As Adobe Summit arrives again, it’s time to share some exciting news about a feature we are announcing in Adobe Analytics. To explain it properly, I have to harken back to when we released Anomaly Detection over a year ago. Anomaly Detection uncovers significant signals, or changes, in your trended data from among the noise. It literally finds the needle in your haystack of data. Once an anomaly has been identified, the next logical question is why. Why did revenue drop by 30 percent? Why did we experience an anomalous drop in video completes yesterday?
Anomalies – A Data Analysis Nightmare
Your boss stops by and asks, “Everything on track?” Tricky question. You may need to crunch millions or even tens of millions of records – site visits, bounce rates, unique visitors, conversion rates, revenue and dozens or hundreds of other metrics. Then, you must figure out if all these data fit their respective patterns. If any drifts outside a 95 percent confidence level margin, where typical noise fluctuations live, you have an anomaly on your hands.
Even if you somehow sift through and analyze enough data, and find some anomalies, you can be sure they’ll land right back on your desk. This time, the boss wants to know, “What’s causing it?” You’re in the hot seat and need to have an answer, and quick.
Data Scientists – Mismatch Between Supply and Demand
Data scientists are hackers by nature. They can throw together a query, code it, run statistical analyses and tests, bring in machine learning, figure out business problems, and even conjure up data visualizations to boil the complex results of their work down to an easy-to-grasp image or two. When a good data scientist finishes her work, not only has she made the problem seem simple, the path forward is clear as day.
Unfortunately, juxtaposing all these talents makes data scientists slightly harder to find than Bigfoot; and as basic economics teaches us, when something is in great demand and limited supply, its cost skyrockets. That’s why your boss is still looking to you for that answer.
Contribution Analysis – Adobe’s Data-Scientist-in-a-Box
Since finding a good data scientist is difficult and expensive, your boss is expecting you, a non-quant analyst, to do the job. Adobe has just released a great tool to bridge the gap – Contribution Analysis.
To help you get the job done without drowning in the details and mechanics of models, regressions and other statistical esoterica, Contribution Analysis does the heavy lifting in the background, and lets you concentrate on getting business answers for your boss and your company. This doesn’t mean you’ll never need a data scientist again, but it does mean you can do many of his tasks on your own.
So how does it work? As the following illustration shows, Contribution Analysis is found by navigating to the Anomaly Detection Report within Reports & Analytics. After opening the trended visualizations containing anomalies, you simply identify an interesting anomaly in a metric within the trended visualizations. Now you want to know what caused that anomaly. Enter Contribution Analysis. You select your anomaly in the trended visualization and then click analyze.
Contribution Analysis automatically kicks off the analysis – querying and statistically analyzing every conversion, traffic (including the pathing versions of the props), out-of-the-box variable, SAINT classification, customer attributes as well as all data collected within Adobe Analytics for mobile, video, Adobe Social and Adobe Target solutions. It applies advanced statistics and machine learning to learn the patterns inherent in the data to find potential causes or contributors within seconds to minutes (perhaps long enough to get a cup of coffee), rather than weeks or months of laborious data scientist effort. The system displays a rich visual contribution analysis that tells the story, easily identifying all relevant factors.
It once took me literally six weeks of intensive statistical analysis to tease out a $400 million opportunity for a client. Now, this true Big Data learning engine transforms you, a non-quant analyst, into a lightning-fast data scientist. With its help, you can generate answers for your boss and still be home in time to take your significant other out on that date.
Instead of flailing though hundreds of possible reports that may or may not be relevant, Contribution Analysis shows you, as the next illustration shows, the top factors contributing to the anomaly. As you move further down in the interactive experience, there are new big data visualizations like the hierarchical tree map and scatterplot visualizations that show all of the contributing factors.
The system includes many other interactive features and actions for further exploring the contributors associated with your anomaly. Finally, you may be interested in understanding which contributing items occur together for visitors or during visits on the day of the anomaly. Contribution Analysis has a “Top Segments” feature and algorithm that automatically uncovers contributing segments for you. These are not your segments saved in Adobe Analytics, but segments that Contribution Analysis has discovered. You can then select a segment and click “Create Segment” to have it saved within Adobe Analytics and make it available across Adobe Marketing Cloud.
Top Five Benefits of Adobe’s Contribution Analysis
This “data-scientist-in-a-box” offers several important benefits.
- Analyzes immense amount of data in seconds, rather than consuming weeks of data scientist time
- Provides automated advanced analytics usable by non-quant analysts
- Offers a rich, intuitive, interactive user experience tied to segment creation
- Analyzes customer attributes including offline ones imported from your CRM solution
- Provides the most up-to-date, comprehensive, and trustworthy analysis on the market
With these powerful features in your hands, you can accomplish in seconds powerful tasks that would otherwise require a data scientist at your beck and call.
The capability will be added to Adobe Analytics Premium in the next few weeks.
Are you periodically asked to serve as a stand-in for a data scientist, and if so, are you happy with your current tools, or would Contribution Analysis make your life easier?