Brickweb

02/09/2026

What we got wrong building it

We’ve spent a good while building AI into The Brick Factory, and since I’ve spent five posts telling people how to judge other suppliers’ claims, it’s only fair to say what didn’t work in ours.

The first version of the quote was too clever. It tried to determine pricing from patterns in historical jobs. Impressive when it was right and quietly wrong in ways nobody could explain, which is worse than useless in a business where somebody has to defend a number to a customer. We rebuilt it to work from explicit rules the customer sets, with history used only to fill in what it already knows about that account. Less clever. Considerably more useful.

We underestimated how bad artwork actually is. We built the checker against the faults we expected. Then we pointed it at real files from real customers and found things we hadn’t imagined, including a logo somebody had scanned from a printed business card. The checker got a lot more paranoid.

We tried to automate approvals and had to pull it back. Routing a job automatically is fine. Approving one on somebody’s behalf is not, and the first time it went wrong it cost a customer a morning. Approvals go to a person now, faster than before, but to a person.
And the honest one. Early on we described this work in the way everyone describes it, as intelligence and automation, and it landed badly with exactly the people it was built for. Printers don’t want intelligence. They want forty minutes back on a quote and the artwork fault caught before the press. Same product, and we were selling the wrong half of it.
None of that is in a brochure, which is rather the point. Anyone who tells you their AI went in cleanly the first time either hasn’t shipped it or isn’t telling you about the bit that didn’t.

Ask us what went wrong. We’ll tell you, and then you can judge the rest of it properly.
30 minutes. www.brickfactory.uk

02/09/2026

Your job history is worth more than you think

Every decorator at volume is sitting on something they’ve never valued.

Ten years of job history. What every customer ordered, how often, at what quantity, on what garment, with what decoration, at what price, and whether it went out on time. Where it exists at all it’s usually spread across a filing cabinet, an inbox and somebody’s memory.

That’s an asset, and it’s about to become a considerably more valuable one, because it’s the only thing that makes any of this AI work worth having. Software that reads your history can tell you which accounts are drifting before they leave, which job types quietly lose money, and what November is going to look like while there’s still time to do something about it.

Software with no history to read can tell you nothing, however clever it is.
Which brings up something worth careful thought.
If your orders come through a marketplace, the history isn’t yours. The customer relationship isn’t yours, the ordering patterns aren’t yours, and the data that would tell you all of the above sits on somebody else’s servers and informs somebody else’s decisions.

You’re doing the printing, and they’re learning the business.
That was a tolerable trade when the alternative was no orders. It’s a much worse trade in a period where operational data is the thing that makes automation possible, because you’re handing over the raw material at exactly the moment it starts being worth something.

Own the storefront, and you own the list, the pricing, the relationship and the history. Slower to build. A considerably better business in five years.

That’s why we sell The Brick Factory as owned infrastructure rather than a fee on every order. Your growth is yours, and so is everything the system learns along the way.
30 minutes if you want to see what that looks like. www.brickfactory.uk

Print-on-Demand Software for Merch Producers

02/09/2026

Will it replace my staff?

Somebody asked me this on a call last month, and I think most people are too polite to.
The honest answer is no, and not because I’m being reassuring. It’s because of what actually gets automated.
The jobs that go are the typing. Copying an order from an email into a system. Opening artwork to resize it and save it under the right name. Retyping an address onto a courier label. Placing the same supplier order every Tuesday. Nobody in your office enjoys any of that, and nobody was hired to do it.

What doesn’t go is everything that needed a person in the first place. The customer who rings up wanting something that isn’t on the website. The job that has to be moved because a delivery slipped. The account that needs handling carefully because they’ve had a bad month. Chasing artwork out of a client who keeps sending phone photos.
So the office doesn’t get smaller. It gets to do the part that was always being squeezed out.

There is a real version of this worry, though, and it’s worth naming rather than pretending it away.
It isn’t that a printer replaces two office staff with software. It’s that a printer who automates the typing takes on 40% more orders without adding anybody, and a printer down the road who doesn’t has to add somebody to grow at all. Over three years that’s a serious difference in what each business can charge and still make money.
That’s the actual competitive pressure. Not redundancies. Capacity.

And if you’re wondering how your team will take it, the honest answer from the ones I’ve worked with is that the office staff are usually ahead of the owner on this. They’ve been asking for it for a year. They’re the ones doing the typing.
Has anyone here been through it? Curious how it landed with the team.
www.brickfactory.uk

Print-on-Demand Software for Merch Producers

02/09/2026

AI can’t automate a process you’ve never written down

Here’s the conversation I’ve had four times this year.
A decorator wants quoting automated. Fair enough, it’s the right thing to want. So we ask a straightforward question: what are your rules?
And there aren’t any. Not really. There’s a rate card from 2023, a discount structure that lives in the owner’s head, a handful of customers on special terms nobody has written down, and an unspoken understanding that Dave in the office knocks a bit off for anyone who’s been decent about lead times.

That isn’t a criticism; it’s how most businesses this size actually work, and it works because Dave is good at his job.
But it can’t be automated, and no software on earth changes that. Automating an inconsistent process just gets you to the wrong number faster.

So before you buy anything, spend a morning writing down what you actually do.
Quantity breaks. The real ones, not the published ones.
Decoration method and position count. What each one genuinely costs you in time, including setup, not what the rate card says.
Customer tiers. Who gets what, and why. This is the awkward one, because most businesses find they’ve got customers on terms nobody can justify any more.
When you say no. The job types you’d rather not take. Writing that down is worth more than any software.
Two things happen when you do this. The first is that you find the leaks, and there are always leaks. Every decorator I’ve done this with has found at least one job type they’ve been quietly subsidising for years.
The second is that you end up with something a machine can actually work with, whoever you eventually buy from, including nobody.

That’s the honest version of getting ready for AI. It isn’t a purchase. It’s a morning with a notepad and an uncomfortable conversation about what you really charge.
Anyone done this and found something that surprised them?

www.brickfactory.uk

01/09/2026

What AI is genuinely good at in a print business, and what it isn’t

Worth being specific about this rather than talking about it in the abstract, because the abstract version is what everybody’s already sick of hearing.

Three things it does properly.
Drafting quotes. Not deciding your prices, drafting the quote from rules you set and history you own. The value isn’t speed; it’s consistency. Three people quoting the same job three ways is a bigger problem than any of them being slow.

Checking artwork. Resolution, colour mode, size against print area, fonts, transparency, missing bleed. Same six faults, over and over, and every one detectable the moment a file lands rather than when it’s on the press. This is the least glamorous and the most valuable.

Reading your own history. Which customers slipped last November. Which job types always overrun. Which decoration method quietly loses money at low quantity. The answers are already in your data, and almost nobody has the time to go look.

Three things it is not good at, and be wary of anyone claiming otherwise.

Judgement on a difficult customer. Whether to eat a mistake to keep an account is not a data problem. It’s thirty years of knowing people.

www.brickfactory.uk

Colour. Anyone promising AI colour matching across substrates and decoration methods is selling you a demonstration. Your operator’s eye and a proper proof still win.

Anything requiring it to be right every time. These systems are probabilistic. Used where a human checks the output, that’s fine and useful. Used where nobody checks, it will eventually cost you a run.

The pattern is simple enough. It’s good at the repetitive work leading up to a decision and poor at the decision itself. Which, when you think about it, is exactly what you’d want from a new starter in the office.

01/09/2026

One question to ask anyone selling you AI

Every supplier in this trade will show you AI over the next 12 months. Machine manufacturers, software companies, ink suppliers, all of them.

Here’s one question that sorts the real from the demonstration, and it takes four seconds to ask.

What is it reading?

That’s it. If the answer isn’t your order history, pricing, or artwork, it isn’t doing anything for your business. It’s doing something clever with somebody else’s data, and you’ll be paying for the privilege of watching it.

Three follow-ups worth having ready, because the first answer is usually vague.
Where does it get my pricing from? If a quoting tool can’t name the source of your rates, it’s guessing, and a guessed quote is worse than no quote. You’ll win jobs you shouldn’t and lose ones you should have had.
What happens when it’s wrong? Every one of these systems is wrong sometimes. The question is whether you find out before or after the garment is printed. Any supplier who tells you it’s never wrong is telling you they haven’t run it at volume.

Who checks it? If the answer is nobody, walk away. If the answer is your team, then ask how long the checking takes, because a tool that saves you twenty minutes and costs you fifteen in review isn’t a tool; it’s a hobby.

None of this is anti-AI. I’m building it into our own product, and I’d be daft to argue against it. It’s that the gap between an impressive demonstration and something that works on a Tuesday in November is enormous, and the only way to see the gap from the buying side is to ask what’s underneath.

Take those four questions to the next trade show. You’ll be surprised how quickly the conversation changes.

What’s the most impressive thing you’ve been shown that turned out to do nothing? No names needed.

www.brickfactory.uk

26/08/2026

Dolly Parton, 1946 to 2026. Underestimated for sixty years, and right about nearly everything.

God bless Dolly, a true legend.

26/08/2026

Dolly Parton, 1946 to 2026. She said more useful things about work and worth than most business books manage.

God bless Dolly, a true legend.

26/08/2026

What you can and cannot print when a celebrity dies https://www.brickweb.co.uk/what-you-can-and-cannot-print-when-a-celebrity

Tribute merchandise is one of the fastest demand spikes in print on demand. It is also one of the quickest routes to a cease and desist. Here is where the line actually sits.