By Ramprakash Ramamoorthy
The fear of being outperformed is pressurising businesses across the globe to adopt agentic AI for any and all business operations. Without pausing to evaluate their actual requirements or readiness, these businesses are making room for a technology that requires solid tech groundwork built on digital maturity and interconnectivity.
At its core, agents are autonomous and dynamic decision-making systems capable of independently thinking and executing operations with minimal human intervention; businesses gravely underestimate the digital readiness needed to adopt such technology.
Is your organisation prepared for agents?
Decision makers should set realistic expectations for agent implementation. AI is not plug-and-play, and it is impractical to expect agents to improve business productivity and generate returns instantly. Fragmented data channels continue to hold companies back. When different technology platforms within the same organisation are unable to communicate with each other, agents lose their versatility to work across platforms, impacting their potential and the final output. Your first step should be to evaluate if your organisation’s digital systems are well integrated, and have clean data.
Agentic AI is better suited for ‘80% problems’. These are tasks where error tolerance is high and an 80% success rate is acceptable, such as addressing L1 support tickets. In contrast, 100% problems, for instance, digitising an invoice or reviewing legal documents, are ones where absolute precision is required. Does this business process need an agent?
You can always test the waters by experimenting with other forms of automation before moving to agents. Before jumping in to implement an agent, ask: Does this process have to be agentic?
Agents are touted to be the future of digital businesses. The reality is that they are not necessary for all processes. AI literacy allows decision makers to expand their horizons on other approaches, including say, automated workflows. These structured and rule-based automation tools are capable of providing pre-determined outcomes. So, though workflows are not adaptable or dynamic like agents, they are well equipped to handle predictable, sequential chains of tasks while allowing humans to oversee the entire process.
For companies considering AI agents, workflows can be a great starting point. They can handle processes that have fewer decision hops, and do not need a high level of digital maturity. Let’s take a look at a ticket-booking process within an organisation. A user raises a travel request on an expense processing platform. The travel request needs to be approved by the user’s manager, after which the travel desk takes over. The travel desk will have to coordinate with the user to know their preferences and make decisions to book the ticket based on various factors, including ticket prices, destination, mode of travel, weather, and state of affairs. The need for dynamic decision making and seamless platform interoperability is high in this case.
Agentic AI excels at dealing with such processes that have a lot of decision hops. Agents can pull data from different platforms, reason, and take the best decisions independently. Organisations can also utilise agents in customer service and coding domains since it has been built to make dynamic decisions to resolve queries, refer to multiple databases, and retrieve relevant data. On the other hand, when a process has a predetermined set of outcomes, workflows are sufficient. For instance, in the case of talent onboarding, since the processes are defined and there is barely any scope for autonomous decision making, deterministic workflows can be utilised.
Agents have an edge over workflows in areas like customer support. For one, it is quicker than a workflow, thereby accelerating the ticket resolution process. Agents can guide customers at various touchpoints, gauge the customer’s tone and intent and provide relevant resolution promptly through its dynamic decision making skills. Unlike workflows that reset every time, agents are capable of retaining the memory and context of previous conversations with a customer. This translates to improved customer experience and, in turn, minimise attrition. Agents are also capable of collaborating with other backend systems like CRM and billings, as well as coordinating with other agents to solve problems multidimensionally.
Decision makers should remain level-headed in playing the AI game and prepare their businesses for agents by first, improving the overall digital maturity of the organisation, and secondly, by brushing up on their AI literacy before deciding to jump into any AI offering.
Understanding how AI works goes hand in hand with figuring out how it can contribute to the business. Not all processes need to be agentic. So, businesses should not rush into adopting AI agents. Instead, they should first understand their business requirements and the nuances of automation to truly evaluate how they can harness AI. It is necessary to remember that the real challenge of intelligent automation is not adopting agents, but rather evaluating the business context to ascertain the best AI approach for different use cases.
