Amazon Personalize: Real-Time Recommendation
aws.amazon.com
aws.amazon.com
More insidiously - why not offer a higher price on graham crackers if they seek them out on their own? They’ve already made a decision.
More or less the same tactic Amazon used for search results: Return useful/functional results for a few years to train you users to trust what is near the top, then start putting promoted products at or near the top and generally ordering the search by what increases your revenue the most, rather than for any user-oriented goal, AKA "sort by relevance".
Also the same thing Amazon did with prices in general, get people used to thinking you're the market with the best price/deal, etc.
If you use their cloud products, ala Netflix, they may decide to compete with you any moment. Netflix may have been too far along to be beat but the next product won’t be. If you sell products on their market place and if they do well you run the risk of them copying you. If you do anything in their ecosystem you are living with the fear that the lion may wake up and swallow you whole at any point and time.
Edit: Maybe I’m a bit paranoid but this feels like a great way for them to get to know exactly who your customers are and what their behavior is. All the while training their own personalization to get much better.
This isn't how the world works.
And the example of Facebook with Onavo is also not plucked out of thin air[0].
Sure, it's not how all of the world works but for the big players, it's a important tool in their toolbox.
Of course, Amazon could lie, but now if Netflix has a reason to suspect wrongdoing, they could sue.
You are not paranoid. That's exactly the business model of the retail website when hosting independent merchants.
The lion waits until you finished eating dinner...
SageMaker was a ground floor service and you should expect to see many more services built atop it.
I’m not sure that’s accurate — and there’s plenty of ways for organizational structure to strangle good ideas.
[0] https://en.wikipedia.org/wiki/Hierarchical_temporal_memory
But if they pay me (well the consulting company I work for) enough I might fly in to hold some session along the lines of
- don't annoy power users for no good reason (a certain ad delivery company with an attached search engine might be interested in this)
- how to please all your customers at once - the forgotten art of making your software configurable. Bonus session: sensible defaults (audience: every modern software company)
- your customers might not be as smart as you but most of them aren't braindead - and other notes from the field (same audience)
- continue to be the best - by not nerfing your almost-perfect product (same audience)
- not storing data you don't need to serve your users [0] - a practical guide to not getting fined by EU while also getting rid of dumb disclaimers that won't help you anyway when regulators get fed up (audience: everyone who has a GDPR popup with default opt in to one or more likely more than 100 different trackers)
- why customers who have already bought a dishwasher won't buy a new one however much you advertise for them the rest of the month (and other great secrets from marketing 101). (Audience: most adtech companies and ad buyers)
- the long-term benefits of serving ads that actually provides value - or why you'd want to sell some ads to other companies than scammy dating sites even if they are currently the easiest ones to fleece by showing expensive ads to customers who aren't interested at all. (Ad-tech companies, and that big one with the attached search engine in particular.)
Master class:
- 1/3 Living in a global world: Europeans are also on the Internet and knows when movie and series releases are + other shocking truths.
- 2/3 increasing digital sales by putting the content up for sale - how to increase sales and customer happines all while reducing piracy just buy removing stupid geo limiting
- 3/3 the synergistic effects of making your content available to customers who wants to pay for it while it is still hot
[0]: Deliberately talking about users, not customers here to avoid the ever-helpful response of "if you are not paying you are not the customer, you are the product". And BTW, I agree, it just isn't always helpful.
Same as YouTube's algo now recommending already watched videos, over and over.
This is either being gamed by bots or someone at those cos is not dogfooding. And if all else fails, for the love of god, give me an option to train your dumb algo manually by blocking shit I don't want to see ever again.
Feel free to recommend my course posted below :-P
"how to please all your customers at once - the forgotten art of making your software configurable"
I don't believe amazon's recommendation engine has ever delivered any value to me, and it would be really nice to just have all that crap removed.
so while they do somethings well there are just some odd choices made with how they present product. oh, with regards to that "see something new", reported it and got back that those thumbnails were not meant to represent actual products. call me confused
I don't remember ever looking at their recommendations and finding them useful. Why would anyone want to use the same algorithms?
Because if I'm an Amazon dev I'm 100% not just showing a 45 year old woman who buys a TV the same "other people bought" as a 25 year old man. I'd definitely run the same ML on each section, because the whole point is to understand the closeness of connections between personal traits and purchases, and it is impossible for a human to outperform in that task.
Something simpler is more effective in this case, in that it gives me much more relevant results. Why is that unfortunate?
Or perhaps you think they might work well for your particular sales domain. They clearly don't work very well for product categories with lots of different items which fulfil the same purpose (e.g. one thousand kettles, of which you only need one, or one thousand TVs) but they might work great for funny slogan T-shirts or for board games.
"hey, Amazon has recommendations and they're a successful shop, so we should have recommendations too!"
I read a good post about this using washing machines as an example but I can't find it.
I mean, I understand Amazon shows me a washing machine while I'm searching for one and I clicked on several without buying, but not when I already bought one! It's not something I would buy every other week...
If I just bought a big pack of batteries I won't be needing batteries for a long time; and they should know this because they know that I buy batteries every 18 months or so.
Also, I don't have a dog, I have never bought anything pet related on Amazon (or any other website for that matter) and yet they keep suggesting pet food.
Amazon recommendation engine is abysmally bad. Picking a random item from my orders history would be just as good and so much simpler to implement.
Patrick explained it why it might be like so: https://mobile.twitter.com/patio11/status/982208307057246209 (thread: https://threadreaderapp.com/thread/982208307057246209.html )
> I don't remember ever looking at their recommendations and finding them useful.
Amazon does various forms of recommendations (no way limited to just recent purchases) based on
1. Purchases
2. Searches
3. Browsing history (via trackers, ads, affiliate links)
4. Activities (on their digital devices, tagging things on their websites in wishlist, adding preferences in your account etc)
5. Global and Local trends (people in your area/country buy..., people with similar buying pattern buy..., It's Eid and you're Muslim, so you buy...)
...And probably many other signals I might be missing.
They must have got pretty good people working with this and enough data to get their models right.
Also, Amazon makes decisions on new retail businesses to start depending on those signals. https://techcrunch.com/2016/11/03/amazons-private-label-bran...
That, and considering the fact that Amazon has an unshakeable culture of making data-driven decisions for everything it does...
I guess what I'm trying to say is, Amazon and its subsidiaries may not be as dumb at AI/ML as it might seem to be by gauging against one datapoint.
Also see: https://www.cnbc.com/2014/04/09/big-data-knows-youre-pregnan...
My only explanations are that that kind of thing happens when they can't figure out anything good to recommend.
Perhaps they are not.
Perhaps the best they can do, is to recommend to all of us the same/similiar product again because thats the only thing they got working: Catching those x percent of people who send the tv back and buy the second recommended one.
You seem to be taking this an an axiom and running with it. My impression too, is that the recommendations are pretty shoddy.
- yes, people who just bought a lot of batteries are actually more likely to buy more batteries soon. same for people who just bought a fridge.
- these recommendations may cause some sort of "burn out", where people stop looking, responding, or decide to unsubscribe. But this wasn't much worse than other recommendations. including the awesome ones.
- the loss from burnout was greatly outpaced by the wins in actually targeting the right users
Yes, these are bottom-of-the-barrel recommendations. But they work. And people don't care that much even if they disagree. Maths said so.
I think this is a similar story to the complains on non-literal searches in search engines. People think they want one thing, but they want something else.
Okay, that makes sense to me. They go from 0.02% likelihood of wanting to buy a fridge to, let's say, 2% likelihood. A marvelous leap!
But even though they are now much more likely to buy another fridge than the average person ... I would think that they would be more likely still to buy other types of products than to buy another refrigerator.
Even if their likelihood of needing to buy more consumable products (e.g. deodorant) is only 5%, that's still double the likelihood of their likelihood to buy a fridge again.
So I would think that there would be some bias toward more frequently purchased items, even if your likelihood of buying a less frequently purchased item does indeed increase after purchase.
I'm sure the math checks out somehow, and that the creators of the recommendations algorithm wouldn't be pursuing strategies that don't work, but I don't think Patrick's explanation fully captures it.
Everything else sucks though.
Recommending for similar products at the end of a purchase raised awareness towards those items for the next inevitable batches. I believe amazon wants its own brands to win or otherwise a strong competition among others producers.
I can't believe it's just math.
Regardless, this could be a useful service for some but I do think you would need to worry about things like GDPR if you are planning to offer this to amazon. I'm assuming they've thought this through and are providing ways for companies to use this without getting in trouble. For the same reason, I'm pretty sure that Amazon is going to be very careful not exposing themselves to legal trouble here as that could become very expensive for them. So, I'm not so concerned with them trying to grab the data for their own purposes.
They are certified under HIPAA and PCI/DSS in terms of their isolation of customers data from other customers and their infrastructure management.
Anyone who thinks they would deliberately attempt to steal customer data is beyond paranoia.
With a bit of effort anyone can find products on Amazon that sell well (there's services that will do this) and then try to compete with those products. Amazon undoubtedly has more insights than those services provide, but that's probably not what makes or breaks their Basics products.
Their strength is probably their manufacturing costs being much lower than the average 3p startup seller and the brand itself.
I don't know why by I love these Basics lines that are being produced by Amazon, Zalando etc. They're positioned perfectly between "brands that are overpriced" and "Chinese brands selling probably crap products with probably fake reviews". I have Amazon basics towels, camera tripod, water filter, toiletry bag and more.
The fact that they can make sure it makes is prominent in any search result in the category, even for a competing product, is what makes it, combined with the fact that it's the one thing on Amazon you can be 100% sure isn't counterfeit, size no one else can sell it so there is no risk of commingling. It's nice that Amazon has found a way to trade on the unreliability of their own marketplace, and to leverage dominant market power in online into marketshare in any particular product market they choose.
Great for Amazon, that is; not so sure it's good for anyone else.
A reason not to use AWS as a competitor of Amazon is that you wouldn't want to give them cash they can invest in their other business lines.
I worked with Amazon recs on one of my teams. We all complained about how inane the recommendations could be on a regular basis. So I doubt it's a perception issue.
Of course, they could opt to not show recommendations due to a lack of data or pad it with what's currently popular with other customers.
Maybe enough people return their fridge right after and buy a new one, or are flipping houses and need to buy multiple washing machines, or are moving into a house with more rooms and new multiple TVs, or decide to buy a mattress as a gift for their parents after buying one for themselves, or any other of the hundreds of events that some exceptional to you or I but happen frequently enough in the aggregate to actually make it a normal enough occurrence to still recommend items that are usually one off purchases.
Or maybe one of the most valuable companies in the world that has gotten to that point largely off of selling us crap is too stupid to realize that people don't usually buy 3 fridges or is unable to hire the talent to tweak the models to reflect this sort of behavior. I've got no firsthand knowledge, but I feel like the former is more likely than the latter.
A/B testing a bad model does nothing. I'm sure Amazon's recommendations are the "best they've got" given their tech. Which sucks.
Google runs a massive recommendation system at scale: their ad placement product (in Search and elsewhere). Google's recommendation tech is astoundingly better than the equivalent at Amazon, rapidly putting the most profitable ads in front of the best potential customers.
That said, Google will never release their ad tech recommendation system "as a service", since it would kill their golden goose. That whole part of their company is a black hole of information.
The rest of the product is largely under MAP policies excluding Amazon, and the products take a level of care and branding a bit more than Amazon's typical white label products. So, could they, yes, but we have plenty of bigger operational risks
Anyone know if there's something in the AWS contracts about Amazon using your data?
It's horrible for moderately popular items though.
Going a step further, if I was buying the Mona Lisa, would it know to recommend me other Da Vinci items, or other renaissance items, or would it fail to make that inference due to there being no co-occurrences in purchasing behaviour.
I built a very basic recommendation engine myself but would love to try out something with actual brains behind it, just being curious about the results.
You can share all the information you want with other entities under GDPR as long as the information you are sending as long as you can't uniquely identify someone from that data _without_ additional information.