From Reviews to Best-Sellers: How Hakuna Matata Became Amazon's No.1 with a Tiny Team and Over 20 AI 'Employees'

[Ebrun Original] How did a company that started with just four people achieve the top spot in a niche category on Amazon across the United States within two and a half years, scaling from zero to $25 million in revenue? At the recent "Next-Gen Cross-Border Summit" hosted by Amazon Global Selling, Sam Sun, founder of Hakuna Matata and a member of the "2026 Forbes China 30 Under 30 in Cross-Border E-commerce," shared his entrepreneurial journey.

Best-sellers are 'heard' from customer reviews; users will guide your path. After observing that the issue of children's eczema affecting sleep remained unresolved, Sam found a breakthrough with the new material Tencel. When "users thought it was too expensive, and suppliers thought it was too difficult to produce," Sam led his team to re-engineer the Tencel supply chain. They produced garments using medical-grade cleanroom standards, achieving stable mass production and a low return rate, ultimately becoming the category leader. The launch of the premature baby product line "was also identified by manually reading thousands of reviews."

"When a small team has limited energy, focus on serving the same group of customers as much as possible," Sam believes. "The biggest competitive moat lies in whether the entire team puts 100% effort into fulfilling that 1% of demand." Growth doesn't come from jumping to new categories. A new brand can succeed by developing products for the same use case, solving the needs of a specific group, which leads to repeat purchases and willingness to pay a premium.

Furthermore, it's worth noting that Hakuna Matata is a prime example of successfully integrating AI into cross-border e-commerce operations. They built a workflow pipeline with over 20 AI Agents, using an AI-native approach to rethink every problem. "AI Native isn't about slapping AI onto old processes. It's about using AI to re-architect the entire business workflow, guided by the authentic voice of the user," Sam mentioned. To reduce AI hallucinations, the team implemented an "AI Business Brain for Unified Orchestration," enabling AI to manage AI, ensuring the smooth operation of the company's systems.

The following is a transcript of Sam Sun's speech:

Good afternoon, everyone. I'm Sam. I run a mother and baby apparel brand—a small team that started with just four people and has now become number one in our Amazon subcategory.

Before I begin, let me ask: Has anyone ever created a best-selling product because of a negative review?

01. A Negative Review Exposed a Fabric Issue, Leading to a Tencel Supply Chain Overhaul

I received a negative review: "My son has always worn your clothes, but this recent purchase gave him a severe rash."

She even wrote an email in the middle of the night, sent at 3:17 AM US time.

When I read the full message, I realized: This wasn't a complaint; it was a mother's cry for help in the dead of night.

It was July 2023. We had just started and had launched a Tencel garment—soft and breathable against the skin. One day, it suddenly sold out. I checked and found out: A dermatologist had posted in a forum, saying she only allowed her own child to wear clothes made from two types of fabric: organic cotton and Tencel. For each fabric, she recommended one brand—and our brand happened to be one of them.

The stock sold out quickly, and I rushed to restock.

Why would a doctor recommend an unknown brand? And why did so many mothers buy it after the recommendation?

Digging deeper, I discovered: Eczema is the most common childhood skin condition in the US—affecting 1 in 5 children. To keep their children from itching, mothers would stock up on any expensive organic cotton or Tencel they could find. But organic cotton has an unavoidable flaw—it traps heat and moisture when the child sweats, making the already itchy skin even more uncomfortable all night. Tencel, however, broke through this dilemma.

But then the question arose: If Tencel was the solution, why did we receive that negative review? I traced it all the way back and found out—for that batch, the supplier had switched materials without telling me: Tencel is soft and stretchy, making it notoriously difficult to produce consistently. To save costs, they quietly used cheaper material.

At the same time, people in forums were complaining it was "ridiculously expensive." Users thought it was too expensive; suppliers thought it was too difficult.

In terms of raw material cost, Tencel and organic cotton aren't that different—but the factory gate price for Tencel was double that of organic cotton. Could we make it better and cheaper, so more mothers could afford it?

We traced it all the way back, back to a single yarn, and realized: That extra factory price wasn't in the raw material. The cost wasn't in "premium" quality; it was in the fact that no one wanted to do it. Fabric with 97% Tencel content had unstable processes, low yield, and small demand. Every link in the chain was fragmented, so each link had to factor waste, inventory, and risk into the price. No one was willing to connect this chain from start to finish themselves.

That year, our team—we spent most of our time on factory production lines and in QC rooms, scrutinizing every single process step: For more stable raw materials, we even traced back to an ecological plantation in Sichuan. We redefined every parameter for dyeing and printing with master craftsmen. We painstakingly raised the yield rate for this knit fabric from an industry low to a level that allowed for stable mass production.

02. Winning on Both Quality and Price with Medical-Grade Cleanroom Garment Production

Once the fabric was sorted, the next step was garment production. Traditional garment factories found it hard to scale with our new fabric. First, they needed to procure new specialized machinery and rebuild production lines. Second, it was difficult for factory operating standards to quickly reach the required quantitative levels because our fabric was so soft and unique.

So, we searched all over the country for factories that could handle it. In 2023, right after the pandemic, factories that had been making medical protective clothing were looking for new opportunities. So, I had a wild idea: Could a protective clothing factory possibly make baby clothes? Luckily, we actually found one! Not only could they do it, but they also came with a set of medical-grade cleanroom standards—the kind used for food, pharmaceuticals, and medical supplies!

I moved the most skin-contact garment production processes into their cleanroom, where airborne bacteria are controlled to under 500 per cubic meter. Together with the factory, we purchased new equipment and established a completely new SOP for baby clothing production and management, ensuring risk management and traceability at every step.

This approach won on both fronts: On quality, while the industry average return rate is around 30%, we achieved 8%. On cost, by minimizing waste at every step, we managed to produce this "ridiculously expensive" chain at a price point others couldn't match. And how loyal were the mothers who stayed with us? Our most dedicated customer repurchased a single style over twenty times.

Today, that style is on its seventh revision—my little niece, our cutest "product manager," has been wearing it since she was one month old, all the way to two years old now.

Soon, we became number one in our category.

03. The 'Heard' Best-Seller: Launching a Premature Baby Product Line

How did our first best-seller come about? The answer isn't in how smart I am; it's all in those 4,711 reviews. Listening to them one by one, we 'heard' the next best-seller—a product I never would have thought of myself.

The second best-seller's keyword: NICU—Neonatal Intensive Care Unit.

An NICU nurse commented: "I work in the intensive care unit. We're always looking for clothes that these tiny babies can actually wear." That's when I learned: Store shelves simply don't carry sizes small enough for premature babies.

Another mother wrote about her Christmas: She stood between store shelves, unable to find a single piece of clothing that fit her child because her child was too small—she said she cried several times in those stores.

So we made it—a premature baby product line that didn't exist on any shelf.

Later, a mother of a premature baby in Ohio wrote in our review section: Her baby was born five weeks early, and she hadn't had time to prepare anything—until her sister placed a box of these clothes in her hands at the hospital. Her baby weighed only 4 pounds 11 ounces.

That box was from us.

None of these best-sellers were "calculated" by data—they were 'heard' by us, one review at a time; then made by us, stitch by stitch.

You have to dive deep to hear it.

Eczema, NICU—these terms that later supported entire product lines were all identified by us manually reading through thousands of reviews back in the day. We used Excel, with formulas to extract high-frequency keywords, reading every single one, revising version after version—supposedly "cross-border e-commerce operators," we were forced to become vlookup experts, the "Excel masters."

04. Building an Agent Pipeline: Over 20 AI 'Employees'

But Excel couldn't contain the voice of the entire internet. With AI models advancing rapidly over the past two years, we applied the same determination we used to overhaul the supply chain to re-architect our entire business workflow: We handed the entire task of "listening to users" over to AI—creating a user-voice-driven AI Native system.

We coded a group of Agents, over twenty of them, each with its own role, like an assembly line.

At the front are three "ears": The "Review Section Agent" monitors all reviews in the Amazon category—that's what users say directly after using a product. The "Forum Agent" delves into overseas social media and mom forums—where they have private conversations and speak most candidly. The "Search Agent" analyzes keywords in the Amazon search bar—what a mother searches for in the middle of the night, told to no one, is the most authentic.

The captured voices are then passed to the "Persona Agent": It semantically dissects our over 6,000 Amazon reviews one by one, clustering them into distinct personas—Is she a first-time mom or has a second child? How old is the baby? Does the baby have sensitive skin? What causes the most anxiety at night? In the past, we could only "roughly know" who our users were. Now we can pinpoint: For each type of mother, what specific problem is she worrying about? Cross-referencing different personas can even reveal—which mothers are we not yet covering?

Next is the "Data Agent," which integrates almost all the officially authorized data from Amazon's backend—32 endpoints from SP-API, the Advertising API, AMC Marketing Cloud.

Now what we see isn't just "how much we sold": Which step is a listing's conversion getting stuck at? Which keyword's conversion is dropping? Which size had a sudden spike in return rate this month? What new terms are the mothers in our personas searching for this month?—Problems that used to require piecing together from multiple reports are now pointed out instantly. Which needs remain unmet, or even unseen, are also detected by it first.

05. Establishing an "AI Business Brain": Using AI to Manage AI

All this converges into an "AI Business Brain." It orchestrates everything, distributing personas and data to the next batch of operational Agents—what's distributed isn't KPIs, but the voice of the user.

Each job function is paired with an Agent: The "Design Agent" tells the designer what the next image should depict—maybe a new usage scenario emerged in the reviews, but the listing lacks that image. The "Advertising Agent" tells the operations manager which keyword to bid on tomorrow—a keyword that isn't even on third-party traffic charts yet. The "Content Agent" tells the short-video creator which pain point to address—often hidden in a real review or a video shot by a mother herself. Social media, supply chain, and product development each have their own Agent as well. This way, the entire company revolves around the same set of voices, with every task rooted in the user's real needs and feedback.

Of course, AI can also go astray: The scariest part isn't making mistakes, but confidently spouting nonsense while fabricating evidence. So, at the end of the line stands a "Spot-the-Flaw Agent"—it reviews all previous conclusions against a unified standard, specifically hunting for contradictions. Every Agent's output must pass this checkpoint: Is this image what the user wants to see, or what the AI thinks looks good? Does this need genuinely serve that type of user from the reviews? Is there evidence?

This is what we understand as AI Native: It's not about adding an AI layer onto old processes. It's about using AI to re-thread the entire business workflow along the lines of the user's authentic voice. It's not about using AI to create more images I "think" look good; it's about letting AI tell me what users actually want to see. It's not just about using AI to write more listings or stuff more keywords; it's about turning "starting everything from the user" from a slogan on the wall into a system that the entire company runs on every day.

But no matter how powerful the Agents are, AI doesn't know how to love. That part is still up to us—it was us who brought that premature baby product line, which made no commercial sense on paper, to life with an abundance of love.

This is how we shed the "Excel master" hats and became a group of "vibe engineers" afraid of losing our hair.

Today, our team has grown from 4 to 30 people. Armed with this entire AI Native system, we are expanding from North America into Europe, Japan, and Australia. This time, we are fully prepared for the needs of every mother.

I want to say to that mother who wrote at 3 AM: Take a deep breath, don't be afraid, everything will be okay.

That's the meaning of our company name: Hakuna Matata, a Swahili phrase from "The Lion King"—no worries, everything is fine.

Thank you.