Before automating Amazon Vendor Central purchase orders, I had the client record the whole process

On a Thursday in January I was on a call with a manufacturer that sells its products wholesale to Amazon. Every Monday, Amazon sends them purchase orders. Every Monday, one person on their team turns those orders into shipments by hand.

We'd already done the first piece of work on that call. I registered them for Amazon's Selling Partner API, the developer side of the same Amazon Vendor Central screens they work in every week, and got us set up in the sandbox. Then I asked a question I ask on almost every automation project: what does the whole process look like?

The answer was a screen share I won't forget.

Why record the whole process instead of describing it?

Because the person doing it describes the steps they think about, and the recording shows the steps they've stopped noticing.

Here's what I'd seen before that call: a recording of part of the process and a task in their ClickUp called the weekly Amazon PO processing and shipping workflow. That was enough to see pieces we could automate. It wasn't enough to see the whole thing.

I said it plainly on the call. "There are certain parts that just by looking at it, we can automate portions of it. But it's not necessarily end to end. So you would have to do certain parts of it manually."

Then the other half. "If we know the end to end process, we can go, this is how they're doing X, Y and Z. Maybe they're not using the right solution. Or here's a tool that we can put in place, or we can integrate everything together so that we can automate this thing end to end."

That's the trade. Automate what you can see, and you get a faster version of a few steps with a person still stitching the rest together. Watch the whole thing first, and you can decide what the process should be before you build it.

What did the recording show that a list of steps wouldn't?

It showed the same data being copied by hand into four places, a hard Monday deadline, and errors that came from the handoffs between steps.

The client walked me through it live, and I asked if I could record it. This is the Monday, as it was shown to me:

1. 1.Sign in to Vendor Central, open the POs that need attention, and confirm or deny each one against inventory.
2. 2.Download a report called Recent Changes.
3. 3.Open the PO import template, save a copy with the new date "so I don't mess up the template," and paste the report in.
4. 4.Check the chart the template builds for boxes, weight, volume and pallets, and fix anything that's off by hand.
5. 5.Paste the totals into a Gmail chat so everybody knows what Amazon ordered that week.
6. 6.Duplicate last week's Google Sheet, clear it, and paste the same data in again so the team can highlight progress and check off what's been invoiced.
7. 7.Hand it to an outside team to create the shipments. One shipment per PO, and two products on the same PO have to land in the same shipment.
8. 8.Fill in shipment details for each one: pickup date, stacked pallets, cartons, total weight, total volume.
9. 9.Build carton labels PO by PO, combine the files and print them.

Then on Wednesday or Thursday, print the shipping labels, and then two different people invoice it, once on their side and once on Amazon's.

That week there were 24 POs, which meant 24 shipments. It was the slow season. "This was 24," the client said. "We can get 70." And one shipment can be nine pallets.

Then the part that changed how I thought about the build. The shipments and carton labels have to be created in Amazon's system on Monday, "or they can cancel the order."

A step list tells you there's a template. The recording tells you why there's a template: someone left, and the process landed on a person who already had a full-time job. "This has been the method that was passed on to me," the client said. And the outside team doing step seven "did good the first time, and it's just gotten progressively worse."

None of that is a technical detail. All of it decides what you build.

Why set up the Amazon Vendor Central API in a sandbox first?

Because the real system has a deadline that cancels orders, and you don't learn an API on live orders with that hanging over you.

When I set up the app, I picked sandbox credentials. The sandbox runs on made-up data. "We do not want that to mess up at all," I said. I also added our developer as a user on their account so the team could work without asking the client for a login code every time.

That's the order I want on any build with a hard deadline. Get access, learn the API against fake orders, and keep the client's real Monday safe until we know exactly what each call does.

What should you ask for before an automation agency starts building?

Ask them to watch the whole process first, and to tell you what they'll leave manual.

A lot of business automation gets sold as a list of steps someone will take off your plate. Whether you hire an AI automation agency or build it in-house, the first thing you should hand over is a recording of the real process, not a summary of it.

We agreed the client would record the next full Monday. "I'm not editing anything," the client said. The plan was simple: we get the video Tuesday, I review it Wednesday, and by the end of the next week we'd know how we were going to automate it.

The client also said the line every automation project should start with: "If we can take the human errors out, that would be great." And then, to be clear about it: "Don't worry, you're not getting me out of a job." It wasn't. The point was to give that time back to someone who never had it to begin with.

It's the same instinct as having the client's staff test the app before I write automated tests. Find out what's true before you build around it.

What this means for your business

If you're about to automate anything, start with one unedited recording of the person who does it. Have them share their screen and talk out loud while they work through a real week, not a demo. Ask them to say what they're checking and why, every time they stop to look at something.

Then watch for three things. Count how many places the same data gets copied into. In this case it was a template, a Gmail chat and a Google Sheet, on top of Amazon itself, which is the same problem I wrote about in signs your business has outgrown its spreadsheet. Write down every deadline that costs you money if it's missed. And mark every handoff, because that's where the mistakes come from.

This is why Skaiward builds around your client data living in one place. When the orders, the shipments and the invoices are all in the same records, your people and your AI agents work from one version of the week instead of four copies of it.