The Realities of Implementing a Cognitive Service Desk (Part 1)
Modern AI history. In 2011 IBM’s flagship artificial intelligence (AI) technology known as Watson beat former winners Brad Rutter and Ken Jennings on Jeopardy! While this event was hardly ground-breaking in the long and storied history of AI dating back to the 1950’s, it does mark the turning point when AI started to become mainstream, capturing the public imagination and motivating a new round of technology startups to take a deeper look at cognitive computing and the potential solutions that it can provide.
Today a number of new entrants in the AI market offer commercially available capabilities such as “natural language recognition”, “machine learning techniques” and “cognitive reasoning engines.” Some of the entrants are combining these features into specific solution sets such as service-desk automation and customer-service support, allowing them to focus on very specific markets.
Within the IT industry, this strategy is generating a considerable amount of hype and excitement with some industry analysts going so far as to declare themselves as “true believers” in the future potential of AI being used in support technology.
It is not difficult to see the appeal of introducing AI into a service-desk operation, because most organizations will be attracted to the benefits of significantly reducing costs, creating a consistent end-user experience and resolving issues more quickly and effectively. However, as with any hype cycle, there is a lot of future-state selling happening in the market. As a result, this 2 part blog series aims to balance the excitement against the reality when it comes to implementing a cognitive service-desk solution.
All incidents are not equal. A quick review of the brochures and marketing material from service providers and AI niche players reveals a wide range of claims as to the potential benefits of using their technology. On the whole, they typically claim that their technology is capable of reducing Level 1 service-desk incidents by more than 50%, with some going as far as 90%. This is exciting but also potentially a little misleading, because they provide scant details about the types of Level 1 incidents they are referring to and not all incidents are the same.
As it happens, some incidents, typically ones that can be described as having a deterministic (predicable and repeatable) resolution, are extremely well suited for an AI solution. An example of a deterministic incident would be an end-user request for a password reset. Every time a service desk receives this incident – typically one of the largest sources of incidents for an organization – the resolution is the same, with the end user having their password reset to a new temporary one that they have to change as soon as they can. The consistency of this resolution means that this deterministic incident can be readily supported by an AI solution. Not surprisingly, organizations that have a greater than significant amount of deterministic Level 1 service-desk incidents are extremely good candidates for AI solutions and will almost certainly realize the associated benefits by adopting the technology.
It gets a little more complex with non-deterministic service-desk incidents, because the solutions to those are more difficult to implement and in some cases not worth considering as part of an AI solution. An example of an non-deterministic incident would be a complaint from an end user experiencing difficulty accessing a corporate web application hosted in the cloud – the sort of application often used to provide payroll or travel and expense management for an organization. This incident has hundreds of potential causes, from browser plug-in version incompatibility to a denial-of-service attack on the cloud provider’s infrastructure. In the case when there are multiple resolutions, the AI system has to be able to determine the best-fit resolution for a specific user at a specific point of time.
The better AI systems on the market today have the capability to navigate through a myriad of options and present the end user with a best-fit resolution that is correct the majority of the time. However, that is possible only if the AI system has access to all the real-time information it needs to make the decision.
Tags- Applied Service Management Artificial Intelligence Automation Autonomics Cloud Cognitive Agents Digital Business digital IT Digital Service Management Hybrid IT IT Operations ITOM ITSM Service desk