The Potential of GenAI in Enterprise Service Management

Peter Zalman
7 min readAug 27, 2023

A new era of AI tools in the realm of generative AI and LLM brings truly fascinating opportunities. Unlike previous technological hypes, such as 3D TVs and the Metaverse, GenAI tools are already making strides in solving real problems. While there is no shortage of futuristic predictions in this crowded space, I’ve decided to contribute with a few points of my own.

Framing

One of the noticeable outcomes of GenAI is the advancements in LLM, offering a realistic chatbot experience. Existing chatbots were essentially glorified answering machines, whereas ChatGPT, Bart, or Bing can engage in real conversations, responding to prompts and inputs.

There is an obvious shortcut to be realized in service management, especially for IT Helpdesk services. Some references indicate that organizations are either aiming for or already implementing workflows where a significant portion, if not the entire support process, is offloaded to LLM-generated chatbot or email responses.

Is Suumit referring to Incident Response or Support Case Deflection?

As with any meaningful discussion, it is essential to establish a shared terminology and framing. I frame this within the world of internal services (ESM), e.g., reaching out to your internal HR team or reporting lost company phone to IT. This is a distinct domain compared to general customer service management, e.g., reaching out to Adobe support seeking details about misbehaving Photoshop editing features. In this context, we need to align on at least four ITIL definitions.

Terminology (ITIL)

Incident

An unplanned event that causes or may cause a disruption to or a reduction in the quality of a service, requiring an emergency response. The primary goal of incident management is to restore the service to its usual operational state as quickly as possible.

Service Request

A service is a type of work that is performed for others. Services can be intangible, meaning they cannot be touched or held, like consulting or HR advice. Services can also be tangible, meaning they involve a physical product or activity, like purchasing a software license or booking a travel. Service request is a formal request from a company employee for something to be provided.

Incident Deflection

This refers to preventing incidents from happening in the first place or stopping them from reaching the support team. It’s about directing employees to self-help resources or automating solutions so that the issue is addressed without the need for direct intervention by a support agent.

Incident Resolution

This refers to the actions and processes taken to solve or address an incident after it has occurred. The goal is to restore normal service operations as quickly as possible while minimizing the negative impact on business operations.

GenAI Opportunities

With these terms established avoiding confusion between Deflection and Resolution, we can discuss how GenAI can be leveraged or if the rush to innovate at all costs might backfire in areas where the decisions are not made in the context of IT Service Management and GBS (Global Shared Services) system design.

Incident Deflection

A ChatGPT-like experience seems like an excellent way to engage with employees seeking instructions on using a specific application or inquiring about company policies, such as travel.

However, there is a risk associated with the overly enthusiastic use of LLM for Incident Deflection in environments with lower ITIL process maturity. In such contexts, the use of Incident Management workflows is often mistaken for service requests. Employees contact service desk to request a service and do not just seek additional instruction or policy explanation. Instead of deflecting, these interactions need to be addressed by a specific team or fulfilled through a workflow automation. Imagine reporting a stolen iPhone, but instead of receiving a replacement acknowledgement and triggering a workflow behind-the-scenes to ship you a new one, a chatbot engage in an endless discussion offering you advice on how to communicate without a phone. Or consider reporting network or VPN outage — an Incident you can’t resolve by yourself. But rather than an immediate escalation to the networking team, the machine starts a conversation about how “you” should further explore the topic of networking, which you might not be interested in at the moment.

From the perspective of a service desk, every deflected incident is deemed a success. However, from the broader business standpoint, even accurate ChatGPT answers can present the risk of a generic user experience that impacts the employee experience, potentially leading to extended incident resolution times and delays in service request fulfillment.

Ages 55 aren’t a fan of chatbots

In automated Incident Deflection or chatbot responses, it is crucial to establish a workflow and data sources to prevent LLM from generating hallucinated answers. When referring to a specific company’s e.g., cybersecurity policy, employees should find exact quotations and references. Answers must also adhere to the organizational access level governance. Not all employees necessarily see the same set of information e.g., for travel, insurance, or even IT device eligibility that depend on their role and employment status.

The Risk of Placing Too Much Faith In AI

Incident Resolution

Workflows that already enable faster, more accurate Incident Resolution powered by AI/ML algorithms hold great potential also for GenAI. The primary responsibility of L1 service desk is often not fixing the underlying problem, like implementing a fix for software bug or increasing network capacity, but merely restoring the service to its normal state, even if it’s through temporary solutions or redirecting to self-service advice bypass method.

Incident vs. Problem

Incident resolution often requires escalation to and collaboration across multiple teams — operations, networking, application development and QA. When an incident is acknowledged, the opportunities with GenAI lie in making support and operations teams more productive. For example, Incident Alerts, when hundreds of employees report a similar Incident at once, can benefit from AI/ML algorithms automatically categorizing incidents. The initial root-cause analysis often involves an all-hands-on-deck war room investigation, engaging dozens of experts across various communication channels — Zoom calls, Slack channels, and Incident Details. If the cause is not identified during the initial phase, the subsequent investigating team can benefit from GenAI text summarizers. These tools synthesize insights about the discussions, what components of the investigation have been ruled out, and the potential next steps.

Incident Auto-Resolution

A specific opportunity for AI/ML and GenAI is in the distinct field of Incident Auto-Resolution. AI/ML algorithms can be effectively used to predict incidents based on historical patterns, either by using an on-device agent or other technologies, predicting possible solutions to an incident with minimal or no user involvement.

The opportunity with GenAI lies in responding to an Incident that wasn’t deflected, probing for more details to speed up resolution, redirecting to a service offering from the catalog, or serving as an effective “waiting-room” experience with self-service initial troubleshooting.

Service Request

True conversational user interfaces, enabled by GenAI and LLM, offer alternatives to traditional service channels. It’s crucial to highlight that the primary objective of service management is to fulfill the request and execute the corresponding action, not merely to respond with structured content or engage in a prolonged conversation. To illustrate this distinction, consider the example chat below.

Service request in conversational UI.

Conversational UI in the example above offered several services from a structured catalog and facilitated the service fulfillment.

Summary

GenAI introduces exciting new opportunities in service management that extend far beyond Incident Deflection and chatbots rephrasing existing codified knowledge. Enhancements in service teams productivity, such as Incident summarization and Incident auto-categorization, lead to an improved user experience and a more productive workflows.

True innovation emerges in the service request fulfillment although conversational user interfaces shouldn’t be the only service channel for service catalog. When employees submit service requests, the primary aim is service automation and fulfillment, not merely to engage in a chat or deflecting the demand with self-service resources. For instance, when requesting a room booking, the response shouldn’t be a chatbot elaborating on the variety of available meeting rooms; the desired outcome is a booked room. Reporting a stolen phone should lead to a replacement order, a sensitive HR issue should automatically be escalated to the appropriate agent, and an incident caused by a software bug should be immediately routed to an application team who can prioritize the fix in implementation backlog.

Overemphasizing Incident Deflection at the expense of more sophisticated and automated Incident Auto-resolution might produce a Google/Facebook customer support experience, where users feel trapped in an endless loop of “more information” resources without ever reaching a resolution. In the realm of internal enterprise services, this does not promote a productive and satisfying employee experience.

Adoption of GenAI tools and workflows is built on the expectation of a high level of service management workflows maturity within the organization. Services must be fully digitized and fulfilled using modern platforms so that GenAI tools can access the data and APIs (Application Programming Interfaces) as well as the service catalog and knowledge bases build with granular and detailed access control mechanism and advanced content personalization.

Enterprise Service Catalog

--

--

Peter Zalman

I am crafting great ideas into working products and striving for balance between Design, Product and Engineering #UX. Views are my own.