As we talk with people — from thousands of companies all over the world — about their efforts to drive growth through smarter, data-informed decisions, a few clear trends emerge. First, analysis tools must make it easier for everyone in the organization — from top to bottom — to obtain insights. Second, and relatedly, analysis tools must assist all data consumers with achieving more relevant insights by bridging the gap between Big Data and human intuition with machine learning and other “smart” capabilities.
I’m excited to be part of the team that is realizing the vision of these two themes. The Adobe Analytics Fall 2016 release, available now, includes a handful of new features and capabilities that will directly help you drive better decisions through “smarter analysis,” aided by powerful machine learning within Adobe Analytics. In the rest of this post, I’d like to walk you through what our team has done to make your analysis tools smarter.
There’s a lot of great stuff in here, so grab a coffee, and let’s get going. By the way, everything I share in this post is available today in Adobe Analytics.
You can now receive an early-stage “heads-up” that something truly significant is happening in your business. We have updated our built-in email- and SMS-alerting capability to use Anomaly Detection as the default mechanism for determining whether to notify you of a spike or dip in a key metric. Anomaly Detection looks at your historical data to determine an expected range of values for a metric, and now it also takes seasonality and major holidays (including Black Friday) into account. This means that you can expect no more false-positives in your alerts — if your traffic always decreases on the weekend, you won’t be alerted every single Saturday morning. Alerts that you set up based on Anomaly Detection will be triggered only when there is a statistically significant spike or dip.
Another huge improvement here is that you can “stack” alerts. Let’s say, you have six different metrics that a certain team is interested in. You can set up a single alert that includes all six metrics and set it to be sent to everyone in the given user group. This carries several advantages. First, you will only need to manage one alert instead of six. Second, if all six alerts happen to spike simultaneously, the recipients will only receive one email instead of six, meaning it cuts down on what we might call “alert fatigue.” Third, by sending to a user group, new employees who are added to that group will automatically begin receiving these alerts; similarly, employees who move out of that group won’t need to be manually removed from the alert.
You can set up alerts a few ways. You can always go to Components > Alerts to create a new alert or manage existing alerts. But, you can also right-click just about anywhere in Analysis Workspace to set up an alert based on the data point you have highlighted. That’s my preferred way, because it preconfigures the alert for you and allows you to stay in your workflow without leaving Analysis Workspace.
Take special note of the “Alert Preview” window in the upper-right corner of the Alert Builder. This view lets you know how many times the alert you’ve configured would have been triggered based on your recent data. By observing how often the alert would have triggered, you can adjust your anomaly thresholds or other criteria so you are not alerting your users too often.
Distribution of alerts has also been improved. The email you’ll receive looks a lot nicer, and you can send alerts by both email and true SMS. International telephone numbers are supported, and you’ll want to enter the country code as well (as shown in the screenshot above).
Automated Anomalies in Analysis Workspace
If you’re like most of us, you value anything that saves you time. This effect is probably multiplied in the world of data, where separating signal from noise can be difficult for even the most polished analyst. That’s why I am so excited about improvements to Anomaly Detection, as we bring this powerful machine-learning technology into Analysis Workspace.
Adobe Analytics has featured Anomaly Detection since late 2013. But to this point, it has only been available for daily data, and it has been confined to a report in Reports & Analytics. With this release, any hourly, daily, weekly, or monthly time-series data in Analysis Workspace — whether in a table or a line graph — will automatically show anomalies based on your historical data and predicted trends.
Create a freeform table, using date as my dimension, and I immediately start receiving insights into statistically significant spikes or dips in my metrics. Add a line graph, and anomalies are called out as hollow data points on the graph. Hover over any of these points in the table or on the graph to gain more details about the anomaly.
As a bonus, if you are an Adobe Analytics Premium customer, you can right-click on these data points to run a Contribution Analysis, which scans hundreds of thousands of values of dimensions to find the likely causes of an anomaly — directly in Analysis Workspace, so you can embed the why along with the what in your projects.
In keeping with the idea of automated insights, there is nothing you need to do to turn this on or to activate this feature — just start building time-series data tables and visualizations, and away you go!
Averages (means) are good for quick snapshots, but as any analyst will tell you, they often hide tremendous insights behind their single-number façade. Histograms, which we’ve added as a visualization in Analysis Workspace, tease out those insights by showing the distribution of your audience across buckets (“bins”) representing escalating tiers within any metric. This makes it much easier to identify high- and low-value segments and market to them accordingly.
To use a histogram, just drag it from the collection of visualizations onto one of your panels. It will ask you for a metric, which you can also supply from the left. The histogram will bucket your visits or visitors according to how much/many of the selected metric they had. For example, if I use revenue as my metric, by default, it will bin my visitors by how much revenue they had. Using the advanced settings, I can adjust the number of bins, the size of each bin, and the starting bin (primarily, so I can choose whether to include visitors who had zero of the selected metric).
The distribution of revenue per visitor in my dataset is mostly normal; however, there is a bit of an uptick on the right side, so the group of customers who are spending $600–700 is larger than I might have expected. If I want to analyze that segment further, I can click the dot in the upper-left corner of the visualization and choose to show the data source. In the data table that is revealed, I can find this segment and begin to explore it directly in that data table. I can also drag it to the top of the panel to apply it as an ad-hoc segment, so I can add other data tables and visualizations that will help me better understand this group of customers who are behaving interestingly.
I’m sure you can see how we are advancing our goals to empower your organization with more recommendations powered by machine-learning intelligence and advanced analytics, but these features, and the benefits they provide, are just the start for us. We’re excited to continue our journey in helping you turn your company into a truly brilliant enterprise.