Skip to content
Building AI Hardware: Product Restraint and the Cost of Traffic

Building AI Hardware: Product Restraint and the Cost of Traffic

Illustrations are AI-generated concepts, not photographs of actual Recolx products or app screens.

Part One: The Restraint It Takes to Build a Product

Do you know how hard it is to start a business now—to build a product?

The direction everyone is competing in is product differentiation. You add a feature; I add a feature. And I've noticed something pretty ridiculous happening with a lot of apps: there are just so many features, right?

Sometimes I open an app and can't even find the entry point I used to use. It gives me four or five choices, and I don't know which one to pick.

But in a way, there's not much you can do. Users are so demanding now. They have all kinds of requests: add this, add that. If a product doesn't meet their needs, they turn around and leave.

In this age of streaming content, all kinds of user needs have been stirred up. People want different things from apps and products. So taking your core features seriously, and building them with real restraint, is actually very difficult.

Our app doesn't even offer recording as a software-only feature—recording directly in the app without hardware. We can only record through hardware.

Why can't our app record on its own? It's because we've exercised some restraint in the product.

I have my own thinking behind that choice. But users might prove me wrong later, right? It's hard to say.

What my own experience has taught me, though, is that you can build features people never use. We, the people making the product, may applaud them—but they don't sell.

I think I've done a great job. I'm happy with it. I use it myself, and friends who get my product use it too. But in the market, it just doesn't catch on.

So a lot of the time, we need to distinguish between the person who buys and the person who uses.

Often they're the same person. But not always. In my own experience, the buyer is the first hurdle we have to get past: someone has to pay first.

Why do I feel this so strongly? Today, I'm not really talking about AI. I'm looking back on some of my experiences.

We started this AI voice recorder project in May 2024. But it wasn't until October 2025 that the product went on sale overseas.

Why was there such a long gap? Even building our in-house team, our sales team, and our sales channels didn't begin until 2026. A lot happened in between. A lot.

Looking at our monthly active users, daily active users, and hardware sales now, we've actually done reasonably well overall. From a business performance perspective, the AI hardware has done all right too.

But looking back on the journey, it's been painful. I've fallen into so many traps.

Of course, there were also some traps we didn't fall into.

AI companionship, for example—we didn't put that much energy into it. AI toys are also a very good industry, but we didn't go into that either. We stuck with recording.

I'm not saying those industries are bad. But there are directions we didn't pursue, features we didn't build. And at least we're still alive today.

From May 2024 until now, I've watched plenty of AI companies go under. Some rose very quickly, raised money, and then collapsed because they couldn't sustain themselves.

So I've done some reflecting on that myself.

Part Two: In AI Hardware, Reaching Customers Is the Real Hurdle

AI-generated editorial illustration of a founder facing the cost of connecting a recording product with real customers.
AI-generated conceptual illustration.

I think one of the biggest issues for AI hardware founders is understanding traffic—and understanding what that traffic costs. It's incredibly important.

For an overseas launch, my view is that you start with a first wave on crowdfunding platforms such as Kickstarter or Indiegogo.

Then you find specialist reviewers on platforms like YouTube, TikTok, and Instagram to give objective reviews of the product.

After that, you use sales platforms such as your own online store and Amazon, with feedback from real users to demonstrate the product's value. And you need to push the product at the right point in an upward cycle—not try to launch the brand when the market is already brutally crowded.

That's one lesson we've taken from our experience with the whole process.

More importantly, when it comes to R&D cycles for application-layer AI hardware, I believe shorter is better—not longer.

That may run counter to how a lot of investors think. When some investors spoke with us about AI hardware, they felt the development timeline might need to be longer, bringing up things like 3D printing or other products.

We don't see it that way.

We believe you should be making money from the start. You should be competing hard from day one.

For example, when I look at Kickstarter, Indiegogo, or sending review units to YouTubers, I ask: how much will I need to spend? How many creators with a million, ten million, or more than five million followers should I send products to for review?

How many people on Reddit need to be genuinely reviewing this product? How much discussion is there?

And on X—Twitter—is that buzz something we're creating ourselves, or are users actually talking about the product?

I think investment in this area should come first.

If the product itself isn't generating a strong enough market response, then create an event around it, or find another way to reach people. At its core, the cost of traffic plays a huge role in all of this.

Looking back at our own experience, if I'd done Kickstarter and Indiegogo well, I would definitely have worked with a large number of creators on YouTube and TikTok. TikTok would have generated some sales too.

As those sales came in, we'd then move on to Amazon and our own online store. Broadly speaking, that's the sequence: step one, step two, step three. After Indiegogo, you can also start Amazon in parallel with the content and sales work on YouTube and TikTok.

Let me tell you something else. In the category we're in—AI voice recorders on Amazon—at the peak this year, I've seen the cost of a click reach as high as US$8.

Think about what that means.

Eight dollars for one click. Roughly fifty yuan.

To put it bluntly, if someone sent bots to click on a competitor's ads for a few days, that competitor would be hurting too, right? It's eight dollars a click.

A lot of people don't make fifty yuan in an hour. But one click can already cost that much.

That's a frightening level of traffic cost, isn't it?

Of course, with traffic costs that high, I'd expect the margins—and the business opportunity in the category—to be fairly good too. So we've developed some of our own thinking around distribution and traffic.

But as soon as I finish explaining this process, people will ask, “Have you been very successful at it yourself?”

That's not how I think about it.

I'm not saying I have to build a huge brand. At this stage, I just want to build a small, thoughtfully run brand whose products people really enjoy using.

I'm not saying I need exceptionally high margins either. What I want more is for everyone who uses my product to feel something good from it. And to make money along the way.

To me, making money is something worthwhile and dignified.

You have to make money first to support R&D. Only then do you have enough money to give users better products. What I really dislike is how shamelessly some people in the industry chase money.

There's another element to this approach: borrowed credibility.

You might need trusted media outlets or public-account publishers to run what I'd call “PR pieces for appearances”—telling everyone how much funding we've raised, how much money we've secured.

But some of the information put out there may not actually be true.

When we've seen these stories in the past, we've wondered: this company says it raised a certain amount, but could the claim itself simply be made up? Making something up seems to cost nothing, and there don't seem to be many consequences. But consumers believe it, right?

On this point, I feel quite a few AI companies may not be all that sincere.

There are many ways to get attention. I'm not saying this approach can't work. But this particular tactic is one I consider unethical.

That's what competition can look like. Someone gets attention this way, and as long as they eventually make the story come true, they end up achieving their goal.

As we say, they've gone around in a circle and closed the loop. They exaggerated, but then they delivered. It just hadn't happened yet when they made the claim.

A lot of people working to generate traffic haven't actually achieved what they're boasting about when they say it. But by the time everyone believes them, it has happened. People really buy into the boasting, don't they?

It's a curious thing about this age of streaming content.

And I find it strange today that simply building a product, keeping your head down, and running a business isn't as easy as it used to be.

You have to step out in front and build a personal brand.

For example, I now record videos, write articles, and put these thoughts out there to reach people. I never used to think I had to do that. Now it isn't a choice—it's a path you have to take.

It's the same overseas.

We've made mistakes along the way too. I'll tell you more about those another time.

Leave a comment

Your email address will not be published..

Cart 0

Your cart is currently empty.

Start Shopping