Under the hood of Adobe Sensei - how Adobe is evolving their analytics with AI

Leveraging artificial intelligence (AI) to improve marketing and customer experience analytics is a primary focus of Adobe, more than any other analytics vendor.

So I thought it was worth a look back over the past few years at the AI-focused capabilities added to the Analytics solution.

With the help of Adobe Sensei

Adobe Sensei is Adobe’s artificial intelligence and machine learning technology. It’s slowly getting embedded into all aspects of Adobe’s digital experience solutions, including Campaign, Target, and Analytics.

It’s fair to say that AI is useful in a wide variety of situations helping marketers and customer experience professionals make sense of volume of data coming from many channels. What it brings to analytics is the ability to not only help us understand what has happened in the past and why, but also give us insights into where marketers can improve future experiences.

Here's a quick look at several key capabilities which use AI as an underlying technology:

• Anomaly Detection: Adobe introduced anomaly detection to Analytics in 2013. Analyzing data, it could tell a marketer if something happened outside normal operations and expectations - good or bad. Did a campaign do better than expected? Did a video go viral or some news cause a stir? You can analyzes anomalies over time. For example, you can compare an anomaly like a surge in sales of a product one year, over what happened the prior year.
• Contribution Analysis: Building on top of anomaly detection, contribution analysis (added in 2015) gives you context into why something happened. So it will look at an anomaly and analyze the data, and give you key areas to focus on that likely cause the anomaly. Was it a web page? An email? A combination of items? You can also analyze contribution segments and further dig down into segments that interest you. The AI does the analysis; you dig into what is important.
• Intelligent Alerts: Intelligent alerts were added in 2016. They are a way to send you automatic alerts when an anomaly occurs, but you can also set them up based on other rules.

All three of these capabilities are available in the Workspace area of Adobe Analytics and fall under what Adobe Help calls the Virtual Analyst:

Your Adobe Analytics Virtual Analyst continuously analyzes data and uses predictive algorithms and machine learning to deliver personalized alerts and insights into unknown anomalies impacting your business.

This week, the Virtual Analyst was introduced as a new feature of Adobe Analytics, with Adobe saying it:

Builds on existing technologies to now automatically uncover insights without the user having to ask. By leveraging Adobe Sensei, Adobe’s AI and machine learning framework, it will empower brands to drive more value from analytics by surfacing signals that would have otherwise gone unnoticed. This can include valuable insights around sudden spikes or drops in key metrics such as online orders, traffic and reservations, along with identifying the contributing factors. This is critical to addressing escalations in real-time and finding new opportunities.

So, is it just a new way to talk about all the prior capabilities listed above? Or is there something new? It’s a little bit of both.

Virtual analyst

The Adobe Analytics Virtual Analyst is the marketing analyst’s super smart assistant who has no life. It’s always running in the background analyzing data and looking for things it finds interesting. With the help of Adobe Sensei, it analyzes all customer data points across all connected channels.

The idea behind the Virtual Analyst is that it helps you find things you didn’t know you needed to know to uncover the “unknown-unknowns”, meaning you aren’t asking questions, it’s telling you something is going on. It’s easy to miss things amongst all that data especially now that so much of it comes in real-time. How can an analyst continually look at data as it comes in and compare it to historical data? AI can do that, saving the marketing analyst time:

Another benefit to the virtual analyst is that it will de-duplicate insights, instead packaging up data that is related to the same event.

And because it’s always learning, the virtual analyst analyzes the behaviors of its users to personalize the insights, showing a user the things they tend to look at most. As other people in the company use the virtual analyst, it will also analyze their usage patterns and surface insights that may be beneficial to the current user. The point here is that the analyst is always learning and improves the insights and recommendations it surfaces as it’s used.

Unlike anomaly detection and contribution analysis, the Virtual Analyst sits on its own and runs in the background automatically. As a user looks that things the virtual analyst points out, they will shift into the Workspace to do more in depth analysis doing things like contribution analysis.

Up next, Adobe is working on something called Prescriptions which will offer recommended actions to take based on the analysis it performs on the data and the insights it surfaces. According to Adobe, “Our vision moving forward is that the AI will not only surface signals that you didn’t ask for (but need to see), but will also prescribe for you the specific actions to take. We are well on the path to realize this in the future, and have lots of research happening now. No one in the market is anywhere close to solving for this. It is a big challenge.”

My take

Adobe isn’t the only analytics vendor to provide anomaly detection and surface insights. Google Analytics also does this, automatically scanning data for anomalies and providing alerts to let you know if it found something. It also allows you to ask questions in plain English without having to create and run a report.

A lot is happening in the analytics market thanks to the advances in machine learning and deep learning. The next thing I’d like to see is if the AI technology can not only surface insights but adjust parts of a campaign or web page based on those insights. These changes might start as minor adjustments, but think of a future where you can teach the system to make critical fixes automatically, or take advantage of positive anomalies by following a decision-tree or rules (and then eventually learning what to do automatically). After all, we can’t always be on, but our virtual assistants can.