ChatGPT and Service Management

ChatGPT is hot topic these days. While the technology is not yet as reliable to become a fully trusted knowledge source it clearly demonstrates the disruptive potential in many application fields. I wanted to think loud about how Service Management could be disrupted by generative models and how the industry and expert community might embrace this transformation to continue being relevant in years.

Knowledge management

Can generative models completely replace knowledge management discipline? How will the role of knowledge managers evolve under influence of AI? I think of a concept of a living dynamic knowledge base, which will be auto generated by AI, and which will infer their knowledge from the huge amount of data sources availability in open internet as well as within corporate networks. Authoring of knowledge articles will become a thing of the past. These models could in future mine all enterprises communications: emails, chats and even voice calls to learn about how common problems are solved by people. They can then present this knowledge to both service customers and agents not only to increase their productivity but also to give them emotional capacity to empathize. It is companies perceived for their empathetic customer service where people return. Focus will shift to monitoring of the model accuracy and ensuring the recommendations of the generative models remain relevant and accurate over time as data drifts.

Request management

Requests are the most common interaction in the service world. Key focus is to organize different types of requests so that they can be easily found and consumed on one end and that service teams can fulfill them as quickly as possible on another end.

Platforms like ServiceNow already provide today ways to recognize customers’ intent and to map it to right request type in the vast catalog of services. But to ensure reliable results these capabilities must rely on a well-structured and machine search optimized service catalog where each service is designed with a user experience in mind. For many organizations it is still more art than a science and a thin balancing act between ability to simplify individual requests and reduce the size of the catalog. Most successful service organizations are known for their catalogs of service offerings being compact, yet effective.

If I think of service request catalog as a structured code that can be represented as JSON object, or XML payload, I anticipate that generative models will help to generate highly interactive request forms even on fly. They could generate forms based on what they would have learned about services by mining communications within a company. To me it means a potentially endless catalog of services that grows with the organization. GPT may be both vice and virtue in this context – it helps humas to “templatize” common requests but will also drive increase of the catalog size. Classification of services into meaningful taxonomies will likely be a challenge, on which humans will need to concentrate with support of machine learning algorithms.

Automation

The lack of qualified service workforce and growing technology job wages put pressure on service organizations, which must deliver better service to more customers and reduce unit costs at the same time. Automation is essential to successful service organizations. Yet the common hurdles to automation include deficit of funding, lack of skilled personnel and siloed mindset of the organizations. Automation is still considered luxury for many service organizations. Companies like ServiceNow or UIPath started to remove barriers by creating citizen development capabilities to increase access to automation for businesses. The ability of generative models to create program code has a potential to bring citizen development to a completely different level. Imagine that 80% of all automated workflows can be generated automatically by an AI assistant even on the fly after a customer request has been received. Service owners who became citizen developers today will become service owners again. They will be able to focus more on leading their teams, and increasing competitiveness of their service offerings, while their AI assistants will be doing the heavy lifting of creating digitized automated workflows.

Challenges for the generative models in service management

Today GPT models come with a big disclaimer – validate outputs before you trust them. While this approach is good for AI enthusiasts and early adopters among knowledge workers, service customers need services they can trust and rely upon. When it is no longer clear whether a human or machine stands behind the computer screen, how can we provide then accurate solutions that customers will not have to verify? Will this be governed by customer communities who will perform the role of free beta testers or service organizations should introduce special GPT QA roles?

Another consideration is the balance between the efforts to achieve accurate models and their ROI. Does it still make sense to invest in such models knowing how much computing power and human efforts are required to collect, clean, label necessary data and train models on it. Not to mention the ethical aspect of this process, like the poor working conditions of the data labelers.

Conclusion

Generative models possess a significant disruption potential in service management. While they are not scalable for mass adoption, the industry needs to start shaping its GPT transformation today. Proofing use cases, addressing ethical concerns and making necessary cultural shifts will take years. I hope with proper intentions GPT technology will turn service management into a competitive power of businesses and make it more empathetic for they customers.