AI as new UI: Driving agentic process automation at enterprise scale

Insights

• AI as UI is enabling users to interact with systems through simple natural language instead of traditional clicks and forms.
• AI has mainly been used to enhance existing processes like summarizing information, writing emails, or improving chatbot interactions but not to transform them.
• Agentic Process Automation (APA) takes this further by allowing AI to understand user intent, take decisions, and complete entire tasks without human help.
• In enterprise use cases, APA has shown it can drastically simplify complex processes, like handling pricing errors or filing insurance claims, through conversational interfaces.
• For APA to succeed, AI needs to build new intelligence layers around invent, event or context, orchestration, and knowledge.
• To make AI as UI a success, companies should redesign processes with AI agents in mind, launch focused pilot projects, and upskill employees to work with these new systems.

When ChatGPT burst onto the scene in November 2022, it offered a compelling opportunity to reimagine conversational user interfaces (UIs). The convincing natural language capabilities of generative artificial intelligence (AI) created a new paradigm for how we could interact with computers and how customers and employees could engage with enterprises.

No more clicking, typing, reading, or searching. Just tell your computer what you want, and it will understand and respond.

AI could become our new UI.

The Reality Check: Where AI Stands Today

Two years on, we seem no nearer to this reality. Generative AI’s success so far lies in the relaying, restructuring, or reframing of information. For instance, generative AI excels at summarizing and explaining online search information, and summarizing and highlighting actions from call transcripts. The technology can be used to improve written text in emails or documents and build more convincing chatbots. And of course, it can create convincing art, including music, images, and poetry.

While useful and often impressive, none of these examples get close to what many had hoped AI would bring, or rather, what it could take away: The drudgery of clicking, typing, scrolling, and submitting. A 2019 Forbes article reported that data professionals lose 50% of their weekly time — 30% on searching for, governing, and preparing data, and 20% on duplicating work. We still spend more time finding existing information, searching data through different portals, websites, or other sources to verify the data, rather than actually creating new knowledge from it.

Rethinking Enterprise Workflows with AI as UI

To eliminate these burdens, companies need to realize this vision of AI as UI: A future where humans can focus on their real work and leave the bureaucracy of working in a large enterprise to AI.

Traditional business process management (BPM) focuses on digitizing physical forms and capturing data through UIs. However, this approach still requires manual input and intervention.

This transformation allows for more efficient, accurate, and faster processing, fundamentally changing how businesses operate and interact with their systems. It's an exciting development that promises to enhance productivity and innovation across industries.

The good news is that this is not a fantasy. Many of the critical elements required to build this future are already in place. They reside within the already established BPM tools, such as Appian, Pega, and ServiceNow, which drive many of the critical business support functions and workflows for large enterprises.

But for APA to truly work, another layer of intelligence needs to be created in and around these systems. Traditional process management tools rely on programmed rules organized in fixed workflows. The user’s experience is defined by the design of the system. Unfortunately, this typically means that — regardless of the query type, urgency, or expertise of the user — all go through a similar one-size-fits-all process.

AI can bring an entirely new and personalized approach to user engagement in a process. Rather than having to rely on a monolithic process design, user queries can be delivered through independent agents that autonomously route queries, working with each other to understand, gather, and validate the correct information needed for a response.

Take, for instance, an employee checking the status of their pension and updating their monthly contribution. Traditionally, this would require multiple searches and menu clicks on an intranet system as well as reading documents and guidelines to understand which policy applies in their region.

With APA, an employee could simply type “increase my pension contribution by 2%” into a chat box. The system would then ask any necessary follow up questions before finalizing the task.

Now, imagine applying that same intuitive experience to high stakes business scenarios like a pricing error during the Black Friday sales period. Traditionally, resolving such issues demands the involvement of multiple teams — such as product information management, order processing, and finance — each working in silos to diagnose the issue, fix it, and adjust past transactions. This manual, fragmented process causes delays, increases the risk of errors, and diverts critical resources at the worst possible time.

AI autonomously orchestrates actions across systems like product information management, order management, and finance. Once the user approves a strategy, the system executes it, handling everything from order adjustments to customer notifications while keeping the user in control. This shifts employees from manually correcting issues to strategically approving AI-driven resolutions (Figure 1).

Figure 1. Supply chain automation through AI-powered transformation

Source: Infosys

The Four Pillars of Agentic Automation

To achieve this kind of faster decision-making and better user experience, four technological pillars are needed. The first two focus on understanding the user, both their intent and the event they are initiating. This encompasses the user’s context and the context of their action. The second two pillars involve understanding the data, processes, systems, and policies required to deliver the necessary information. These can be categorized into orchestration and knowledge. Each of these four areas benefits from predictive and prescriptive capabilities of AI systems that can operate autonomously and work together seamlessly.

Let’s take a closer look at how these pillars work together through an example from the insurance industry. When a customer reports an incident, “intent” enables the AI to understand the customer’s goal, such as filing a claim and seeking a resolution. At the same time, “event” captures the context of the interaction, such as the urgency of the situation, the sentiment expressed, and the type of incident. It accomplishes these tasks through analytics tools like sentiment analysis, voice-to-text, and image recognition. As this data is gathered from documents, photos, and channels, “orchestration” comes into play, automatically notifying the appropriate claims specialist and generating an initial assessment package by coordinating across various backend systems. In the background, “knowledge” powers the decision-making process through semantic search and knowledge graphs, enabling the AI to locate relevant policy coverage, estimate repair costs, and even schedule repairs. It also suggests next steps or autonomously triggers related actions, which reduces manual efforts and accelerates claim resolution (Figure 2).

This approach not only automates the decision-making process but also ensures seamless data transfer between various systems, including policy administration and claims management systems. The result is the elimination of manual entries and the reduction of errors. Specialists can then review AI’s recommendations and provide expert judgment on complex aspects of the claim, leading to high-quality customer service at every step of interaction.

Figure 2. Claims automation process through UI as AI

Source: Infosys

Building the Future: From Vision to Execution

AI as UI is a near-term option for many companies and is poised to revolutionize software development and architecture. In the future, providers will deliver services through prebuilt AI agents — creating an AI-driven workforce — and will adopt the service-as-a-software model. By building small language models that don't require extensive hardware and can outperform larger models in specific domains, providers can effectively test AI agents for various use cases. As traditional manual operations become obsolete, it's imperative to adopt these autonomous models to increase overall efficiency.

Companies should follow a three-step approach to expedite this service delivery transformation:

1. AI agent design: Start with an AI agent design process that replaces traditional design and business analysis with an AI-first mindset. Prioritize platform readiness by focusing on data readiness, data quality, and data engineering.
2. Pilot projects: Begin with small-scale experiments and pilot projects to demonstrate value. Identify low-hanging fruit by targeting areas with complex processes and high business value and then choose areas with organized and clean data for initial implementation.
3. Skill transformation: Equip employees and customers with the skills needed for success. Educate users on the changes and how to build and improve the systems themselves. Emphasize continuous learning and adaptation to new AI-driven processes.

It’s crucial to recognize that this transformation isn’t some far-off future concept, it’s happening right now. Legacy systems and outdated approaches simply can’t keep pace with the demands of today’s fast-moving, data-driven environment. Organizations that act decisively to adopt this three-step approach will not only accelerate their AI adoption but also position themselves to deliver more agile, efficient, and customer-centric services which are essential to secure the competitive advantage for long-term success.