Our recent webinar, From Hype to Action: How to Build with Agentic AI (The Right Way), brought together two practitioners who are applying agentic systems in live customer environments every day.
Deon Nicholas, Co-Founder, President, and Executive Chairman of Forethought, and Phil Lynch, Senior AI Program Manager at ActiveCampaign, spoke candidly about the difference between generative tools that merely respond and agentic systems that understand, reason, and take action.
Their discussion offered a clear view of how customer experience teams should evaluate emerging technologies and where the real operational value is beginning to show up.
What Agentic AI Actually Represents
Nicholas opened by noting that although “agentic AI has become a buzzword in the industry in so many ways,” the important question is whether a system can truly interpret customer intent, reason over business rules, and carry out actions inside core systems. He explained that organizations quickly discover a ceiling with generative tools that only search, summarize, or restate knowledge. In his words, “over 70 percent of issues require some form of agentic AI or agentic action” while only “about 30 percent… require say FAQ or knowledge-based AI.”
Lynch agreed, describing how teams rapidly expand their ambitions as soon as they understand what agentic systems can accomplish. As he put it, “once someone kind of gets the… whole of like AI, like immediately they’re just, mind kind of goes a bunch of different areas of like ideas of things they can try and do.”
Moving Beyond Traditional Chatbots
Lynch described ActiveCampaign’s earlier chatbot efforts as heavy and inflexible. His first experience required “creating 500 different decision trees” and constant upkeep. A later version “was a glorified search” that “just spits out the article… and then you have to click into it and find the answer.”
The experience shifted when they adopted agentic capabilities. The new system “will follow the tone of voice… summarize and highlight the key points… and deliver a very personalized experience.” Customers no longer receive general links. The agent draws on account details, past interactions, and internal systems to produce answers aligned with each user’s situation.
Training Without Technical Barriers
A central theme throughout the webinar was accessibility. Lynch noted that Forethought’s autoflow builder is “just a prompt-based system. like you go in pretty much, if you can write it, it can do it essentially itself.”
Instead of coding decision trees or intricate rule sets, teams describe processes in natural language and allow the system to operationalize them.
Nicholas emphasized that this distinction is core to agentic AI. He pointed out that leaders should consider whether a solution forces them to write rules and code or whether it “can learn through natural language.”
Using Organizational Data to Drive Real Actions
ActiveCampaign’s deployment highlighted how agentic systems leverage data from multiple internal sources. Lynch explained that they connect to Tatango, Snowflake, Zendesk, and other systems, with some workflows “carry[ing] through about seven of these… unique distinct agents.” Each agent performs a specific analytical or operational step, ultimately producing a unified recommendation or action.
One example addressed customers who request updates on open cases. When this occurs, the AI “can grab that ticket… and then… post a message to that ticket bumping that with the agent that has it internally.” It functions as an operational partner, ensuring nothing stalls and customers remain informed.
Expansion Across Channels and Departments
Although chat remains the heaviest channel, ActiveCampaign has extended agentic capabilities into email and is exploring voice. Lynch explained that email can now support full, multi-step exchanges, with the AI collecting details and progressing the case.
Internally, adoption is equally strong. The team now uses several agents inside Slack, and Lynch shared that they recently built “a competitive Intel kind of co-pilot” to support sales teams with insights during calls.
Governance, Trust, and Auditability
Both speakers stressed the need for visibility into how agentic systems behave. Lynch explained that Forethought provides clear audit logs, workflow-level details, and error logs that indicate precisely where an adjustment is needed. If something appears unusual, the team can trace the reasoning and the source material immediately.
He added that hallucinations have not been a meaningful issue. Because the system identifies the sources of its answers, the team can quickly refine an article or instruction whenever something needs clarification.
What Results Look Like in Practice
ActiveCampaign saw immediate impact upon implementation. Lynch noted that they observed “a pretty large amount of deflection… up near 50 percent by just turning on the system itself.”
Longer term, AI adoption reshaped internal operations. Lynch described the growth of a dedicated AI department and widespread reliance on automated agents for daily work, with the technology now integrated into workflows across the company.
Key Considerations for CX Leaders
Nicholas summarized the framework CX leaders should apply when evaluating agentic systems:
- Determine whether the system is genuinely agentic rather than rule-based.
- Assess how easily it can be trained and adapted through natural language.
- Confirm that it learns from ongoing and historical conversations.
- Ensure that it behaves consistently across chat, email, voice, and other channels.
- Require transparency into its reasoning and auditability across every action it takes.
These considerations shape which solutions scale effectively, and which produce measurable improvements in customer experience and operational performance.
Watch the full webinar here: https://aileaderscouncil.org/events/from-hype-to-action-how-to-build-with-agentic-ai-the-right-way/
Download the 2025 AI in CX Benchmark Report
For a deeper view of the trends and metrics discussed during the webinar, download the 2025 AI in CX Benchmark Report.
