Expand this concept into an "Amazon power search"?

[allquixotic]

AmazonBrandFilter is a very useful tool! But it is just the start of what I think could be an even more useful tool for Amazon shoppers.

This is a long, sort of "meta-issue" that I hope will start a discussion about how to possibly expand upon this tool, and weigh some design decisions at the top level before this project becomes too big. I think it's likely to gain a lot of traction soon, and it will be good if the community of users and developers around this project have discussed where we are, and where we're going, before this hits Slashdot or Ars Technica.

Thesis

I observe that:

• Amazon likely profits significantly from the way things are (i.e., allowing a lot of "junk" brands on their site), or they would be investing in shutting them down. A significant portion of sales probably result in products not being fit for purpose for the consumer, but they don't deem it worth the effort to return, so they just eat the loss and throw it away. So, this is unlikely to change any time soon on its own.
• It is so easy to set up a "brand" for products -- often, without even manufacturing anything yourself -- that the market encourages this "excess competition" for products with high profit margins, e.g. because they're small and easy to ship, insanely cheap to mass produce (made of plastic, etc.) and benefit from economies of scale and global trade market forces and policies. This type of thing will continue to worsen over time.
• "Junk brands" are not the only problem consumers have when shopping for products. Other challenges include fake reviews, false advertising, products that are designed to fail just after the return or warranty period, the environmental and geopolitical impact of consumer buying choices, lack of information about products and companies that make them, lack of transparency around supply chains, etc.
• Most people who shop online, whether deliberately or subconsciously, have "mental filters" that they use to manually sift through the BS, depending on their level of expertise, whether they've been burned by bad purchases, and whether they care about economic or environmental impacts, etc.
• The filters of each person vary greatly, depending on their political views, the extent to which they care about various factors, past experience, etc. Some have sworn off brands that others think of as completely legitimate and trustworthy. As a result, nobody can come up with one single "trusted brand list" that works for all people and all products.
• Sometimes, people want to get very nuanced with our product purchases. For example, someone might be totally fine with a certain product segment from a given company, but have a bad experience or other reason not to buy a product in a different segment by the same company. As a result, we probably also need to look at the products, not just the companies.
• Due to shifts in the supply chain, with more and more stuff going "online-only" (or at least, with online being many times more convenient than driving to the nearest brick and mortar store that has what you want), it's often the case that Amazon may be the only online marketplace -- with any semblance of reputation -- that offers certain product categories.
• Other times, you can order direct from the manufacturer or a distributor, but they charge an arm and a leg for shipping, or require many weeks to get you the same product that Amazon can deliver in 1-2 days.

As a result, shopping for seemingly simple household items can become a real chore. Finding a product that meets your needs, can physically (or digitally, in the case of software) perform exactly as needed, for as long as needed, at a price you can afford, that can be delivered in the time you need it, is an extremely hard problem to solve.

The fact is, Amazon has come closer than anyone in human history at solving that problem for most people in the industrialized world.

But, just like the Internet in general, we're suffering from information overload. There's just too much stuff, and it takes too much of our time to sift through it all to find the "signal" -- the brands that stand behind their products, are made in countries we want to support with our business, with the right interests in sustainability and fair labor, with attention to quality and reliability, and without gouging the consumer with excessive prices.

Considerations in Developing Solutions

When thinking about how to solve these hard problems, we in the software engineering community can envision many technical approaches. But a few common themes that come to mind include:

• Trustworthiness of the people and data: Even with the scope of the AmazonBrandFilter{List} project today, there is a big question mark about how to verify the trustworthiness of the data. As a simple example, how could we filter out "shills" submitting pull requests to include brands in the whitelist, while secretly being paid by brands to do so? Obviously, someone submitting the brand "WANGWA" with a completely new GitHub account is suspect; but if this project really takes off, they may start paying otherwise reputable GitHub contributors a few bucks to submit PRs or issues. Or they may hide a junk brand in a PR with a ton of other legitimate companies to make it hard for us to spot the "needle in the haystack".
• Possible roles of Machine Learning: It remains an open question: are there "emergent properties" about brand names (or the entire text of product description pages) that would enable sophisticated ML models to somewhat reliably detect junk/fake/scam products? Is there a pattern in their brand names that ML could find that we couldn't express with a classical algorithm? Perhaps an ML algorithm could compute a percentage confidence that a given, unknown brand is junk, and label it accordingly (or hide it entirely) on Amazon product pages and searches? Do we even want to go down this path, or will it just lead to an arms race with "brand creators" finding new and inventive ways to break the meta?
• Since many different data about vendors (people who post products on Amazon), manufacturers (companies that actually make stuff), brands/labels (names products are sold under) and products themselves can be relevant (or not) to different consumers, what data should we collect? Right now, we're going for the lowest of low-hanging fruit, but we could go so much further to shorten the product search/research time for folks looking to find the right item for themselves, quickly and easily.

The 20th century offers us little in the way of solutions. Organizations such as the Better Business Bureau and Consumer Reports might purport to be helpful in this area, but they are seriously flawed and mismatched with the current marketplace:

• BBB and CR tend to only cover major corporations, and most of the time only ones headquartered in a country like the US, Canada, UK, Germany, Japan or South Korea. There are plenty of brands selling legitimate, high-quality products from other countries, from China to eastern Europe to Africa and South America.
• BBB and CR are dependent on consumer-submitted ratings to make their determinations, but it turns out that most people don't spend time submitting anything to these sites unless they have a problem. As a result, you end up with a bad signal to noise ratio anyway, because most companies end up looking bad on these platforms. Very few people spend their precious time going in and rating a company highly for delivering a good product, but they will eagerly go and submit a complaint when they mess up, even if they only do so very rarely.

One Possible Solution

Given all this background, here is one possible approach to a solution that makes some compromises while trying to provide maximum "return on investment" for the community's time and effort:

Filter Maintainer Trust

The question we should be asking is, what is the means by which we can maximize confidence that someone is submitting data in good faith (without the intent to trick us into whitelisting junk brands), while minimizing the barrier to entry? These two factors seem to be directly related; meaning, an increase in confidence comes with an increase in the barrier to entry, while reducing the barrier to entry will often (but perhaps not always) chop down our ability to detect and combat bad actors.

As a first attempt, perhaps we should require submitters to provide evidence of the reputability of a brand? This doesn't require them to have purchased anything; they just have to link to something like a website, Wikipedia page, etc. that helps to solidify the brand's identity.

By putting the burden of proof on submitters, this will save time for the "core team" (the PR reviewers / those with committer access) so they don't have to go doing research on their own.

Data Quality / Data Maintenance

We also have the concern of data quality independent of trust. Someone can, in perfectly good faith, submit a brand that turns out to be junk. Maybe they're excited because they just received a cool product, and immediately PR the company's name, only to find out that their product broke on day 31 of ownership, right after the return period, and the company is nowhere to be found when trying to RMA it... oops.

Over time, as the list expands, it will become increasingly difficult to "prune" the list of companies that either (a) used to be fine, but have become junk brands (or otherwise no longer fit our criteria for whitelisting); or (b) no longer exist -- bankruptcy, renaming, re-branding, mergers and acquisitions, etc.

In the case of both establishing submitter trust and continual data maintenance, assuming we're keeping this fully "human in the loop" (no AI/ML), we will need a growing core team to review both new submissions and the existing dataset.

Oh, and, if we decide to expand this tool beyond just keeping a whitelist of brands, that vastly increases the volume of data we have to curate, which is a huge ask for even a moderately sized team of dedicated humans. Yikes.

To help with this burden, we could perhaps employ some technology. It's easy enough to determine if a brand is still relevant: if you search Amazon for that brand and don't get any (good) results, or all their products are indefinitely out of stock, that's a sign that you can count them out. Then we can have manual reviewers look at the shortlist of companies that came up "empty" and figure out which ones to nuke from the list to shorten it.

Multi-faceted Data Warehouse

This one will require a lot of coding, data collection and maintenance, but I think the minimum important data to collect (to allow users to filter on using the browser extension) are:

• Countr(ies) of origin of the brand
• Brand reputation parameters: do they offer minimal warranties or better than required, like Belkin? How are they on sites like Trustpilot or other "brand reputation" sites we might like to pull from? What about the opinions of AmazonBrandFilter contributors?
• Age of company: Some might prefer to only buy from companies X years old or more.

I'm sure folks might also be interested in things like company ESG initiatives (Environmental, Social and Governance), sustainable packaging, etc. but these are much harder to determine/collect outside of a few big brands that make a big deal out of this, like Apple and Patagonia.

Once we collect this data, we can implement UI elements to let folks narrow down their searches to exactly what they're looking for. But we might not be able to justify expanding the data like this until we have a much larger, more vibrant product/community, where we'd have the development, review and data submission resources to make it happen.

Conclusion / My Help

For now, I'm mainly going to contribute by occasionally sending in PRs to the filter list, until it's clearer what the overall direction of the project will be.

But, if this gains the attention that I think it legitimately deserves, it might go a lot further than that. I've got decent JS client-side development experience, and can probably help with designing the data collection process for a multi-faceted data warehouse that isn't just a text file of brand names (not that this is a bad start; it isn't! This is more than we had before, which was nothing!)

Anyway, thanks for working on this project.

[chris-mosley]

First off. It's making me SO happy that people are as excited about this as they are. I honestly expected that this project would be a tiny little obscure project I worked on for a while and no one else used before I personally got bored maintaining the list and abandoned it.

I'm going to attempt to address most of your points but the short version is that generally I'm hoping to keep this addon relatively focused for the time being.

At the moment this list is not meant to be a stamp of quality. It is simply meant to filter out a brand that might not be there tomorrow. If a brand makes crap products but sticks around then it is on them to sink or swim on whatever reputation they develop. I think something like fakespot is a better arbiter of who is "good" and who is "bad." I'm here to tell you who is "real."

AI/ML is outside my wheelhouse right now and I think quite out of scope in the short term. I would be interested in trying something but that will be well after v1.0

Much of the rest of these are things that would be address by the features I want to implement but currently only have in mind. I will probably publish a roadmap sometime soon.

Some of the things I have in mind:

• Personal allow/block lists. Which would give individual users the ability to block a "real" brand for personal reasons. And to add a brand that is missing from the current list
• Allow users to report missing brands through the addon popup itself
• Per department enablement. Currently the addon will filter things like video games and I don't want to have a list of every game
• Dovetailing off of that would be per-department lists. This would significantly help filtering by not considering brands that dont operate in different product sectors
• And going off of the personal lists. I will probably implement something like ublockOrigin where a user can add arbitrary lists from anywhere and if someone wants to maintain a list of brands that they feel are more ethical/environmentally friendly etc, allow users to subscribe to that list.

But this is all pretty far in the future. Right now I have an addon that can't find brands that have spaces in their names. Once I get to v1.0 we can start thinking about the moonshots.