When it comes to data, we tend to think in dichotomies: quantitative vs qualitative, objective and subjective, messy and curated, science and story.
Using data for product development does not have to be an either/or; instead, it should be YES, and…
We know that the product development process usually consists of a few important phases: planning, design, development and launch. To increase engagement and conversion rates, this process undergoes multiple iterations which developers seldom navigate by looking at the data. Usually there’s somebody else, be it a product owner, marketing or sales person, analyzing it and feeding developers a feature list needed for the next product release.
There lies the gap between developers, marketers and users which leads to lots of guess-work. Consequently, the product development feedback loop is broken.
That’s why organizations have started developing metrics that go beyond the default off-the-shelf analytics provider. Rather than relying on single signals — say the number of page-views or the number of clicks to determine conversion — the trend is now toward using Machine Learning that draw insights from multiple signals or event streams.
Let’s look at some AI-enhanced products and initiatives in this space and while we do that, I want you to think of your very own mobile or web applications and how you could apply some of this stuff to improve your product’s UX.
Use Case #1 — Predictive App Actions
Google Duplex — a new technology for making calls on our behalf was presented this year at Google IO.
And if you are like most people you’d either be super excited or really scared 🙂 I have yet to see a person that’s genuinely both excited and scared at the same time …
Now, obviously this technology is under development, but the thing that I want to talk about which is more applicable today is App Actions and specifically Predictive App Actions in Android P.
Last year Google introduced the concept of Predictive Apps for Android: a feature that places the next apps the OS anticipates you’ll need on the path you’d normally follow to launch that app. And it’s proven to be very effective with a 60% prediction rate.
With Android P that was just launched a few weeks back, Google took it a step further into predicting the next action you wanna take, within those apps.
For example, if you plug your headphones in, your phone will suggest to resume your favorite Spotify playlist. This is incredibly powerful for user retention … and for those of you that build mobile products, I would definitely check this out.
This is happening at the OS level, but how would you go about predicting user actions inside your very own mobile or web application? Could you apply the same concept within your eCommerce website, for example, and improve performance?
Use Case #2 — GuessJS
Apparently your can and what I’m about to show you blew my mind when I first saw it in action. What if we could take this two worlds of ML and Web and blend them together to offer increased performance?
Enters GuessJS — an open source project focused on data-driven experiences for the Web. I won’t go into technical details, although those are fascinating, instead I’m going to explain the high level concept.
GuessJS takes advantage of the Google Analytics data and uses Machine Learning for modeling Next Page predictions to dynamically prefetch next pages a user is likely to visit as they browse. It improves page load performance by prefetching/preloading resources for that page. Thus, GuessJS introduces the notion of data-driven loading for websites.
As a result, the loading is instantly … imagine what this can do to your eCommerce conversion rate 🙂 You can actually figure out where you are on a Conv. Rate vs. Load Time chart and put a dollar value on the amount your business is losing today or compute the extra revenue you’d be getting by improving your page load times.
GuessJS works with Angular and React and you can use it in your own web applications today: https://github.com/guess-js/guess
Use Case #3 — Predictive Marketing Campaigns
Another interesting use-case is building predictive marketing campaigns — this is exactly what the marketing team at Invision did. And if you don’t know about it, Invision is a prototyping and design management platform with over 3 millions users.
As a marketing team, their focus is specifically on acquisition and activation of users on their various plans. Last year alone, they sent over 500 email campaigns and ran 250 ads, across 6 different channels to drive growth.
Rather than inefficiently using all their marketing spend on all users, they used Machine Learning to identify the top 25% of users most likely to create a prototype in a given week.
They then set up persistent marketing campaigns to target only these 25% of users in their Facebook ads and Hubspot emails and in just one week of running the campaign, they found that email CTR increased by 40%, and ad CTR in Facebook rose 70%.
This is impressive! I bet you’re already thinking of applying the same approach for your own business … and you’d be right.
Use Case #4 — AirBnB
According to the McKinsey 2016 report, travel companies and airlines, in particular, have 23x greater likelihood of customer acquisition, 6x customer retention, and 19x larger likelihood of profitability if they are data-driven.
AirBnB is a good example of such a company.
In a dynamic pricing feature, AirBnB show hosts the probability of getting a booking (green for a higher chance, red for a lower chance), or predicted demand, and allow them to easily price their listings dynamically with a click of a button.
That’s only one example, but AirBnB implemented a lot of ML strategies that optimize users travel experience: from detecting host’s preferences to predicting customer lifetime value and listing recommendations.
You can read more about them here: https://airbnb.io/
Use Case #5 — MorphL
Lastly, I wanna talk a bit about MorphL — the project we started at the beginning of this year. MorphL is an open-source initiative that uses machine learning to predict user behavior in mobile & web applications and enables personalized user experiences to increase engagement & conversion rates.
How do we do that specifically?
- We integrate with Google Analytics, Facebook Ads, Mixpanel, Kissmetrics and other platforms to identify user behaviors.
- We’re developing machine learning algorithms to build predictive models.
- And then software engineers can programmatically consume these models to build personalized user experiences in mobile or web applications.
Here are some use-cases that we’re currently working on at MorphL:
- Predicting churn rate, especially for SaaS businesses
- Anticipate if a user is going to purchase or not, especially for eCommerce businesses
- Predicting LTV for eCommerce businesses
- Recommender systems for Publishing or eCommerce businesses
More available here: https://github.com/morphl-project/
This enables businesses to be proactive instead of being reactive: meaning that you can offer various incentives (discounts) before the user actually churns or if you identify a user with a certain probability to make a purchase, you can offer a voucher, on the fly, to help with that conversion.
We’re super excited about MorphL and even if we’re just getting started, we’ve already partnered with companies in US & EU to AI-enhance their products. If you wanna follow our progress and see how this product is evolving, this is where you can find us: http://morphl.ai
How to Start with Machine Learning
By now you’re probably wondering: “Yes, I want to implement some of that into my products, but where do I start?”
The first principle of building a great product using machine learning is to focus on user needs. In other words, you need to figure out if the problem you are trying to solve needs machine learning or not.
How do you tell which user/business problems can Machine Learning help you solve?
Usually these problems can be categorized into a few different types (and we already talked about some use-cases):
- User is inundated with too much data
Google and Bing us various machine learning algorithms to surface the best results for the user
- Problems that require complex cognition abilities
A gallery app that automatically sorts photos — needs to be able to detect places, people and things.
- Predictions and Forecasting
Would a user like a story in their News Feed, would a user who bought a subscription churn.
- Anomaly Detection
Fraud detection is a major application.
Like we’ve seen in the case of AirBnB.
- Experiences that interact with Humans
All the assistant technologies being built — Alexa, Siri, Google Assistant.
- Augmenting/Creating New Experiences
SnapChat filters are a great example of how experiences can be augmented using ML.
Once you have an idea of what you want to solve, usually these are the steps for implementing a ML/DL strategy for your business:
- Preparation Stage
- Business Modeling
- Model Rechecking & Adaptation
- Repeat steps 1–4
In closing, I would say this: NO, machine learning is not the holy grail to your problems.
AI-enhancing your product might not even be possible for you. Today.
AI-enhancing your product doesn’t make marketing department optional. In fact, it’s the opposite.
But AI-enhancing products is the future, in fact … as we’ve seen it, it’s already in the present. As product owners, managers or developers, the smart thing about it is to understand it & embrace it … not fear it or fight it, hence my invitation to you all: “Let’s make AI happen!”