6 Levels of Automated Analytics & Action on Data with AI

What should we expect AI to deliver for automating analytics and action on data? In the era of self-driving cars and Gen AI, our expectations run very high for AI Analytics.

The Automotive industry’s 6 levels of autonomous driving can provide inspiration

Years ago, in a similar situation, the automotive industry defined 6 levels of driving automation. Specifically, these six levels are documented as a trademarked standard by the SAE, i.e. “6 Levels of Driving Automation”. Most automotive brands today offer cars capable of Level 2, and so far there is apparently only one brand that has been certified for Level 3 in some limited geographic regions is , i.e. Mercedes Benz.

What if we defined 6 levels of Automated Analysis & Action on Data?

With full credit to the SAE for the template, here is what 6 levels of automation could look like for data analytics and action on data.

The last row of the graphic above provides an example for one out of a gazillion use cases of data analytics. Namely, something that everybody can relate to: Using data to resolve website and mobile app user experience frustrations.

Here are two more examples to show how the 6 levels play out for other use cases.

What use cases can you think of?

Setting expectations high for automated analysis and action

Just like the hierarchy for autonomous driving, this proposal also sets the bar very high. For example, a car driver assistance technology such as “lane centering” feels pretty darn astonishing to me. But within the levels for autonmous driving by the SAE, lane centering ranks very low, just at Level 1.

That’s because Level 5 are vehicles that can operate autonomously “everywhere in all conditions”

In the same spirit, we need to set our vision high for analytics too.

First, of all, let’s focus on synthesis and insights, not analytics

Data and analytics are tools or means to an end. First, we want synthesized insights, so that data is turned into something that is actionable and prioritized.

Secondly focus on automated action, not just automated insights

Focusing only on insights would be like focusing on cars that provide warnings and recommendations to drivers, e.g. “You should break hard now”. Clearly, our expectations are higher than that. Same with Analytics. Our hierarchy goes beyond “Co-Pilots” to automating action.

But is it too utopian?

Do the examples proposed for automated action in my proposal above sound utopian? They sure do. But so do autonomous vehicles of Level 5, i.e. “everywhere, in all conditions”

Critical success criteria

The automotive industry has invested billions in self-driving cars and has probably trained them over millions of miles in test vehicles. And still, we’re barely at the cross-over into Level 3.

What do we need for analytics automation, besides billions of dollars of investment and rocket scientist engineers?

Here are some critical factors:

• Sensor inputs that are more complete: Could a car drive autonomously without the addition of cameras as sensors? No! Same is true with self-driving analysis & actions, i.e. we’ll need more than the traditional data inputs. For example, for self-optimizing websites and apps, we’ll need more than just traditional web analytics data at the level of clicks. We also need data on the 90% of behavior and experiences that happen in between clicks, e.g. scrolling, exposure, engagement, hesitation, frustration, errors, slow speeds, etc.

• Integrating data with action:  A car lane centering assistant could not do its job if it wasn’t able to control your steering wheel. Same with assistance for automating A/B testing, personalization, and marketing automation. Need to integrate data, AI, and action tools.
    
• Business outcomes focus:  The SAE didn’t call it “6 levels of self-driving cars”, but “6 levels of driving automation”. So it’s not “6 levels of AI analytics tools”, but “6 levels of autonomous analysis & action”.
      
• End-to-end use case coverage:  Just like autonomous driving requires more than just “autonomous braking”,  same here. We need more than just automated finding and fixing of issues. We need automated finding and reaping of opportunities too
    
• Real time: Not just post-mortem analysis and course correction. By that time the car or business have already crashed! We need to help customers while they are in the middle of the interaction with us.  
  
• Informative, explainable AI :  How many people can drive a car? Billions! How many can repair a car? Very few.  Same here, we don’t need the humans to understand how exactly the AI does what it does, but the output needs to be easy to explain. 
  
• Accountability for Actions — with cars and analytics alike, accountability remains with humans for the actions of the AI, even if they are on vacation while the AI is driving their web business.

The first letters of the criteria above spell out S, I, B, E, R, I, A. Easy to remember!

The SAE defined it for Automotive, who is going to define it for Analytics?

I love that the 6 levels of driving are a formal standard, defined by an industry body. That way, we’ve got a common language for classifying any car manufcaturer’s self-driving cars into L0 – L5.

We need the same for the 6 levels of Automated Analysis & Action on data. For example for digital analytics, this would have been perfect for the Digital Analytics Association. We need a similar industry body to take this over and align analytics providers on a standard. Any suggestions?

Next step: Request for comments (RFC)

While we think of an industry body that could take this over, all comments and critique are welcome so we can make this better. I’d also love to see more examples for use cases beyond the three discussed above.