How to test SideQuest on your real POs before you commit to a subscription

The free tier processes 20 POs per month, no credit card. Here is the playbook we send to evaluators. About an hour from "I downloaded the zip" to "I know if this works on my messiest customer's POs."

Most automation tools want you to subscribe first and trial second. That makes sense for the vendor and not the buyer. We do it the other way: free tier, full functionality, capped at 20 POs a month so you can actually evaluate on the POs that matter without burning a card.

Here is the exact sequence we suggest.

Step 1: Install against a sandbox QuickBooks first (15 minutes)

Set up a QuickBooks Online sandbox at developer.intuit.com — free, takes about three minutes. Use the install prompt at sidequestautomation.com/install-prompt and have Claude Desktop walk you through it. The sandbox catalog is generic (landscape services), but it's enough to prove the OAuth flow works end-to-end.

Don't connect to your real QuickBooks yet. Validating the plumbing in sandbox first means you find environment-level issues without polluting your real books.

Step 2: Process the demo PO (5 minutes)

Once installed, ask Claude in your terminal: "process the demo PO." SideQuest has a baked-in scanned PO that exercises every stage of the pipeline — handwriting recognition, fuzzy matching, price variance flagging. The whole demo runs against the sandbox without touching your data.

You'll see the draft Estimate appear in the sandbox QB at the URL Claude provides. Open it, confirm it looks right. If anything's off here, the rest of the test won't work — stop and ping us.

Step 3: Forward three real POs from three different customers (15 minutes)

Pick three POs from your inbox that represent the spread of what you receive:

1. A clean structured PO from a sophisticated buyer (their MRP system emits a typed PDF with predictable layout)
2. A messy scan from a smaller customer (handwriting, scribbles, signed)
3. A PO from a customer who uses their own internal part numbers that don't match your QB SKUs

Forward them to your sandbox Gmail with the purchase-orders label, or use the forward-PO trick to push them in directly. Then in Claude: "list incoming POs and draft estimates for all three."

Watch what comes back. The clean PO should be auto-matched at 0.95+ confidence with no review flags. The messy scan should populate most lines and flag the questionable ones (this is correct behaviour — you want lines flagged, not silently wrong). The custom-part-number PO will mostly be flagged as needs-review.

Step 4: Resolve the custom-part-number lines and watch the connector learn (10 minutes)

This is the moment of truth for the matcher. Take that third PO, the one with custom part numbers, and tell Claude: "set line L1 to [the QB item it should be]." Repeat for each unmatched line.

Then ask: "list the cross-reference rules SideQuest has learned." You'll see the connector wrote each mapping you just made into cross_reference.csv. Forward another PO from the same customer with the same part numbers. This time the matcher resolves them all at 0.99 confidence without you doing anything.

That moment — the second PO from a difficult customer auto-matching — is the actual value prop. If it doesn't happen for you, the rest of the trial isn't going to convince you.

Step 5: Connect production QB, run for a week (free tier covers it)

Once you're convinced the sandbox flow works, re-run the OAuth flow against your real QuickBooks (about 90 seconds). The free tier gives you 20 POs a month, which is roughly a week's worth for a single rep at a small distributor. Process your real POs for a few days.

The signal you're looking for: how long does it take a rep to process a PO through SideQuest versus typing it from scratch? If the average is 3 minutes vs 12 minutes, the ROI math is obvious. If the average is 9 minutes vs 12 minutes, the math is harder and you should tell us — we want to know.

If something breaks during the trial

The first day of any new tool has friction. Run the diagnose prompt in Claude when anything looks wrong. It walks the most common failure modes in plain English and points at the fix. If diagnose can't resolve it, email [email protected] — same-business-day response, even on the free tier.