Ten Considerations for Applying Agentic AI
I had a very enjoyable and thought-provoking conversation with Michael van der Steen yesterday.
Michael is a senior executive in Global Business Services, Digital Transformation and Data, having led change and operational excellence initiatives at Adidas, BMI Group and AkzoNobel.
It seems only yesterday that Generative AI was leading the transformative technology “Premier League”.
In 30 short months GenAI has ceded poll position to Agentic AI in the minds of futurists, management consultants, technology marketeers, and the C-Suite.
The promise of autonomous business processes with minimal human intervention is highly appealing and, from all the forecasts, could prove to be either an economic miracle, a societal disaster or another over-hyped, costly technology strategy.
Michael and I had a fascinating discussion, exploring the potential, the use-cases, opportunities, challenges and recommendations in applying Agentic AI.
The reference to “Agency” is at the root of this innovation, although still in its early days.
“Agency” is the capacity to make individual choices and take actions that influence our lives and surroundings. It’s a fundamental aspect of what it means to be human, involving the ability to set goals, make plans, and pursue them.
This is the aspiration of Agentic AI.
- A class of system that can autonomously make decisions and take actions, often without direct human intervention.
- To understand context, set goals, reason through steps, solve problems and adapt actions based on changing conditions, all with “limited” human oversight.
- Multiple “agents” working together to achieve a desired business outcome.
- Not all the “agents” are necessarily Generative AI, or even Analytical AI, and will likely also include classic, deterministic code agents or “bots”.
- In one sense, an evolution of enterprise workflow, “connecting the dots” in an “Action Flow” requiring less human input than today, with higher quality outcomes.
Of course, the human “augmentation vs replacement” discussion is central here.
The Automation “Toolbox” has not simply reduced to one option now, rather we have an additional power tool to add to those we already use. The key is to recognise and understand the classes of “business task” to which we wish to apply automation, to establish best solution fit. AI, Agentic or otherwise, is not a “silver bullet” for all cases.
Michael and I explored the 3 most commonly discussed scenarios that may occur over the coming decade as a result of Agentic AI. The outcomes for business, the economy and society.
We also touched on example use-cases, such as;
- Marketing campaign planning
- Legal research and document creation
- Customer credit management
- Complex customer order configuration
- Collections / Accounts receivable
- Fraud detection & prevention
- Vendor management
- Supplier sourcing, procurement to payment approval
- T&E expense reporting and reconciliation
- Logistics optimization
- IT service management and issue resolution
We had a great discussion, off camera, about the potential use of Agentic AI for master data governance and continuous cleansing, based on triggers that could initiate an Agentic AI “Action Flow”, touching customers and suppliers directly as the “single source of truth”. Master data is the major, often hidden, problem that is the root cause of many downstream quality and productivity issues. This one, among others, is a potential big win.
Finally, we talked through the “Ten Considerations on the Road to Agentic AI” . .
- The “end to end” Business Process beyond the Agentic flow (Revenue, Product, Spend cycle etc ) – if we don’t understand it, we are not automating, we are just hoping.
- The “topology” of Agentic Action Flow – understand the classes of target business task and relevance for AI and probabilistic outcomes, and for classic deterministic, binary outcomes – “Horses for Courses”.
- “Think like a CFO” – Where is the P&L or Working Capital impact or the Customer impact that creates them? We should be solving genuine business problems.
- Consider Human Behaviour, Habits & WIIFM – There is always a “Human in the Loop”, whether or not we have designed one in.
- The Digital Dichotomy – not all information or knowledge is digitized!
- Explore where Human Relationships add value to the task or brand, and what the implications are.
- Accountability – whose job is it when things go wrong? They will. 😉
- Data Quality & Integrity – Critical to any automation, especially AI.
- Augmentation vs Replacement – What outcome do we want?
- The Future of Talent – What skills should we be hiring for the future?
You can watch or listen to our 30 minute conversation here . . .
Thank to Nic Halley for sharing that.
Thanks for reading . . .
Ten Considerations for Applying Agentic AI