End-to-End Personalization Means Real-Time User Insights Everywhere

Unlike many startups, we didn’t start Miso.ai with the idea that we wanted to build an entrepreneurial solution to a big market problem for fame and fortune. We started it with cornerstone technology. Specifically our CTO Andy’s doctoral dissertation research on how to build real-time personalization systems with just the data of a given site’s clickstream logs and product catalog. The ability to learn what users wanted, even at cold-start, not incrementally but continually, in real-time, click-by-click was incredibly exciting and reason enough to start the company.

In those early days, Andy and I spoke often about what to build and who we could help with this tech. Inspired by the power of personalization-first platforms like Spotify, Netflix, Google, and Amazon, we could see how even community-centric platforms like Etsy, Airbnb, and DoorDash were advancing their success through personalization. These companies had all won their markets, and they all shared one key aspect: they are data-driven to the core, treating granular user insights as the lifeblood of their business.

In this sense, product management and product marketing are two sides of the same coin when it comes to launching and promoting a successful platform-scale product. Product managers (PMs) need to understand their customers and develop a product user experience (UX) that meets their needs, while marketing teams need to effectively promote that product to the right audience at the right time. Both teams rely heavily on access to a steady pipeline of user insights to power, personalize, and optimize their efforts. 

In this post, I’ll break down how Miso.ai is supporting this from a very simple point of view: every user insight we generate is your data and should responsibly and securely be accessible at any time to power and support your product and business from end-to-end. 

How Does Miso.ai Generate Users Insights?

Miso.ai generates real-time user insights on every user, whether they’re a returning champion or a visitor, by combining two key signals: 

1. Your product or content catalog, which we examine on both an unstructured basis (i.e. semantic analysis of your raw text or PIM descriptions) and on a structured basis (i.e. the metadata and explicit features and aspects of the listing)
2. Your clickstream logs, which give us a clear picture of how real users and visitors searched, browsed, and navigated their way to any given product, and what actions they took after that. This user behavior data is critical to personalization because it’s the clearest signal of not only what each user engaged with, but on the aggregate level, which products or content were trending and popular at the moment

With these two inputs, we’re able to generate a fine-grained model of each user or visitor’s interests and intent. First, your catalog listings themselves are modeled into embeddings that we can cluster and then map in vector space. Similar listings, those with similar features and attributes, will cluster together. Those that aren’t similar won’t. We take the same approach initially with users— those with similar attributes cluster together and others won’t.

However, it’s the behavioral data of the clickstream that ties everything together. Rich signals about how users search, browse, and click through on a specific product or piece of content tells a story about their interests and intent at the moment. Like a gravitational pull, every clickstream event pulls a user closer to their favorite products and brands. These affinities have a decay function, but you get the idea. By understanding a product or content catalog richly, and by having the ability to model historical and at-the-moment user behavior, we’re able to extrapolate a lot about a user’s affinities.  

As such, for any given user, we know their top interests and the features of every product or piece of content that will make them attractive to a user or not. We even predict which top products they’re likely to be most interested in, their probability of making a purchase, or churning their streaming or content media subscription. With Miso.ai,  all of this data is yours. But more on that later.

How Can Real-Time User Insights Shape Your Product and Marketing?

Having an engine that predictively intuits what any user or visitor is interested in and likely to engage with means you have an engine for real-time personalization anywhere. A great example of this is our ShibaMart marketplace demo. As a product team, we sit down every two weeks and brainstorm new personalized features, search updates, and SEO upgrades we can build into this demo—many of which you can try for yourself here. 

Bringing these ideas to life isn’t a months-long process of ML model development, API building, and haphazard QA/QC and user testing. With a clear picture of what we want the product UX to look and feel like, we just have to parameter dial in an API call to one of our search and recommendation engines and then code up the front-end and go. Across all of these touchpoints, every user has a UX that is tailored to and reflective of their interests and what is likely to generate conversions. Compared to popularity-based rankings or simple personalization schemas, this type of 1:1 real-time personalization converts. In fact, Miso.ai partners typically experience a 10-20% ARPU lift in their first 30 days.

The same goes for marketing personalization. Being API-first means our Miso.ai engines are agnostic to what your frontend UX and marketing channels. If you need to personalize the ranking or selection of which products or content to display in a given email, Dynamic Product Ads, or native ad banner, we got you. Just like your platform UX, every marketing touchpoint can be highly personalized and reflective of what a user is most interested in and likely to engage with. 

This is true for personalization of your first-party marketing channels: such as emails and push notifications. Retargeting ads have gotten crazy expensive over the last few years and first-party marketing has never been more important. Rich user insights and a dynamic personalization engine means your emails or push notifications can be personalized to build on a user’s last actions or current interests, leading to their next key action. 

Emails and push notifications are ultimately free. Their true cost is that if you abuse or mismanage them, users will turn them off and label them as spam. But with a solid, nuanced, and personalized strategy, we’ve seen CTR increase by as much as 2X.

There’s another aspect to having an insights datastore in the first place: smarter product marketing campaigns. Like I mentioned earlier, digital advertising has gotten crazy expensive, and you can’t just light money on fire retargeting to every visitor or user on your site. With Miso.ai, it’s easy to generate a cohort of users who have a high probability of making a purchase within 7-14 days, with a high predicted AOV. The incrementality of retargeting starts to make sense because the ROI is tangibly and predictively positive, considering we can use Miso.ai’s personalization API to select which products or content to show users in dynamic retargeting ads.

Additionally, good user insights mean good business intelligence (BI). Understanding what’s trending, who’s interested, and the potential upside on sales and retention is invaluable. It can spark new ideas for new products to source, features to build, and marketing strategies. This isn’t about taking the creativity out of our jobs, but instead, the guesswork of non-data-driven trial and error. 

How Can PMs and Marketing Teams Access User Insights? 

Miso.ai partners have 3 ways to access user insights from us:

1. User Insight Reports: This is the most common and easy-to-use approach for teams. PMs can choose the cohort parameters they want to focus on and generate an export that shows each user's top interests, strength of these interests, top listings of probable interests, and likelihood of a purchase. Marketing teams can use this data to segment their target audience and tailor their messaging and promotions accordingly. Stay tuned as we’re working to up the ante on this in our Dojo dashboards very soon!
2. Datastore Model: In this model, the analyses are refreshed regularly and stored in a simple access point, like an AWS S3 bucket. This allows PMs and marketing teams to quickly and easily access the latest user insights directly and perform any extract, transform, load (ETL) and BI they want.
3. Segment CDP: Segment is a customer data platform that allows teams to manage their customer data in one place. By sending user insights data back to Segment, PMs and marketing teams can get a more comprehensive view of how their customers are interacting with their products and use this information to guide their product and marketing decisions.

Whichever way you choose to use Miso.ai, our philosophy is that your data is yours. A critical part of our commitment to being an awesome infrastructure partner is to support you at all levels.

Final Thoughts

We’ve unpacked a lot here, but the general rule is simple. Having a great personalization engine can be an amazing driver for growth for your platform and your business. Ultimately our goal is to fly wheel insights about your users into a better personalized product experience, which drives more insights about your users, and thus better marketing, more retention, and engagement. And it’s all built responsibly and seamlessly with the data you already have.

If this is what you’re looking for, drop us a line anytime. If you have questions or would like to trade notes, you can reach us at hello@askmiso.com or feel grab time to chat with one of our growth PMs here. This is an exciting space with endless potential, and we’d love to get your take.