Let's start with the good kind of problem: we can't pack fast enough.

Today alone we shipped close to 30 orders. There are still 25 sitting on the floor waiting to go out. And every single morning this week, my phone has been full of the same message from customers: "Where is my order?" "Can you ship today?" "I need it before the 20th."

None of this was a carefully planned launch. There was no ad budget, no marketing calendar. We asked AI what to make for the holiday season, made it, put it online — and it took off. Here's exactly what happened, including the part that went sideways.

The Old Way: Guessing in September

Christmas has a hard deadline. Unlike a canvas tote you can sell all year round, a Christmas bag is worth close to nothing on December 26th. That makes the selection decision brutally important: pick right and you sell out by mid-December; pick wrong and you're holding stock nobody wants until next year.

Traditionally, a small factory like ours decides this way:

• What sold last year? Safe, but you're always one season behind — and if everyone else did the same thing, you're fighting on price.
• What is the boss's gut feeling? Twenty years of instinct is real. It's also one person's opinion, and it can't be checked.
• What are the neighbours making? In Yiwu, you see what's moving in the market. By the time you see it, half the market is already on it.

The weakness isn't that these are bad signals. It's that they're never written down, never compared, and never checked against what we can actually produce.

What We Did Instead: Let AI Do the Boring Cross-Checking

So we tried something different. Not an AI that magically invents a product — an AI that does the tedious cross-checking we never had time for.

We fed it three things:

1. The season and the market — Christmas and New Year, the platforms and regions our customers sell into, the search terms and themes buyers actually use (snowman, snowflake, red-and-green classics, "gift bag that looks like decor", New Year stars-and-stripes).
2. What our factory can actually do — our base bag styles (canvas totes, drawstring pouches, gift bags), our printing methods, our realistic MOQs and lead times.
3. The constraint that matters most — the holiday deadline. Anything that can't be produced and shipped in time is not a candidate, no matter how pretty it is.

What came back wasn't a single genius idea. It was a ranked shortlist with reasons attached — and, more usefully, a few things it talked us out of. That's the part I didn't expect.

What We Actually Made

Four products went into production. All of them are now on our holiday products page — but one turned out to be the workhorse, and it isn't the one we spent the most time designing.

Christmas Stripe Gift Bag — our workhorse

One bag, four colourways: red, yellow, green and pink stripes on canvas, with a drawstring top. Nothing about it is clever. It is the plain, obviously-Christmas bag — and that is exactly its strength: a buyer can pick the stripe colour that matches their tree, their shop window or their own brand. We also run a larger 66×48×15 cm size for bulk orders.

This is the bag the order list is full of — red, yellow, green and pink stripes, in twos, fives, tens, and one order of a hundred. When you reach the screenshot further down, this is what you are looking at.

Red Christmas Snowman Tote Bag

The classic. Red ground, snowman front and centre, tree, stocking and stars around it. It's not clever — and that's the point. It reads as "Christmas" from three metres away, which is exactly what a shopper scanning a shelf or a phone screen needs.

Pink Christmas Snowman Tote Bag

The one I would have cut by instinct. A pink Christmas bag felt wrong to me — Christmas is red and green, isn't it? The reasoning was simple: pink is a colour lane almost nobody occupies at Christmas, so it doesn't compete head-on with the red ones, and it reaches a completely different buyer. It has been one of the strongest performers.

That single recommendation was worth the whole exercise. Twenty years of experience is exactly what told me pink was a mistake.

New Year Stars & Stripes Gift Bag

This one extends the season past Christmas: a drawstring pouch with stars on top, red and white stripes below, a bold red bow and "Happy New Year" across the front. Christmas bags die on December 26th. A New Year bag keeps selling into January.

Then It Sold. Faster Than We Could Make It.

We photographed them, listed them, and that was more or less it. Orders started coming in — not in a trickle, but in the shape you only see with seasonal goods: a slow first week, then a wall.

Right now the packing table is the bottleneck of the entire business. Shipped close to 30 today, 25 more waiting, and the message notifications have become part of the morning routine.

Our shipping back end mid-season. Every line is a paid order still waiting to be packed — red, yellow, green and pink striped canvas gift bags, in twos, fives, tens, and one order of a hundred.

The Bottleneck Nobody Puts in the Case Study

Everyone writes about how AI helped them pick a winning product. Almost nobody writes about the week after.

Here's what we got wrong: we planned the selection, not the capacity. We knew what to make. We did not reserve enough production slots, fabric, or packing hours for the case where it actually worked. When demand showed up, the constraint wasn't ideas or customers — it was hands on the packing table.

A seasonal product makes this worse, because you cannot solve it with time. A bag that misses the holiday window has lost most of its value. There is no "we'll catch up next month".

What We'd Do Differently Next Season

1. Decide earlier, and book capacity at the same time. Selection and production planning are one decision, not two. If AI says "this will sell", the next sentence has to be "so we need X metres of fabric and Y packing days reserved".
2. Trust the counter-intuitive pick — but test it small first. Pink snowman totes were right. We should have run a small first batch deliberately instead of discovering it after the wall of orders arrived.
3. Plan the end of the season, not just the start. Have the January product (New Year, Valentine's) ready while Christmas is still shipping. Otherwise the line stops the moment the holiday does.
4. Keep the human in the chair. AI widened the option set and caught my blind spot. It did not decide, and it should not. The judgement about our own machines, our own workers and our own customers is still ours.

Is this a huge business? No. It's thirty orders on a good day in a small factory in Yiwu. But it used to be that the hardest question of the season — what do we make? — was answered by one person's hunch in a noisy market. Now it gets checked, reasoned, and written down. And this time, it meant we spent our days packing orders instead of wondering whether anyone would want them.

---

This is Post 15 of our AI journey. If you're planning a holiday line and want a factory that can make canvas totes, drawstring pouches or gift bags to your own artwork — with honest MOQs and dates you can actually plan around — have a look at our holiday products or our custom process, and come say hello on the contact page.