Six Degrees of Personalization

I spend a lot of my time talking and thinking about personalization. If you know that data driven personalization and experimentation is something you should be doing but don't really know where to start or how to get there, then you're decidedly in the majority. Most organizations are doing a fairly good job of collecting a lot of data, a smaller number are excelling at understanding the data. Very few are activating that data in real time.

In this post I aim to lay out six steps that go from the very beginning to using data like the best of them.

1. Get the Data

That sounds easy enough, right? You are probably collecting mountains of data already in the form of web analytics, CRM, email marketing, POS, and more. That's a great start, but there might be a missing step. If all of that data is living in separate silos, it's going to be quite difficult to gain the level of insight that you really need.

There are a lot of great tools on the market that help solve this problem. The (somewhat) new Universal Analytics from Google is a great place to start. There are also a myriad of Business Intelligence tools out there focussed on the problem from different angles such as Looker, Domo and GoodData just to name a few.

A major consideration here is getting all the data into a consistent format that makes sense when you query between these different sources. Depending on your tool, you will either need to do that before pushing the data into the system, or the tool may have data cleaning capabilities. If you have a lot of data to clean up, you might consider a tool like Paxata or Alteryx to make that a lot easier.

2. Understand the Data

Now that you have data, what do you do with it? This is probably the hardest question of this whole post, and the one where organizations are most likely to go wrong. Having loads of data often leads people to find "truths" that aren't there. Spurious correlations run rampant in business and in life, and if you don't take the time to dig deeper, you may be in trouble down the line.

Truly understanding your data, and understanding it well, is going to take time. If you are fortunate enough to have a full BI team in your organization, then you can ask lots of deep questions. If you're new to this and are working within a smaller team, take your time and look at the data from a lot of different directions. Slice, sort and pivot the data in as many ways you can think of until you start to see the patterns like Cypher reading the Matrix.

3. Run some experiments

Perhaps the very best way to gain knowledge about your visitors, or just about anything for that matter, is through scientific experimentation. On the web, we typically call this an A/B test, but your mindset should be that of a scientist seeking to gain insight. This mindset has implications that not all practitioners of A/B testing tools will follow.

Here's what I mean: You can run an experiment on your home page to see whether "Free" converts better than "Get it today" or an orange button beats a blue button. You can start that experiment, and wait until your tool says you have a "winner" (more on this dangerous topic in a future blog post). Once you have decided the blue beats the orange, you can implement that on your button and move on. What have you just learned? That your visitors like the color blue? Maybe. It is very unlikely that you've learned enough at this point if you didn't start with a well formed hypothesis.

Let's say you're interested in increasing clicks on a call to action (CTA) on your home page. Rather than simply making changes, start by stating some assumptions and some possible changes that might effect users behavior. Is the button you want them to click on sufficiently visible? What else might be competing for its attention? If your visitor does notice the button, is it clearly communicating what will happen when they click on it? Do they have any reason to be nervous about what will happen? If you can answer these questions first, then you can move to changes that you could make that, if your hypothesis is correct, would improve your conversions.

Now you're ready to run an experiment, but don't just stop at one. If you think the button isn't visible enough, try changing its color in one experiment, making it bigger in another, and making something else smaller in another. If you're right that people were missing the button before, then you should see significant results with each one of these changes. If you don't see those results, then you've also learned a valuable lesson: It's not about visibility. You can now safely move on to changing wording or other messaging that might be creating friction in order to improve the results, and you may have even learned some valuable lessons about your visitors themselves.

4. Segment and target your audience

Now I'm going to let you in on a little secret: Not all your visitors are the same. Of course, you already knew that. If you stop and think about it, there probably aren't many things more obvious. But due to technical limitations of the past and lack of resources in the present, we often think about our visitors as one big homogenous group and give them all the same content.

Really great websites in the past have addressed this problem through great information architecture, allowing all the different visitors to quickly find the content that they are looking for. Really great websites of the present do that as well, but they go a step further by personalizing a web site to the needs and desires of the current visitor, in the present moment.

More than just serving your visitor's needs though, segmentation allows you to dive deeper into understanding. So once you've built a segment, say drilling down to geography, past behavior or device type, go and run more experiments within that segment and refine the experience for each group.

5. Repeat steps 1 through 4

If you've ever thought personalization or A/B testing were tools you installed on your site, you were mistaken. Personalization is a mindset; a process; a way of life. Ok, maybe not a way of life .. but it's most definitely a process and one that takes time and iteration.

Don't loose heart though, small experiments and small personalization can have huge impact on your bottom line. Focus on learning, refining, and always moving forward, and the return on investment of your time and resources will be well worth it.

6. Trusting the machine

A lot of buzz in the personalization space centers around machine learning and automated content recommendation techniques a la Facebook, Netflix and Google advertising. I find this area of research to be incredibly exciting and in the right situations believe it can be incredibly powerful. I want to give a few caveats here before the excitement runs away.

First of all, machine learning carries with it a trade-off. On one hand, you can have complex blended models that are incredibly accurate in predicting user behavior. On the other hand, you can have predictions that are easy to interpret. Sometimes you can have a nice blend, but the algorithms used by the best prediction algorithms will teach you very little about the behavior of your users, or how to plan your next piece of content.

Second, how many times have you been frustrated with the inaccuracy of a recommendation from Netflix or Facebook? If you're anything like me, it's rather frequent. Now, not many organizations have a data team on par with these, so if you expect that switching on a black box recommendation algorithm will instantly give your site through the roof conversion rates, you will likely have a bad day when the data comes in.

That said, if you have a large amount of great content and a lot of different kinds of visitors, then a manual process of segmentation and targeting for each type of visitor can quickly become impractical. If that is the case, then you should be looking at automated recommendation software, of which there are some really great options out there. Automated recommendations should always be used in tandem with a solid practice of data collection, understanding and experimentation though, so that you don't miss out on the evolutionary process that your site can take described in the rest of this post.

Conclusion

You can do great things with data. Whether you're just starting to think about personalization or have been doing it for a decade, there are always ways of improving. Focus on the process and the learnings that you can achieve from using these tools. Ask good questions. Dive deep into the data. And above all, have fun!