to the customer to confirm the price. One is the scanned PDF, which you open and type in the seven line items yourself. Total time: under 20 minutes for 23 POs.

For comparison: at the manual-entry baseline of 4 minutes per PO, 23 POs is 92 minutes of typing.

What "human review" really means

Every draft pauses for you before it writes to QuickBooks. That's the default. You can opt into auto-submit later (it's a single environment variable) if you want clean drafts to flow through without a click, but the connector ships locked. The clean-gate has to pass first: every line matched, no price variance above your tolerance, customer found in QBO, no needs-review flags. Anything short of that pauses for a human.

This is the part most automation pitches skip. AP-side software (Bill.com, Tipalti) handles invoices going out — you trust the vendor enough to send them money. Inbound POs coming in are higher stakes the other direction. A wrong price on an Estimate becomes a wrong invoice, becomes a customer dispute, becomes an aging receivable. SideQuest's bet is that 40 seconds of human review per PO is the right unit of trust. It's still 5x faster than typing it from scratch.

Setup, end to end

The hardest part of the setup is the 20-minute mental shift from "this should be a SaaS in a browser" to "this is software that runs on the same computer as your QuickBooks Web Connector or your QBO browser." Once you accept that, the install is straightforward.

• 10 minutes to install the connector (Mac or Windows installer)
• 5 minutes to authorize Gmail (one OAuth click-through)
• 5 minutes to authorize QuickBooks Online (one OAuth click-through)
• 5 minutes to set up the Gmail label/filter for incoming POs
• 5 minutes to process the first real PO end to end

The detailed walkthrough is in the quick-start guide. About 30 minutes total. You're processing real POs the same morning you install.

What if a PO doesn't parse cleanly?

This is the question every distributor asks first, because everyone has at least one customer whose PO format is its own snowflake. Three things happen, in order:

First, the parser falls through to Claude (the AI model running in Claude Desktop) for interpretation. The model reads the email body and any PDF attachments natively and writes structured line items the connector can match against your catalog. This handles the narrative shapes — "Hi Mike, send 30 of those 4-inch valves" — that no deterministic parser will ever cover.

Second, the draft is built from whatever the parser plus Claude came up with, then flagged for review with the reason. You don't lose the PO. You see exactly what got extracted, exactly what didn't, and you fix it in the draft UI.

Third, SideQuest learns the pattern. The next PO from the same customer with the same shape parses cleanly the first time. The cross-reference auto-learning kicks in: when you fix a customer's part number to your SKU in a draft, that mapping gets stored. Over four to six weeks, your "needs review" rate drops as the cross-reference table fills out.

The pricing math vs. typing it yourself

The honest comparison isn't to enterprise AP automation. It's to your current state, which is your team typing POs. At 4 minutes per PO and a $25/hour blended labor cost, every PO costs you $1.67 in pure typing time. A distributor doing 600 POs a month is spending $1,000/month on PO entry alone. SideQuest's Growth tier is $99/month for 750 POs. That's a $900 monthly delta in your favor, plus the time goes to actual customer work instead of data entry.

Run your own numbers on the PO time calculator. It uses your volume, your hourly cost, and the realistic time-per-PO range we see across our customer base.

Try the parser on your own PO

Before installing anything, paste a real PO email into the parser playground. It runs in your browser; the email content never leaves the page. You'll see the buyer, ship-to, line items, and confidence scores in real time — the same parser that runs in production. If it reads your customer's POs well, the rest of the workflow is just the QuickBooks plumbing.