AI in your product: to build or not build. Is not the question.

AI in your product: to build or not build. Is not the question.

Not using AI today is a mistake. And still, not every AI application is a chat window. How we recognise the form that fits.

Not using AI today is a mistake. And still, not every AI application is a chat window. How we recognise the form that fits.

Portrait of Marc Siefert

Marc Siefert

Director Product & UX

4 min Reading Time

AI

Share this article

The honest answer up front: anyone not using AI today is making life needlessly hard, both in your finished product and in the building of it. For me that stopped being a trade-off a while ago, it is a precondition, which is why the interesting question is no longer whether but in which form.

And that is exactly where it gets messy. Because AI has become a commodity by now, it is available to everyone and costs almost nothing, and a commodity differentiates you from nobody. Here is how we notice at Silberpuls that a planned AI feature has the wrong form, and what has to be in place before we start building.

Two sentences that make me prick up my ears

The first is the one where AI is everywhere, so we should be doing something with it too. That is not really an argument, it is an observation, and if no second one turns up in the whole conversation, then we are building a feature because other people have one. Features like that usually survive a quarter, after which nobody looks at the numbers behind them again.

The second is subtler and comes up far more often: a product idea gets taken from somebody else and rebuilt in house. Just because ChatGPT and Anthropic managed to build an AI interface as a chat window does not mean every AI application from now on has to happen inside one. That chat is the form which worked first says something about those products and fairly little about yours.

And this is not a matter of taste either. The Nielsen Norman Group has observed that people in chat almost never get what they wanted on the first attempt and have to feel their way through several rounds, which is perfectly fine for research and gets quite tiring when there is a purchase decision at the end of it.

What has to be there before we build

From our point of view it mainly takes a clear product vision, so really just an answer to how this helps users and what concretely gets better about the product afterwards. Once that answer holds, the form usually follows by itself, and when it does not, people discuss models and vendors while the actual question sits unanswered in the room.

An example I use pretty often: just because I can chat with a language model, my shopping experience does not get better, because nobody wants to chat when it comes to fashion. That is a visual matter, you want to look, compare and get inspired.

AI can still contribute a lot there, just differently from how you typically know it, namely in the search, in the filters, in the order of the products and in the question of which size suits this person. In our work for Gitti exactly that decides whether somebody places a second order, and in e-commerce projects that is where the money sits these days.

The Baymard Institute has been researching how people search in shops for years, and the pattern is stable: they type incompletely, they mean something other than what they write, and they expect a usable result anyway. That is an excellent job for a model, and in the end it looks like a good search rather than like AI.

Sometimes the answer is the whole product

So that this does not sound too cautious: there is the other case, and we have had it. At a client in education the product used to be a toolbox, a collection of useful functions sitting next to each other, and in the course of a vision design we moved it entirely to a real AI product, chat interface and everything around it. There the chat window was the right form, because the users genuinely think and phrase things in language, and because it changed how the company saw itself, from a tool vendor to an AI vendor.

The difference between that case and the copied chat window is not the technology, it is the order. Here the question of how people actually get helped came first, and the form came after. And that is exactly the order we work in at Silberpuls when somebody comes into the conversation with a finished feature in mind.

What that means for your product

Write down one sentence that gets by without the word AI and still explains what is better after the feature. If that sentence does not come together, you have a technology and not yet an idea, and you will notice that later anyway, only then with a backlog attached.

Also check whether your users want to write at all in that situation. Somebody researching at a desk is quite happy to type, somebody looking for a pair of trousers one-handed on a phone less so. Our AI principles start with the situation for exactly that reason and not with the model.

And have a look at what is already tedious in your product before you put something new next to it. In almost every UX Audit we find two or three places where a model could take real work off people, and none of them is a chat window on the homepage.

Just because I can chat with a language model, my shopping experience does not get better.

Marc Siefert

FAQs

Should we rather wait until this settles down?

No, waiting is the worst option. Not using AI today costs you speed in building and quality in the product, and the gap gets expensive later. This is not about holding off, it is about picking the form that fits your product.

How do we recognise that a chat window is the right form?

By your users thinking in language in that situation anyway, by the task being open with many right answers, and by them being allowed to feel their way through several rounds without it becoming annoying. If one of those does not apply, the function is probably better placed in the search, in a form, or in the recommendations.

Is AI still a competitive advantage then?

The model is not, because everybody has it. The advantage sits in your product vision, in your data and in the form you find for it, and that is exactly the part nobody can buy in.

Is AI your solution or your argument?

We sort that out before the first feature.

Our AI principles, including the cases where we advise against it.

Our AI principles, including the cases where we advise against it.

E-Commerce Design that solves search and advice visually, not in chat.

E-Commerce Design that solves search and advice visually, not in chat.

How can we help?
How can we help?
Line illustration of a hand holding a card
Line illustration of a hand holding a card
English