NBS Knowledge Lab Webinar: Agentic AI in Marketing: From Buzzword to Blueprint

Key Takeaways for Marketing Leaders

• Agentic AI shifts the focus from single-prompt outputs to goal-driven workflows, where multiple AI agents coordinate across tasks, tools and feedback loops.
• Successful adoption starts with a clear business problem, not the technology itself. Not every workflow requires an agentic solution.
• Human oversight remains critical, particularly in strategy, creative judgement, governance, evaluation and brand risk management.
• Organisations must invest in the right foundations, including data pipelines, workflow design, upskilling, guardrails and system access, to unlock meaningful value from agentic AI.

From Chatbots to Goal-Driven Workflows

Opening the session, Ramanathan Vythilingam set out the distinction between conventional AI chatbots and agentic AI workflows. While chatbots are typically rule-based and output-focused, agentic AI is better understood as a learning-based and outcome-focused systems. Instead of asking AI to complete a single task, users define a goal, which is then pursued through a coordinated ecosystem of AI tools and agents.

This distinction is important for marketers. In a standard chatbot interaction, humans remain responsible for deciding the next step after receiving an output. In an agentic workflow, however, the system can perceive information, reason through possible actions, execute tasks within predefined guardrails, and learn from feedback over time. As Ramanathan illustrated, a simple workflow could involve an AI system reviewing a user’s calendar and email each morning, identifying priority meetings, drafting preparatory notes, suggesting replies and improving its recommendations based on how the user responds.

Krishnan Menon extended this comparison through the example of customer service. Traditional chatbots are often rules-based, drawing from fixed responses to answer what the system thinks the customer has asked. The leap towards agentic AI, he argued, lies in “knowledge and learning”: the ability to understand the customer’s context, relationship with the brand, and the problem to be solved. Rather than simply directing a query along a decision tree, an agentic system can work towards a solution by asking better questions and drawing on richer organisational knowledge.

Building Blocks of Agentic Marketing Systems

Emir Caglayan provided a practical view of how agentic systems are designed in marketing contexts. He explained that an agent is not a single technology layer, but a combination of several components: the base model and its settings, the knowledge base it can draw from, the persona and system instructions assigned to it, the tools and capabilities it can access, the guardrails that define its operating boundaries, and the task logic or execution sequence that determines how work moves through the system.

These building blocks become especially powerful when multiple agents are connected within a workflow. Drawing from an example of an in-the-moment reactive social campaign, Emir described how agents could be designed to monitor social conversations, identify cultural moments, incorporate relevant historical context through a “sports archivist” agent, generate potential marketing ideas, which are tested with synthetic focus groups, and winning ideas feed into a creative ideation agent trained on brand tone and campaign requirements. Instead of relying solely on a large group of people in a room to produce one or two posts, the multi-agent workflow enabled more ideas to be generated within a shorter cycle, with the full process completed in about 15 to 20 minutes.

The example underscored that agentic AI is not just about speed. It is about designing systems that can bring together different forms of intelligence, cultural listening, audience simulation, historical knowledge, brand guidelines and creative generation, in a structured and repeatable way.

Applying AI Across the Customer Lifecycle

David Low shared how Singtel is approaching AI within lifecycle marketing and customer lifecycle management. In the telco industry, churn is a major commercial lever, but predicting churn is only part of the challenge. The harder task is acting on churn signals in time, at scale, and with offers that are relevant enough to influence customer behaviour.

David described Singtel’s PRISM system as a hybrid, layered architecture rather than a purely agentic platform. It combines traditional machine learning and deep learning with generative and agentic components. Data science agents extract and ingest hundreds of behavioural features from telco data, including customer interactions, usage patterns and historical behaviour. Marketing agents then reason over these signals, incorporating customer personas, behavioural cues and past campaign interactions to produce customer segments. Creative agents generate copy, visuals and templates for customer communications.

This layered approach allows predictive components to answer who should be targeted and when, while generative components help determine how those customers should be engaged. Early pilots showed positive results, including a close to six percentage-point improvement in identifying at-risk customers compared with traditional churn models alone. At the same time, David stressed that agentic and generative layers are only the “last mile”; the underlying data pipelines, feature engineering and machine learning operations remain essential foundations.

Choosing the Right Tool for the Right Problem

A recurring theme throughout the discussion was that not every marketing problem requires an agentic AI solution. The speakers cautioned against adopting agentic systems simply because they are the latest technological trend.

Krishnan emphasised that organisations must first have a clear grip on the challenge they are trying to solve. Agentic AI should not be implemented merely to reduce cost. Instead, teams should ask whether it can improve effectiveness, strengthen decision-making or unlock better outcomes. Only after defining the problem should organisations decide what role, if any, generative or agentic AI should play.

Emir echoed this point, noting that teams cannot automate or improve a workflow they do not understand. A practical starting point is to identify a repetitive or mundane use case within one’s area of expertise, map out the steps clearly, and then assess where intelligence is genuinely needed. Some parts of a workflow may be deterministic and better handled through rules-based automation. Others may require reasoning, judgement or a human-in-the-loop review.

David added that businesses should measure success by outcomes rather than by how “agent-shaped” a solution appears. In some cases, a deterministic pipeline may deliver the same result at lower cost and with less complexity. The right metric, he argued, is business return on investment, whether through increased revenue, reduced cost, improved customer engagement or stronger operational efficiency.

The Human Role in an Agentic Future

The panel also addressed a common concern: whether agentic AI will remove humans from marketing workflows. The speakers were clear that human judgement remains central.

For David, the human-in-the-loop function is especially important in creative gatekeeping. At Singtel, marketing communications experts review AI-generated copy, sign off on visuals and decide which templates the system is allowed to use. A poor targeting decision may affect a small number of customers, but a flawed piece of copy sent to millions can create significant reputational risk.

Krishnan argued that AI should push marketers and agencies further upstream. If marketing is reduced to producing images and copy, then generative AI can replicate much of that downstream output. But true marketing value lies in strategic judgement: understanding the market, the business, the product, the consumer and the role a brand plays in people’s lives. AI can support research and execution, but the ability to distil insight into a powerful strategic or creative idea remains a distinctly human contribution.

Emir added that humans are still needed to decide what agents to build, what priorities they should serve, how they should be evaluated and how they should be maintained. Agentic systems require ongoing testing, updating and refinement as models, tools and business needs evolve. Evaluation frameworks, including test questions, scoring criteria and output benchmarks, are essential before organisations can rely on these systems at scale.

Managing Adoption, Upskilling and Organisational Readiness

The speakers also highlighted several blockers that could limit successful adoption. Emir identified workforce upskilling as a critical issue, noting that while the technology is already powerful, it remains underutilised. The conversation has evolved from prompt engineering to context engineering and now to agentic workflows, but organisations still need to build the skills to design, implement and manage these systems effectively.

Krishnan cautioned that organisations are made up of people with different roles, knowledge levels and comfort with technology. Rushing into agentic workflows may complicate rather than improve business efficiency, especially if existing workflows are replaced without regard for how people actually work. Adoption should therefore be paced, iterative and sensitive to organisational maturity. He also raised the risk of cognitive offloading: as AI takes over more tasks, marketers must be deliberate about preserving strategic, creative and analytical capabilities.

Conclusion

The webinar underscored that agentic AI is not simply a more advanced chatbot, nor is it a universal solution for every marketing challenge. Its value lies in carefully designed workflows that combine data, tools, reasoning, guardrails and feedback loops to pursue clearly defined business goals.

For marketers, the opportunity is significant: agentic AI can support customer lifecycle management, reactive content creation, research, campaign deployment and personalised engagement at scale. But success depends on more than technology. Organisations need strong data foundations, clear workflow design, human oversight, evaluation frameworks and a realistic understanding of where AI adds value.

Ultimately, the shift from buzzword to blueprint requires marketers to ask better questions before building faster systems. Agentic AI can help teams move with greater speed and precision, but its most meaningful impact will come when it is anchored in strategic clarity, responsible implementation and human judgement.