Layers That Learn

Reflections on an iterative design process for physical product

“Have you got any more of that other yellow?” asked Pete.

I thought for a moment. In the early days of Artefact Cards there was the occasional variation in tones, as hands and eyes worked out what they were looking for, and the yellow ink would fade off the litho printing machine down in Axminster.

Not anymore, though. Yet perhaps though there were a few old rogue packs still rattling around, and Pete had got a hold of one.

“Do you mean a shade or two different?” I replied, “just the barest change in colour?”

“No” said Pete “it’s definitely a different colour. A dirtier yellow. Almost mustard, maybe.” His hands and eyes turned back to the table, where he was playing with a set of cards. Pete was talking about a different yellow. He was talking about ‘Pumpkin Orange’. Which we’d always seen as a different colour. But not different enough. Interesting. We had an orange problem.

Or to be more specific, we had a colour problem, and Pumpkin Orange was just the indicator event. It’s not as much of a problem as it might be, if you’ve constructed a product that has ‘layers that learn’.

Let’s backtrack a bit, for those unfamiliar with the form…

Artefact Cards are a product that fell out of our studio, Smithery. It started when I was running a workshop using blank cards and Sharpies, to encourage people to draw their ideas, rather then just scribble with biros on sticky-notes. Drawing, I thought, would help people express their ideas beyond just language. It worked really well, but not just in the way I’d expected.

Because people had drawn their ideas onto blank cards, not post-its, they automatically started sliding them around. Grouping them. Stacking them. Dealing them. Treating them as if they were playing cards.

It meant that ideas would find new context instantly, fluidly, bumping against each other, acting in the way that all information does when buffeted through the internet. Small Pieces, Loosely Joined, as Weinberger wrote.

The cards we’d found were pretty scraggy though, so we started making better quality cards for ourselves. Ones we could be proud to pull out and use. Designed for a purpose we’d unwittingly discovered.

Then, we realised at the end of every workshop, people would sidle up to us, eyeing the leftover cards. “Errr…” they’d start “…are you finished with those? Can we keep them?”. We’d made a thing, without wholly meaning to.

[What happened next is a story for another time… let’s fast forward]

It’s late Autumn 2013. Buoyed by the reception to a new pack design, we launched a range of eight different colours. From having just one colour, Sunshine Yellow, we added Pumpkin Orange, Apple Green, Peckinpah Read, Peacekeeper Blue, Venetian Teal, Hot Pink and Radiant Orchid.

Why eight? Well, basically, it was an infrastructure decision, determined by an idea that we’ve come to know as ‘layers that learn’. How do you make physical products like you would web ones, changing small pieces at a time depending on user feedback?

The components of Artefact Cards is as follows; make the cards, then cartons for the cards to go in, the sleeves to go around the carton, and boxes to ship them out in.

This modular approach allows you to make individual decisions — how many of each shall we make at a time, and what should they evolve into. It’s like adding more different types of bricks to your lego box over time; what can we make from this now?

The layers in the product

But it’s more than just having different components that fit inside each other; it’s about taking user feedback as you recieve it, and folding it in to the next decisions you take about the different layers.

The layers of the product are improved, little and often, as you learn from user feedback, without you having to change the whole thing time and time again, or get it right from the very start.

For instance, we designed the first version of the sleeve with a notch cut out of the side, a call-back to the notch on the inner carton:

Artefact Cards Sleeve, v1.0

All you had to do was remember that, when putting the sleeve back on the carton, which way round it went. But, of course, people didn’t.

I’d watch people hold the two pieces in their hands looking for clues, and decide that the two notches must go on the same side

Lining up the notches

Which then caused problems, as suddenly there’s all sorts of snagging that would occur when you tried to slide the pack back together that way:


With the layers as set up though, we didn’t need to start again. We just needed to tweak either the sleeve or the carton to make those two layers work together. The new version of the sleeve uses the fact that people with try and match the two notches to its advantage — the debossed white section will slot into the notch underneath:

Second generation sleeve

Most of the time, the layers are a useful way of working, especially when building a business up from scratch. It allows you the flexibility to evolve the thing you get a little bit of feedback on, and solve problems one at a time.

The idea of working in this way is arguably the bread and butter of good web design, where you can tweak and change layered and modular elements independently of each other as user feedback arrives. Trying to work like this in physical materials is harder though, but I think a lot of the same principles can remain if you’re careful about the decisions you make as you go.

Which wasn’t the case when it came to the eight colours. In hindsight, it was a lazy conclusion on my part; people were asking for more colours, we had shipping boxes that held eight packs, so I decided to to eight colours.

I knew it had gone wrong as, during the months after launch, I found myself developing favourites. I’d go to the cupboard where I kept a stockpile of all the colours, and I’d veer away from some of them. If I was doing that, I reasoned, then all the other users probably were too.

A simple survey amongst users proved it; three of the colours, Sunshine Yellow, Apple Green and Peacekeeper Blue, were rather popular. The rest had more niche fanbases, to be polite about it.

Again though, the layers came to the rescue. It’s about solving the problems you find at exactly the layer you need to. The next run of the cards just had to be the colours that people loved. The yellow, the green, the blue.

Plus just one more.

“That is our Orange” I said to Pete. “It’s Pumpkin Orange”.

“That’s not really orange” said Pete. “Orange is Orangey, like an orange. That’s what Orange is, and this isn’t it…”

Isn’t it funny how you realise some things just start popping out at you when you’re looking for the? Every where I turned for the next week, there was a real orangey orange…

I’d share everything I came across on twitter, looking for the perfect shade. And that’s when Ben & Abi at The Conscious Project reminded me that their custome Artefact run used the most fantastic orange imaginable:

So there we have it; the birth of Conscious Orange, settling down wonderfully into the best Artefact Cards yet, nestling through the layers.

I’ve loved working this way, getting live feedback from live products, but in a physical manifestation rather than digital.

It makes it easier, I think, that the root materials of all of the layers are paper and card. They don’t spoil, so there’s no deadline to use them by. They’re pretty robust, so storage is easier. You can run them in relatively low numbers at sensible economic investment, and you can do it locally in the UK.

But it’s going to be interesting in future in taking the ‘layers that learn’ concept into other projects where the materials and processes are a bit harder to deal with. Onwards, and upwards.

More about Artefact Cards can be found here —

Artefact for Desk kit