Defining Systems of Outcomes part II: AI Augmentation vs AI Automation vs “AI Outcome-ation”?

I covered the concept of Systems of Outcomes in this introductory post that led to a lot of 1:1 discussion. Thanks to many of you for that.

The essence of that post was that Agentic AI is a piece of a broader rewiring of enterprise software that evolves the Systems of Record construct (recording stuff) to Systems of Outcomes (getting things done).

Here is part two.

To start to make sense of Systems of Outcomes, we need to look back before we look ahead. Right behind Agentic AI was Automation and Augmentation.

1. Automation is predicated on standardization. RPA and other workflow-based tools offered automation for very repeatable processes. From enterprise workflows to the automated coffee pop-up stands you see in many cities that can make a cappuccino. Love it or hate it, it's going to taste and deliver the same every single time because the needed inputs are constant. Standardized steps and formulas let the machines take charge. But inherent in its success is that it’s standardized process, working in very slim swim lane. To be fair, there are plenty of use cases for this.
2. Then there is augmentation. This gave birth to co-pilots. They are your wing-men and woman, collaborating with you to get stuff you done. Like an AI assistant always making your life easier. Kyle Wiggers over at Techcrunch, explains it well. . Co-pilots are good because they blend automating and augmentation which adds the needed subjectivity for certain tasks that humans can provide. And they are getting better and better.
3. But we are here to discuss Outcome-ation, which of course is not a real word. Outcome-ation, powered by emerging Systems of Outcome requires intelligence to be able to take multiple routes to execute an outcome. They obsess about getting to the outcome and are designed to look at many variables. Much like a human would do and gets paid a lot to do. Those different routes mean checking on a selection of different data sets, be that an ERP system or unstructured text in an email, to ensure a proper outcome. A simplified example would be as follows:

Let's look at executing a sales order for 10,000 tons of Florida Oranges (our thoughts and prayers are with you today, Florida). Based on the season or time of year, the AI agent or hoards of them would ensure that trucking routes and delivery times account for potential hurricane paths, or based on the month, ensure that inventory is adequately unripe so that it doesn't spoil on a trip all the way to Seattle. And be able to run a bid and change suppliers on a dime to optimize margins one last time before shipment based on the most current rates. And be able to accommodate instructions that came in in an email 30 min before shipping about a customer's loading dock change. And be able to change shipping quantities based on dynamic demand. And on and on. You get my drift. That's the work of Systems of Outcomes. Not Systems of Record. Or Augmentation. Or standardized Automation.

As you can see, Systems of Outcomes very often will have to traverse data across the demand and supply chain to execute an outcome. Something that has been far far too difficult for humans to do for systematic, org design, or internal political reasons. I have some hope that machines going straight to the data might be more feasible than getting us humans to get past the political posturing of turf hoarding or the feasibility of continuous and expensive systems-level engineering.

Outcome-ation, or the Systems of Outcomes that will drive it, are not designed to define the completion of a set of rigid steps. They are designed to obsessively optimize for a particular set of outcomes, looking across structured and unstructured data. Over time they learn and get more experienced and look at more and more variables that can improve the outcome.

Systems of Outcomes that drive outcom-ation, just like Systems of Record, will be highly functionalized or verticalized so that they get continuously smarter about a narrow set of problems. It's no surprise that a huge number of natively AI-driven application start-ups are being funded right now to solve a host of vertical problems.

It's a future that is far from here, but the plumbing for this future is being laid out right now with much simpler constructs to get out the gate. And messaging and positioning this coming radical shift to customers is going to be incredibly daunting but also exciting.

(Special hat tip to my stupendously smart pal and ex-colleague, Navin Budhiraja , who pushed me to articulate the difference between the promise of the first generation of RPA and what I called Systems of Outcomes in my opening salvo on this topic.)