Artificial Intelligence in Service Desks

DZone 's Guide to

Artificial Intelligence in Service Desks

Let's take a look at Artificial intelligence in service desks as well as explore a knowledge base and a business case.

· AI Zone ·
Free Resource
“Flawless customer service facilitates opportunity more than anything else; the opportunity to exceed any and all expectations.” ~ Than Merrill, CEO & Founder, FortuneBuilders

Resolving customer issues at the earliest is as critical as delivering a new product or service to customers. While organizations strive to achieve better customer service by optimizing key metrics such as Mean Time To Resolution (MTTR), Defect Removal Efficiency (DRE), etc., Artificial Intelligence comes in handy in catering to our needs to be faster and accurate in providing resolution.

Knowledge Base

Knowledge Base

The foundation for any service support system is to have a solid knowledge base. The knowledge could be of structured information around customer data, details of issues, or unstructured information such as audio or video attachment to the issue.

The knowledge base constantly evolves based on learning through new issues, added and updated content, and improved accuracy of information.

Service Desk

Service Desk

While the Information Technology Infrastructure Library (ITIL) has a different definition for service desks, help desks, and call centers, the purpose is the same for all in providing timely support to customers.

The perceived service quality of an organization is mostly through the effectiveness and efficiency of an organization’s service desk. While incidents that disrupt the availability of a service takes a higher priority to solve, there are service requests that could potentially influence customer satisfaction.

Challenges with the traditional approach:

  • It’s not just about talented Support Analysts: It is essential to have skilled analysts to keep up the Service Level Agreement in resolving issues. However, if the knowledge base isn’t organized effectively or readily available, it’s still a challenge for the support analyst in performing their job.
  • FAQ doesn’t help all the time: Every organization provides "frequently accessed questions," which helps to a certain extent. But if the end user is in hurry to get their problem sorted out, they mostly prefer to call up the help desk.
  • Beyond content management: Most of the time the focus seems to be on available content than understanding for whom the service is being provided. This is one of the reasons why self-service functionality is not as effective as it was intended for.

Business Case

Customizing the content and support based on the end user’s perspective and usage of the system is critical in resolving issues. Some key elements that help in proactive incident and request identification are:

  • Pattern of incidents by system and components
  • Pattern of service requests by persona and category
  • Analysis of historical misclassification of service requests that contributed to more cycle time
  • Changes to existing software or hardware that might potentially increase the inflow of requests
  • Added or updated load to the existing system that might trigger new incidents or service requests.

 AI Techniques

AI Techniques


Time series can be leveraged to predict and forecast the inflow of incidents and service requests. Since the model can be further decomposed to trend seasonal effects and random variation, it helps in understanding the pattern of inflow further and plan accordingly to address them.

Classification and Clustering

While classification techniques such as the decision tree, Naïve Bayes, can help to categorize the inflow of issues into known theme/types, clustering techniques such as k-means and hierarchical can assist grouping them based on the pattern. Supervised learning helps in expediting the resolution by moving the new issue into the appropriate queue. Unsupervised learning is a boon when we don’t have prior knowledge but need information retrieval quickly and accurately.


Everyone prefers a personalized experience, and the service industry is no exception. In fact, based on the profile of the customer, products/services they use, historical requests, ecosystem, and several other key factors, the organizations can make the self-service powerful.

For an effective recommendation, building the feedback system on past recommendations is essential. We can use scoring (such as net promoter score with a scale of 1-10) or simple boolean ("did it help") to obtain the feedback quantitatively.

Let us explore some of the techniques that can potentially help increase the accuracy and precision of the recommendations:

Collaborative Filtering

  • Persona-Based: Most of the time, we tend to focus on the content and type of requests when we try to resolve an issue. If we understand each persona, their needs and usual transactions, and the challenges faced accomplishing their tasks, we can provide appropriate recommendations to another similar persona. For example, the primary responsibility of a project administrator role is to grant or revoke access. If one administrator historically has challenges with an application, there is a higher probability that the resolution might work for another administrator with similar credentials and challenges.   
  • Asset/Component-Based: Due to the dependencies among different systems and components, the issues may be interrelated. Identifying similarities and associated impact help in troubleshooting the issues quicker. For example, network latency could cause erroneous transaction in a website. Hence, it’s important to understand the holistic view of the system and the relationships among different entities prior to troubleshooting issues at a component level.

The K Nearest Neighbor algorithm can be leveraged to find clusters of similar personas, assets, and components based on a number of related requests and recommend using average rating of top-k nearest neighbors. If there are hidden features underlying the interactions between personas, assets, and requests, we can apply singular value decomposition, one of the Matrix Factorization models for identifying those hidden factors for building suggestions.

For a deep dive on recommender system through collaborative filtering, I found this blog to be very informative and hands-on.


As much as the knowledge base, content management, and efficient support analysts are essential, the tools and techniques of Artificial Intelligence play an important role in resolving the service requests and issues effectively. Recommendation systems along with continuous improvement through feedback makes service desks customer-friendly.

analyst, artificial inteligence, classification, clustering, customer satiisfaction, forecasting, recommendation, service desk management, support

Opinions expressed by DZone contributors are their own.

{{ parent.title || parent.header.title}}

{{ parent.tldr }}

{{ parent.urlSource.name }}