Here's what Daniel sent in this time. He says he's terrible with money — his words — but he's stumbled onto one strategy that seems to actually work. He buys a lot of niche, obscure stuff online, way more than in physical stores, and he's made a deliberate choice to concentrate almost all of that spending on just two platforms: Amazon and AliExpress. His reasoning is that even though these marketplaces are run by algorithms at massive scale, he can build up a solid order history that gives him credibility when something goes wrong. He's careful about disputes — only opens them when an item is defective — and he's actively opposed to anyone who games the system. His intuition is that being a good customer pays off. He gets decent support on Amazon even though the baseline has gotten pretty bad, and he doesn't run into the horror stories other people report on AliExpress. He admits he doesn't have hard evidence. But he's noticed the experience is good enough that he increasingly sees online marketplaces as a genuine alternative to buying locally. So the question is: is there any actual mechanism behind this, or is it just a comforting story we tell ourselves?
The short answer is there are mechanisms. Real ones. And they're not loyalty programs in the traditional sense — no points, no tiers, no birthday coupon. What Daniel's describing is something more like... a shadow reputation score that follows you around inside these platforms, built entirely from your behavior, and it determines how the system treats you when things go sideways.
A shadow reputation score. That sounds vaguely dystopian and also exactly right.
Let me start with Amazon, because their system is the most documented. Amazon tracks something internally called cost-to-serve. It's not a single number you can look up — it's a composite metric that pulls in your return rate, how often you contact customer service, and even the language you use in those contacts.
The language. As in, whether you're polite versus whether you're screaming at a chatbot.
Yeah. And whether you're contacting them about things the system considers legitimate versus things it flags as... let's call it high-maintenance. The cost-to-serve score is the backend number that decides whether, when you report a twelve-dollar item as defective, the system issues a returnless refund — keep it, here's your money back — or makes you box it up and drop it at a UPS store.
So the returnless refund isn't random generosity. It's a calculation.
It's a calculation that's been running on you for years. The algorithm has a threshold, and the threshold moves based on your history. A clean record with few returns and few disputes means you're more likely to be trusted on the margin. The platform is literally deciding whether the cost of processing your return, inspecting the item, restocking or disposing of it, and handling the customer service interaction exceeds the value of just giving you the money and moving on. And your history is the variable that tips that calculation one way or the other.
Which means Daniel's intuition about not gaming the system is doing real work. Every time he could have filed a dispute and didn't, every time he ate a small loss rather than making it Amazon's problem, he was making a deposit into a credibility account he didn't even know existed.
And there's research that quantifies this on other platforms. A twenty twenty study of eBay's dispute resolution system — this looked at both human arbitrators and automated decisions — found that buyers with a history of filing disputes are significantly more likely to lose borderline cases. Same facts, different outcome. The past behavior was a measurable penalty.
Wait. Same facts, different outcome. That's the part that should make people sit up.
It's not supposed to work that way in theory. In practice, the arbitrator — human or machine — is looking at a screen that shows your dispute history, and that history colors how they interpret the evidence you're presenting now. If you've filed twelve disputes in the past year, the tenth one gets a lot less benefit of the doubt than the first one did. Even if number ten is completely legitimate.
So the system isn't just tracking whether you're a fraud risk. It's tracking whether you're expensive to serve, and it's pricing that into every decision.
And that's Amazon and eBay, where the mechanisms are relatively visible. AliExpress is where this gets interesting, because AliExpress doesn't have a loyalty program. There's no Prime. There's no obvious incentive to concentrate your spending. And yet Daniel reports a disproportionately good experience there.
Which is the puzzle. If there's no loyalty program, what's generating the good experience?
The escrow system. AliExpress holds the buyer's payment until the buyer confirms receipt. That's fundamentally different from Amazon, where you pay upfront and the money is gone. On AliExpress, the seller doesn't get paid until you say you got the item and it's acceptable.
So the power dynamic is inverted.
Completely. And here's the part that connects to Daniel's strategy: every time a buyer confirms receipt quickly and doesn't open a dispute, that's a positive signal. In a marketplace where a huge portion of buyers are price-shopping aggressively, opening disputes over minor cosmetic issues, or letting orders sit in confirmation limbo for weeks, a buyer who consistently confirms receipt within a day or two and only disputes genuine defects... that buyer is an outlier. And in a low-trust environment, being an outlier in the trustworthy direction generates disproportionate returns.
Because the signal is cleaner.
There's less noise. On Amazon, returns are normalized — the system expects a certain percentage of returns, and your behavior is measured relative to that baseline. On AliExpress, the baseline buyer behavior is more chaotic. A lot of buyers are gaming, a lot of sellers are gaming, and the whole escrow mechanism exists because trust is so low. So when the platform sees a buyer with a long history of clean transactions, that buyer's dispute gets fast-tracked. Not because there's a loyalty program, but because the system has learned that siding with this buyer is a safe bet.
And the seller side reinforces this. AliExpress has buyer satisfaction ratings that are visible to sellers.
Yes. Sellers can see your rating as a buyer before they even ship the item. If you've got a history of disputes or slow confirmations, some sellers will cancel the order rather than risk dealing with you. If you've got a clean history, sellers are more likely to prioritize your order, ship faster, handle issues directly rather than forcing you into the formal dispute process.
So Daniel's good experience on AliExpress isn't a mystery. He's built a reputation in a system where most participants don't bother, and that reputation is paying out in faster resolutions and fewer fights.
And I think there's a knock-on effect here that's worth pulling out. Concentrating spend doesn't just build a good history — it makes you a predictable customer. Predictability is valuable to algorithmic systems that optimize for low-variance behavior. A customer who always buys niche items in specific categories and never returns them is a low-cost, high-margin user. The platform wants to retain you, even if it doesn't have a named retention program.
The algorithm doesn't need to know your name to know you're profitable.
It knows your behavioral fingerprint. And that fingerprint is what determines whether the chatbot offers you a refund in thirty seconds or routes you through three escalation tiers.
So we've established the mechanisms are real. But let me poke at the limits of this. Daniel's strategy works for someone buying niche, discretionary products where he's not price-shopping every order. What happens if you're buying the cheapest option from a different seller every time?
You never build a history with any platform. The algorithm has no signal to reward. You're a ghost — every transaction is your first transaction as far as the system is concerned. And the system's default posture toward ghosts is skepticism.
Which means the people most likely to need good customer service — the ones scraping the bottom of the price barrel — are the least likely to get it.
That's the uncomfortable implication, yeah. The system is regressive in that way. The customers who cost the least to serve get the best service. The customers who cost the most get the most friction. It's not a moral judgment by the platform — it's just math — but it produces outcomes that look a lot like moral judgment.
And this connects to the gaming-the-system counterpoint Daniel raised. He said he's actively opposed to people who try to game Amazon's return policies. There's a whole subculture of that — people who buy a camera for a vacation and return it after, or order three sizes and send two back.
And the platforms are watching. Amazon has banned customers for excessive returns. Not just flagged them or sent a warning email — actually terminated their accounts. There are documented cases. The system tracks return rate as a percentage of purchases, and when you cross a threshold that Amazon doesn't publish, you're done.
So the gaming advice is actively counterproductive. The data proves it backfires.
It backfires in two ways. First, the obvious one — you get banned. Second, the subtler one — even before you hit the ban threshold, your cost-to-serve score is climbing, and that means when you have a legitimate problem, the system is less inclined to help you. You've burned your credibility on fake disputes, and now the real one gets treated like a fake one.
The eBay study we mentioned earlier is basically that dynamic quantified. Past dispute history predicts worse outcomes on current disputes. The system learns that you're expensive, and it prices that in.
And what Daniel's doing is the inverse. He's building a history that says: this person only contacts us when something is wrong. That's a valuable signal. The platform can trust that when he raises an issue, it's worth investigating.
There's something almost pre-modern about this, which is what makes it counterintuitive. We think of algorithms as faceless and impersonal. But what Daniel has accidentally done is replicate the dynamic of a small-town shopkeeper who knows you. The shopkeeper doesn't need a loyalty card. He just knows you've been coming in for years and you don't cause problems, so when you say the bread was stale, he gives you a new loaf without asking questions.
That's exactly the right analogy. And the mechanism is the same — repeated interactions building trust — but the substrate is different. Instead of face-to-face memory, it's behavioral data. Instead of a human making the judgment, it's an algorithm. But the trust-building dynamic is identical.
Which brings us to Daniel's claim that this strategy makes online shopping a viable alternative to buying locally. I think the hidden variable there is that local shopping also has a relationship component, and what he's replicated online is a digital version of that relationship.
The local shopkeeper knows you. The algorithm knows your behavioral fingerprint. In both cases, the person on the other side of the counter — human or machine — has a reason to believe you're acting in good faith. And that belief translates into better outcomes when something goes wrong.
The difference is the local shopkeeper can also tell you that the jacket doesn't fit right. The algorithm can't do that yet.
Not yet. But it's getting closer. Amazon's recommendation system is already factoring in your return history when it suggests products — it's trying to steer you toward items you're less likely to return. That's not the same as the shopkeeper's eye, but it's operating on the same principle: reduce friction for the customer by learning what works for them.
So where does this strategy break down? Daniel's been doing it intuitively, and it's working for him. But there have to be failure modes.
The biggest one is platform risk. If you've concentrated all your spending on Amazon and Amazon decides to ban you — whether justified or not — you lose your entire purchase history and the credibility that came with it. There's no appeals process that looks at your good-customer track record.
And there's no portability. Your Amazon reputation doesn't follow you to AliExpress. Your AliExpress history doesn't matter on eBay.
That's the fragmentation problem. Every platform's trust score is siloed. You're starting from zero on each one. Which is actually an argument for Daniel's strategy — pick two and go deep rather than spreading across six platforms where you never build enough history on any of them.
The other failure pattern is that the algorithm's definition of a good customer can change without notice. Amazon could decide tomorrow that the threshold for returnless refunds is now twice as strict, and your history doesn't earn you what it used to.
That's the fundamental asymmetry. The platform owns the scoring system. You don't get to see your score, you don't get to dispute it, and you don't get told when the rules change. You just notice that customer service got worse for you, and you don't know why.
Which is where the comforting-story risk comes in. Daniel might be right that his strategy works, but he might also be experiencing a run of good luck and attributing it to his behavior. The human brain is very good at finding patterns.
It is. But in this case, the mechanisms are real enough that I think the pattern holds. The eBay study is the smoking gun for me — same facts, different outcomes based on dispute history. That's not luck. That's a measurable effect.
So we've got mechanisms on Amazon, mechanisms on AliExpress, and research from eBay that confirms the pattern. The question Daniel didn't ask, but I think is lurking under the surface, is whether this strategy scales. What happens if everyone starts doing it?
If everyone concentrates spend and everyone is a good customer, the signal collapses. The platform can't distinguish between users anymore, so it reverts to treating everyone the same — which is the baseline experience Daniel's trying to avoid.
It becomes an arms race. The platform raises the bar for what counts as trustworthy.
And that's already happening in some ways. Amazon's return policy has gotten stricter over time. The window is shorter for some categories. The returnless refund threshold has probably moved. As the average customer behavior improves — or as Amazon's tolerance for cost decreases — the bar for good customer rises.
So Daniel's strategy works now, but it's not a permanent hack. It's a moving target.
Everything in these systems is a moving target. The algorithm is being retrained constantly. The thresholds are being A/B tested. What worked last year might not work this year.
But the underlying principle probably holds even if the specific thresholds shift. Being less expensive to serve than the average customer is always going to be advantageous.
I think that's right. The specifics change, but the direction doesn't. Be predictable. Be low-cost. Don't generate exceptions. The algorithm will reward you because you're cheaper to keep than to lose.
That's a slightly bleak way to think about your relationship with a store, but it's accurate.
It's accurate. I don't know that it's bleak, exactly. It's just... the relationship was always transactional. The algorithm just made the transaction explicit.
Hilbert: Forty percent.
...Forty percent what?
Hilbert: Higher chance of getting a partial refund approved on a defective item. If you'd confirmed receipt quickly and never opened disputes. We ran a pilot. Early two thousands. Company called TrustBuddy.
TrustBuddy. I've never heard of that.
Hilbert: Nobody has. It failed. We were trying to build a buyer reputation score that would port across marketplaces. eBay, AliExpress, a few smaller ones. The idea was you'd carry your trust score with you. Platforms hated it. Nobody wants to share data.
But you got far enough to run a pilot.
Hilbert: Small electronics seller on AliExpress. About eighteen months of transaction data. The dispute team had a trust score on their screen — we weren't allowed to show it to buyers, but the agents could see it. And the numbers were clear. Buyers who confirmed fast and didn't dispute had a forty percent better chance on partial refunds. Controlled for item price, controlled for seller. The score also factored in how long they took to confirm receipt. Every day you sat on an order, the score ticked down a little.
The confirmation speed was a direct input.
Hilbert: Biggest input, actually. More than dispute history. The logic was that a buyer who confirms immediately is signaling they're not going to sit on the item for two weeks and then claim it arrived broken. Fast confirmation meant low risk.
Which means Daniel's behavior of not sitting on orders — which he probably thinks of as just being organized — is actively building his credibility in a way he doesn't even realize.
Hilbert: It's the single most effective thing he's doing. More than concentrating spend, more than being polite to customer service. Just confirming receipt within a day. The algorithm loves that.
This was visible to the AliExpress dispute team. Not to buyers.
Hilbert: Internal only. A number between one and a hundred. The agents didn't have to use it — there was no rule that said score below forty gets denied — but they looked at it. Everyone looks at the number. You put a number on a screen and people factor it in. Even when they're told not to.
Did the pilot change anything at AliExpress after TrustBuddy folded?
Hilbert: They kept the confirmation speed tracking. I don't know if they still weight it the same way, but the infrastructure didn't go away. It's too useful. Tells you which buyers are going to be problems before they become problems.
That's the part that sticks with me. The infrastructure persists even after the company that built it disappears. The data is too valuable to throw away.
Hilbert: The data is the whole thing. TrustBuddy had a nice logo and a terrible business model, but the data pipeline we built is probably still running somewhere in AliExpress's backend. They'd never admit it. But why would they turn it off? It saves them money on every dispute.
Daniel's intuition isn't just a comforting story. There was literally a number on a screen.
Hilbert: There was a number on a screen. He's right. He just doesn't know why he's right.
Forty percent. That's not marginal. That's the difference between getting your money back and eating the loss.
Hilbert: On a twelve-dollar item it's not life-changing. But if you're buying niche products regularly, it adds up. He's probably saved himself a few hundred dollars over the years without knowing it.
The confirmation speed thing means he's been doing the most important thing by accident.
Hilbert: Most people don't confirm receipt at all. They let the system auto-confirm after the timeout. Those buyers have the worst scores. The system assumes they're either negligent or waiting to dispute. Either way, not trustworthy.
The advice — if you were going to give advice — would be: confirm receipt immediately, dispute sparingly, and concentrate your spend.
Hilbert: That's it. Three things. Everything else is noise.
Daniel's going to feel very validated when he hears this.
Hilbert: He should. He figured it out without a data pipeline and an eighteen-month pilot. Most people don't.
The open question, though — and I keep coming back to this — is what happens as more people adopt the strategy. If everyone starts confirming receipt quickly and disputing sparingly, the signal collapses. The platform can't distinguish between trustworthy buyers and people who've just read the advice.
Then the platforms raise the bar. Faster confirmation windows. Lower tolerance for disputes. The arms race continues.
Or they make the score explicit. Turn it into a product. Your buyer reputation score, visible to you and to sellers. Portable across marketplaces. TrustBuddy's idea, resurrected.
That feels inevitable, actually. Someone's going to do it. Maybe not the platforms themselves — they have no incentive to share data — but a third party that aggregates your purchase history across platforms and sells access to your score.
A buyer credit score. That's what you're describing.
A buyer credit score. And Daniel's strategy of being a good customer would translate directly into a high score. He'd be ahead of the curve.
Which is either reassuring or terrifying, depending on how you feel about credit scores.
I think it's both. The mechanism is real, the strategy works, and the future is probably more of the same — just more explicit and more portable. Daniel's intuition was right. He just got there early.
This has been My Weird Prompts. Thanks to our producer Hilbert Flumingtop.
If you enjoyed this episode, leave us a review wherever you get your podcasts. It helps other people find the show.
Email us at show at my weird prompts dot com with your own marketplace experiences — we'd love to hear whether concentrating spend has worked for you, or whether you've run into the limits of the strategy.
We'll be back soon.
