Definitions of agentic AI
Both roundtables kicked off with participants defining what agentic AI means both to them as data leaders and how it may be understood – or misunderstood – by stakeholders in their organisations.
The agreed description of agentic AI is the use of LLMs to execute actions like making API calls, pulling data, or updating databases.
The agreed goal of agentic AI is to drive intelligent automation and decision-making at scale. Each organisation has specific problems that can be solved with agentic AI, and the important first step is being able to describe through storytelling how the issue can be addressed with an agentic AI solution.
One participant running a proof of concept for agentic AI in their retail business likened the concept to robotic process automation (RPA) but with more intelligence and the ability to execute tasks that are exclusively unique to each business. Another participant had explained it to non-data professionals as “RPA on steroids.”
Starting the agentic AI journey
Participants emphasised the importance of total clarity on the objectives on any agentic AI deployment as well as a holistic understanding of its implications on connected processes and the wider business.
For example, a data leader from the insurance industry explained that they are looking to deploy agentic AI to help them manage temporary call-centre spikes during disaster events, which today cannot be effectively or economically remedied with additional staff. Agentic AI offers the possibility to prioritise vulnerable customers most in need of immediate human intervention while organising less critical calls into a manageable task list.
Furthermore, agentic AI can help identify at-risk geographies prior to disaster events happening, meaning the insurance organisation can prepare for likely scenarios before they happen, such as being able to focus attention onto a specific insurance claim type or geography while maintaining quality for insurance claims that are happening outside of the effected region.
However, it was acknowledged that automation introduces the risk of degrading the skills and expertise of insurance handlers who traditionally gain experience through resolving low-level claims, developing skills and knowledge required to handle today’s complex claims. Organisations, therefore, need to understand the skills and learning journeys of experts, balancing this with the excitement around the potential for new automated solutions. The importance of defining clear business problems and the evolving role of data teams in AI-first strategies were emphasised throughout and must be regarded as a starting point for an agentic AI journey.
Agentic AI examples from the community
Participants shared their current use cases with agentic AI, including:
SEO
One member was using agentic AI to navigate Google’s SEO landscape following multiple changes to the way ranking algorithms worked which impacted click-through advertisements. Google’s ranking algorithm has changed to use LLMs, leading to a need for first-party information to offset the impact of generated content by the member, and agentic AI was being trialled to solve this issue.
Content transcription and optimisation
Another participant from an event’s organiser described how their team was trialling agentic AI to interpret recorded sessions from events which would then adjust articles based on SEO backends and direct registrations.
Call centre experience
A similar concept was being used by a member that has call centres. Agentic AI is being used extensively in their call centres to interpret unstructured conversations and store information in CRMs or operational systems. This enables the team to prioritise at-risk customers, monitor sentiment, train new hires, and much more.
One participant echoed the use by the call centres in that they were using agentic AI to identify customer complaints and make proactive offers to ensure higher quality satisfaction and experiences.
The examples demonstrate that agentic AI is primarily being trialled to improve customer experiences and wait times. These are also seen as simple ways to introduce agentic AI based on the existing LLMs organisations have been running and it would not require much change to existing architecture.
Concerns with Agentic AI
Al participants noted that there are risks with agentic AI, including issues within highly regulated industries, architectural changes to existing operations, and talent required to make it work, underscoring the explicit need to identify and analyse the problems that need to be addressed and examine how agentic AI can be used as a solution.
Race to the bottom
One FMCG participant raised the point that data leaders need to ensure agentic AI solutions are not just creating a new baseline that all competitors must match but are truly driving strategic differentiation. There is a risk of setting a new “race to the bottom” environment as it seems likely that, much like the implementation of generative AI, agentic AI tools will soon be ready as plug-and-play or no-code additions to existing solutions.
Additionally, while there is the potential for agentic AI to revolutionise data acquisition by handling unstructured data more efficiently, an organisation may either be too slow or too fast in adopting agentic AI, and the new baseline of operational efficiency may become a disadvantage if everyone adopts it.
Beware reputational damage
Elsewhere, a word of caution from one participant was raised as they explained the challenge of deriving new information using agentic AI. The member cited a use case where AI-generated enrichment data might hallucinate and provide incorrect information which would completely undermine the solution, as well as damage the reputation of data as a solution within an organisation. The suggestion was to start with simple use cases and building upon them to avoid overcomplicating the implementation and to repeatedly demonstrate value to non-data leaders.
Human-in-the-loop
Participants underscored the need to blend automation and human interaction, especially in moments of truth where empathy is crucial. This is particularly important for those in highly regulated industries and critical customer-facing organisations. There is still an essential need to keep humans in the loop and the ultimate decision-makers for instances where empathy and a human understanding of situations is needed.
