7 Agentic AI Lessons We've Learned

The world is awash in agents, and every enterprise customer experience vendor is now agentic. At Constellation's AI Forum DC in September and AI Forum Silicon Valley in March, more than 150 artificial intelligence practitioners shared many best practices and lessons learned in agentic AI deployment. These CX professionals showed how they were able to achieve a real return on transformation investment (RTI) for sales, marketing, service, commerce, and advertising.

Here are the seven lessons that matter:

1. Manage the mind-shift in culture.

Leaders must make clear the intent and purpose of agentic AI deployment. Agentic AI used for cost reduction operates differently than agentic AI used for revenue and growth. Constellation’s RTI looks at cost, benefit, probability, and project type to classify projects by these metrics:

• Regulatory compliance.
• Cost reduction.
• Operational efficiency.
• Maintenance and incremental improvement.
• Revenue and growth.
• Business model transformation.
• Brand transformation.

Expectations on outcomes should be clearly stated up front. Enable your teams to make decisions and empower them to succeed. By setting the right cultural tones, agentic AI projects will create a win-win.

2. Invest in a defensible data strategy.

A good data strategy serves as the foundation for successful agentic AI deployments. It's about more than just data governance and quality; the companies that have invested in a coherent and clear data strategy can build the foundation to support personalization, right-time offers, infinite ambient orchestration, and more. Data and distribution are the clearest business moat. Organizations that underinvest will feel the pain in every deployment and consequently accumulate a sizable technical debt.

3. Reinvent your business processes for decision velocity.

Collapse decision trees and replace them with agents. Automation is assumed, but when and where you insert a human becomes the key linchpin of success. Redesign business processes according to these steps:

• Trust intelligent machine automation via agents.
• Augment the machine (agents) with humans.
• Augment humans with a machine (agents).
• Trust human judgment (with humans and agents now in a symbiotic trust relationship) fully.

CX leaders have a once-in-a-generation opportunity to remake the customer journey.

4. Always automate the deterministic.

Start with an inventory of deterministic processes. These are well defined and often have a clear regulatory foundation. Binary as yes-or-no decisions, deterministic decision trees can rapidly be collapsed and automated for the biggest short-term wins. The biggest gains come from automating deterministic processes with agents so that probabilistic processes can be augmented by human or machine.

5. Improve precision in the probabilistic.

Probabilistic processes are the most problematic with agentic AI. Leaders must determine a comfort level for precision. At 85 percent accuracy for customer service, most leaders could be happy. At 85 percent accuracy for order completion, millions of dollars will be lost. At 85 percent accuracy for financial transactions, jail time is on the table. For every probabilistic process, the level of precision determines when a human is inserted. People are inserted when confidence in precision is low, regulatory requirements mandate a human decision, or a human touch is called for. Understanding when and where to insert a human is the secret sauce for probabilistic processes.

6. Prepare for cross-platform agentic AI.

Agents have different levels of granularity in process, and often teams force-fit agents to a department. Successful deployments break through functional fiefdoms to cut across departments and support end-to-end processes. The romantic notion of one agent per role makes no sense when agentic AI can scale massively across roles, departments, and business processes. The rise of cross-platform agents requires a holistic view. Successful leaders go end to end as far as they can.

7. Index for wisdom before knowledge.

In the age of AI, expertise is a commodity, but experiences are not. The ability to connect the dots across industries, markets, regions, and offerings comes from decades of pattern recognition and experience. Most large language models have learned from success while most successful humans have learned from failures. This lens allows for the acceleration of human judgment. The level of people's expertise can be judged by the questions they ask, not the answers given. Wisdom trumps knowledge, and staffing levels should reflect this.

The Bottom Line: Start Now

Time is of the essence. CX leaders who map out processes and journeys can start on their agentic AI paths ahead of the technology. Applying these seven best practices, organizations can fast-track their selection of a technology vendor and services partner. 

R “Ray” Wang is author of Everybody Wants to Rule the World: Surviving and Thriving in a World of Digital Giants (HarperCollins Leadership) and founder of Constellation Research.

Related Articles

Boosting CX Worker Productivity in the AI Age

Agentic AI combined with customer-facing human agents will transform service.

AI ‘Exponentials’ Will Rule the World

Highly effective tiny teams will be nimbler than the legacy behemoths.

Exponential Efficiency in the Age of AI

Harness the power of AI in every aspect of the enterprise.