Predictive Analytics 101: How to Use Data to Peer into the Future
Even with the best predictive analytics in the game, we can't see into the future with 100% accuracy. But in spite of that—and luckily for us—humans are predictable creatures. That means that you can use your users' past behavior to get a fairly accurate understanding of what they'll do next.
In product management and data analytics, we call this process “predictive data analytics,” and it is among the most valuable tools for building products with amazing UX.
So, let’s examine predictive analytics and how you can use it to optimize your UX.
What is Predictive Analytics About?
Predictive Analytics is an advanced data science approach to analyzing the product and user data at your disposal to predict (hence the 'predictive' bit) the actions that your users will take and the way your product will perform in the future.
Essentially, it entails looking at your data with the help of statistical models, specialized tools, and machine-learning algorithms to find certain behavioral patterns. Assuming that these patterns will stay relatively stable in the future, you can extrapolate your data based on these patterns and make educated guesses about future usage.
The benefits of predictive analytics include:
- Better quality of decision-making: With a clear vision of what’s coming next, you will be able to make decisions that are more likely to lead to success.
- Better efficiency: Forecasts will also let you see the potential efficiency problems that you will encounter in the future if you keep everything as-is. So, you get the opportunity to improve your efficiency in advance. This can help you lower costs, lower pricing, and gain a great competitive advantage over others.
- Better risk management: Certain types of predictive analytics help you see upcoming business/legal/financial risks. Again, you can prevent these risks before they materialize.
- Better user personalization: Personalized features and content are the pinnacle of UX. With predictive analytics, you can delight your users with suggestions and experiences that best match their needs.
Sounds fantastic, right?
Well, yes, but there’s a caveat. It’s impossible to have a 100%-precise prediction of future outcomes. We’re not fortune tellers from a fantasy RPG game, after all.
Whatever forecast you get will be 70-80% precise at best. So, please keep this in mind when making important decisions based on your results.
Examples of Predictive Analytics by Well-known Products
Lack of absolute precision, however, doesn't mean you can't benefit from these predictive learnings. In fact, predictive analytics is responsible for some of the most famous product successes out there.
Here are three success stories from big tech companies to prove my point.
Example 1: Netflix’s Famous Recommendation Engine
One of the main reasons people love Netflix is its ability to recommend great shows. In fact, the recommendation engine is one of the two key reasons behind Netflix’s success (the other one being its fantastic content delivery architecture).
Netflix uses artificial intelligence and statistical techniques to understand each user's preferences and recommend shows that they are likely to enjoy.
What they essentially do is look at the shows you have watched in the past, whether you skipped them or watched them until the end, and your general interests in your area to predict the categories and types of shows that you might like.
Heck, they even give you a personalized thumbnail based on your interests.
If Netflix sees that you watch lots of movies/shows with a certain actor in them, it will give you a version of the thumbnail where that actor is in the foreground. (And if you didn't know that, now you won't be able to un-see it!)
Amazon’s Anticipatory Shipping
Remember when I mentioned how you can improve efficiency with predictive analytics? This is exactly what Amazon does with its anticipatory shipping process.
As it turns out, human purchasing behavior is quite predictable. We tend to buy certain products at regular intervals (e.g. laundry detergent every month) or with certain seasonality (Christmas decor, etc.). Lucky for Amazon, they have all of this data and can run predictive analytics on it.
With the help of their statistical models, they can pre-ship the right types of products (and the right amount) to their warehouses. This lets them significantly decrease the shipping time to end consumers and lower the quantity of products that they need to ship back to another warehouse where they are more in-demand.
Here’s what the process looks like on their patent papers.
In this particular diagram, Amazon shows the process of changing the destination address of the product in real-time.
Google Maps’ Traffic Predictions
I love the traffic feature in Google Maps. It’s incredibly accurate and fast.
And do you know how they have achieved that level of accuracy? That’s right—predictive analytics!
Google has data on nearly everything that happens in any given city. The vast majority of people behind the wheel use a device powered by Android or any other Google service with geo-location turned On. So, if they see a group of smartphones moving slowly on a highway, they tag it as a traffic jam.
But Google is also capable of far smarter solutions when it comes to predicting traffic. Because they have data on the major events happening in just about any given city (e.g. a large sports game), they will consider that when planning driver itineraries as well. For instance, if you have a 1-hour ride across the city and there’s a game starting soon, Google’s algorithm will keep you away from the major roads that are likely to be overrun with sports fans by the time you reach that area.
So far, we have looked at predictive analytics as a generic tool in the hands of companies. However, today's main topic is applying this fascinating technique to improve your UX. So, let me move on to explaining how you can take your UX research to a whole new level with predictive analytics.
Predictive Data Analytics Techniques For UX Research
The area of predictive analytics is quite wide when it comes to the different techniques that you can apply to solve everyday problems. But today, I want us to focus on the three that are well-applicable to UX research—regression analysis, decision trees, and time series analysis.
Regression
If you have a business or math background, this term might be familiar to you. Yes, it is the classical statistical regression model that can recognize patterns in your data.
In more practical terms, regression (more specifically, linear regression) is your trendline—a line on the chart that shows whether the user activity you are analyzing is going up or down.
Trendlines are something that my UX team and I use all the time on our Amplitude charts because they help us turn messy data into visible trends and see if our UX changes have had any positive results.
Here's an example of regression:
We streamlined our acquisition flow with the UX team three months ago and want to see if the number of signups has increased since then.
This is a user sign-up chart for the last 90 days, can you tell me if it is growing or not?
Of course you can’t, it’s just too messy.
But, let me change the chart measurement type to a formula from the sidebar.
And add a trendline.
Now, the signup chart looks like this.
With the trendline added, we can see that our changes had a positive effect on our acquisition. However, this change is too small to be statistically significant, so I would argue that our UX changes have not made any real difference.
Decision Trees
Decision trees are among the most popular AI model structures for analyzing your data and making useful conclusions out of them.
A neural network, in this case, takes your original data set and starts splitting it into subgroups based on certain conditions, then splits these subgroups into other groups, and so on. As a result, you get something that looks like an inverted tree.
After building this tree, the algorithm will give you an answer based on the data you have. In the example above, the decision tree will tell you whether you should take the bus to school or walk.
Now let’s see how we can solve UX problems with these models. Imagine that you are trying to decrease the cart abandonment rate of your eCommerce store. When you run the decision tree algorithm on your user data, you get the following result.
Looking at the decision tree above, we can understand the core factors affecting the abandonment rate. In this case, we need to add features/UX that:
- Encourages people to check out before their cart gets bigger.
- Make sure that shipping costs are visible and easy to notice on all sorts of devices.
- Encourages people to log in.
- Speeds up our pages.
We can then continue by using predictive analytics tools to see which of these factors will affect the abandonment rate the most—helping us prioritize these solutions.
Time Series Analysis
Time Series Analysis is the more advanced set of modeling techniques. As the name suggests, this type of analysis is done on specific types of data (mainly historical data that is bound to time).
One of the best ways to incorporate time series analysis into your UX process would be to predict your key UX metrics (e.g. engagement, churn, retention, etc.) based on the user behavior from the past data and see if they are bound to improve in the future or not. If the predictions show you a decrease in your metrics, then you take it as a signal that something is wrong, and you need to fix it.
I won’t go too much into detail on how to do time series analysis here, as we’ll be exploring the Forecast feature of Amplitude up next that utilizes this methodology.
Best Machine Learning Models And Services For Predictive Analytics
Along with all the benefits you get with predictive analytics, there is one major downside—they are notoriously hard to do manually. Luckily, the internet is full of various research tools and solutions that can run these analytical models for you with the click of a button.
So, let’s see how you can take advantage of different types of tools to predict the future of your UX.
Using Product Analytics Tools
The fastest and easiest way of doing predictive analytics is by using relatively advanced product analytics tools at your disposal. For regular tasks, it’s usually Amplitude. For more advanced workflows involving data mining, you can consider advanced predictive analytics tools such as Domo or Tableau.
Currently, my daily driver for such use cases is Amplitude. So, let me show you how you can use some of its features to predict your product’s future.
Predictive Cohorts
The first feature found in Amplitude that can help you with this is called “Predictive Cohorts”.
Amplitude will use its prediction AI models to create user cohorts based on their likelihood to perform certain actions or behave a certain way.
You can later use these cohorts in your analysis by creating charts and comparing the behavior of people in your cohort with the rest. You can also run A/B tests based on these cohorts to see if the user experience changes you make affect their behavior or not.
How to Create a Predictive Cohort in Amplitude
- First simply open the cohorts section in Amplitude and click on “Predict a Cohort”.
2. Then, you can start defining the behavior you are interested in.
In this case, we want to group people who are most likely to purchase $25+ worth of songs and videos.
Based on your requirements, Amplitude’s models will analyze your past and new data and give you a prediction that you can access in the cohorts section.
3. We can turn this prediction into a usable cohort by clicking on it and saving it as a cohort.
4. If you want, you have the option to fine-tune the selection by picking which percentile of the user base you want to include in your cohort. But I would recommend you stick with the default selection and save it by clicking on the “Save as Cohort” button.
Voila! Now you have a dynamic cohort based on predictive models that you can use practically anywhere in Amplitude, including your charts, analyses, and A/B tests.
Forecasting
Another feature at Amplitude that has come in handy for me and my UX team is their native forecasting engine. It automates the process and uses both regression, time series analysis, and its own set of AI models to give you a prediction for your metrics of interest.
From my experience, one of the most practical applications of this feature for UX is the forecast of activation funnel conversion rates.
How to Forecast Activation Funnel Conversion Rates in Amplitude
1. Imagine that the activation funnel for your Financial Services CRM tool consists of three events—signup, add integration, and create opportunity. If we look at the conversion rate over time, it will look something like this.
Now, you have already implemented a couple of UX improvements in the journey a week ago and want to understand if it has made any tangible improvements. For that, let’s ask Amplitude to give us a forecast for the next 30 days.
2. Click on the “Anomaly + Forecast” button and then click on the “Add forecast” button afterward.
Amplitude’s models will load and give us a chart looking like this.
As we can see, there is no visible growth in our funnel conversion rate (which is more common than you think). The actionable insight here is to have another round of UX improvements.
Using Cloud Predictive Model APIs
Using product analytics tools for predictive data modeling is quite handy when the problem you are trying to solve is relatively simple. However, if you need something more advanced, you can opt for third party machine-learning model APIs.
Here are two providers that my team has found useful in the past.
AWS Machine Learning API
Amazon’s cloud service is rich with various business intelligence AI models that you can connect to via API. The stack also includes predictive analytics models.
- There’s Amazon Forecast, which can run on time series data—letting you see the effect of your UX in the long-term future.
- There’s also Amazon Personalize, which gives you a scalable and customizable personalization engine that you can use to suggest relevant products and services to your customers (just like Netflix and TikTok do).
Google AI Infrastructure
Unlike Amazon, Google does not provide pre-made prediction AI models. Instead, you get access to their massive set of tools that can help you do big data analysis and build your own custom deep learning prediction models with the help of your current data scientist team.
Google’s tools are an especially good fit for handling large volumes of data and findining future trends within it (their data management features are amazing). Apart from predictions, you can also use their toolset for your descriptive analytics needs, lifecycle marketing campaigns, business analytics, financials (e.g. credit scores), and more.
So, if Amazon’s ready-made models are not able to solve the specific problem you have, Google is your go-to service.
A Sneak Peek Into The Future Might Save Your UX
We all wish we could predict the future. If we could, we would create perfect products with the most enticing UX in the world—for better or for worse. But, in the meantime, predictive analysis is the closest alternative for helping us make data-driven UX decisions and delight our customers with great experiences.
Don't forget to subscribe to our newsletter for more product management resources and guides, plus the latest podcasts, interviews, and other insights from industry leaders and experts.