Amazon Personalize: What Marketers Need To Know

A reader asks:

“What is the deal with Amazon Personalize? Most of the information I see about it is from Amazon or sites regurgitating what they have on their website.”

Amazon Personalize allows your developers to build real-time recommendations, in the same(ish) way that Amazon does. This includes specific product recommendations, similar items, and product reranking. It also includes personalized offers/promotions, personalized content, and personalized notifications for your email, SMS, and offline/direct marketing efforts. You can read more about Amazon Personalize here.

 

 

How does it do this?

In a nutshell, it does this by coupling your data with its proprietary machine learning models.

Amazon Personalize is a fully managed service from data processing to building, training, and hosting the models.

 

Can everyone use their models, or is it only the biggest of companies? The ones I see listed on their site are much bigger than we are.

This question comes up a lot because of how they showcase studies/examples. If you use developers (in-house or outside vendors), you’re likely golden. If you’re concerned, you can always get a free trial account and play around with it. One of the coolest things is that your developers don’t need machine learning (ML) experience to use Amazon Personalize. (This is a BIG benefit.)

If you are a small company using one of the more common eCommerce platforms, you’re likely to find apps/extensions/plug-ins that say they use Amazon Personalize. Some of these are complete garbage (having nothing to do with AMZ Personalize) or big performance drains, so do your homework before adding one.

 

There are so many other personalized services out there. What are the benefits of using Amazon’s service?

Amazon Personalize is made of the same machine learning models Amazon has used for years. Chances are you buy from them or have played around on the Amazon site, so you already know they’re solid. Just as a reminder, nearly 35% of Amazon’s sales come from personalized recommendations (McKinsey).

It’s quick and easy to get started. Instead of waiting while your IT team scopes, develops, tests, and trains the recommendation models, you’ll skip the line and be up and running in just a few days. (It generally takes longer for your in-house legal counsel to sign off than it does to deploy the models!)

It’s designed to be as easy for your developers as possible. As mentioned above, your team doesn’t need any ML expertise to use it. Amazon takes care of the training, optimizing, and hosting the models. The APIs are streamlined. The experience is pretty plug-and-play allowing you to go live in days, not months/years. (This is way more beneficial than it sounds.)

There are no minimum fees or upfront commitments, and it’s a pay-by-use pricing model. They also have an AWS free tier valid for usage for the first couple of months. Want more pricing information? Click here to get yours. (And no, this is not an ad/sponsored post.)

Incidentally, two other big pluses are that they offer real-time AND batch recommendations (instead of just batch.) Plus, the integration with existing tools is typically quick and easy.

 

What are the negatives of using Amazon to do this? {Redacted Vendor} said that I’d basically be putting a high-speed IV drip of all my blood directly into Amazon’s veins.

Amazon states that “all of your data is encrypted to be private and secure and is only used to create recommendations for your customers. Data is not shared between customers or with Amazon.com.”

Amazon Personalize is a game-changer in many ways, and their competitors are striking back HARD. Sending all your data to Amazon on a silver platter is a controversial topic. Some people have absolutely zero concerns, and others would rather have lobotomies. You know in your gut what’s right for your business. Please remember, AI is a long game, but the sooner you get started, the better you’ll be able to join in and play it. Even their harshest critics can’t disagree that their suite of machine learning tools can give you a VERY big jumpstart.

Back to your original question… Some companies prefer to use other vendors because they want Marketing to be in control, not their internal development people. Companies with many rules and regulations (financial, healthcare, etc.) often don’t use any outside vendors for AI/ML tasks. Neither are necessarily Amazon’s problem, but they are reasons you might choose to build your own thing.

Other companies want more insight into how the sausage is made, and you don’t get that with Amazon. (In other words, they don’t give you the recipe with all the ingredients and steps to recreate their models exactly.) Some companies also feel they need to be more in charge of the training and optimization instead of using Amazon’s secret sauce. (You often see this when people want to bias their model(s), especially at the beginning.

 

Ok, I get all that, but what would we use it for? Got any specifics?

Developing hyperpersonalized product recommendations based on the users’ browsing/carting/buying/etc. behavior. This includes Recommended for You, Frequently Purchased Together, and Customers Who Looked at X Also Looked at This/These.

Personalized reranking. Personalized ranking builds lists of recommended items re-ranked for a specific user. Amazon consistently features this as one of its top benefits. Many people gloss over it, but it’s one of their best features especially compared to the competition. If you’re looking at more than one vendor, you should pay close attention to this. Real-time ranking/re-ranking can make a measurable difference, especially during peak times. (Holidays for B2C retailers and direct marketers, for example.)

Creating highly customized content. With Amazon Personalize, you can show your users individualized content based on various criteria. This isn’t just an eCommerce thing – you can use it for various projects in multiple industries. Ecommerce, Publishing, Hotels/Travel, Banks/Financial Services, Dating Sites, Social Networks, etc.

Delivering personalized home pages for every individual. (This can be a real money-maker when you combine your channel information.)

Building individualized promotions. You can serve bespoke offers to your users based on their behavior.

Developing batch recommendations for your email and SMS messages.

Building offline direct marketing/catalog campaigns. Can be very useful if you are trying to integrate your online and offline efforts better.

Designing a better onboarding experience. Building more robust recommendations for new users. (This is more critical than ever.)

Offering contextual recommendations. This is another one of the things that Amazon does better than many of the other players.

Personalizing your eCommerce and/or video streaming apps.

 

My vendor said that I could only use old/past data?

Incorrect.

You can use historical and real-time data.

That said, there is a minimum number of data transactions required for specific events. Depending on your use case, you may need to get additional data before generating some of their recommendations.

One of the things that Amazon’s machine learning models excel at is delivering personalized recommendations and content for new users and new items. They don’t pitch this well, but it’s something to consider if customer acquisition and/or buyer reactivation are areas where you struggle.

 

Can I only use Amazon Personalize on my website?

You can use it on your websites and apps. You can also use the information in your email and text messages and your direct marketing/catalog efforts. (Because this seems to be a source of confusion, you’ll receive the information so you can port/funnel it to funnel into your existing email/SMS/backend/etc. software.)

 

Is everything done ONLY in batches? That’s what our current provider does.

Depending on the project, you can do it in real-time or in batches, whichever you prefer.

 

How hard is it to get started?

It’s not.

You create an account and go through the set-up wizard in the developer console. The steps are easy, and you have solid choices along the way. (For example, you can let them choose the suitable model or manually do it if that’s your preference.)

The biggest hurdle most folks have is getting their data ready. This is not an impossible task by any means; it’s just one many squawk about. The models are trained based on your data, so it’s vital that you get them the different types of data that they require. (Item data, user data, activity data, etc.)

Have questions about Amazon Personalize? Already using it and have a tip you’d like to share? Tweet @amyafrica or write info@eightbyeight.com.