• Written by Sonali Subhadarsini, Product Manager

A customer service representative, working late on a Friday, needs to answer a complex query from a long-standing client, and turns to the enterprise AI system for help, but the system was built by a team that has since moved on, and the representative is not sure what they can and cannot ask, or what the system can and cannot see.

Permission at retrieval

The traditional approach to controlling what an AI system can do is to instruct the model itself, but this can be problematic when the system is used by people who did not build it, as the customer service representative may not understand the nuances of the model's permissions, and may inadvertently ask a question that reveals sensitive information, or one that the system is not permitted to answer. A better approach is to enforce permission at the point of retrieval, so that the system checks what the user is allowed to see before returning an answer.

This requires a deep understanding of the organisation's security policies, and a system that can integrate with existing permission structures, such as active directory or role-based access control, and it also requires a system that can handle the complexity of multiple permission levels, and can provide clear and concise error messages when a user attempts to access information they are not permitted to see.

Auditability

Auditability is critical in an enterprise AI system, as it allows the organisation to track what questions are being asked, what answers are being returned, and on what data the answers are based, this is particularly important in customer service, where the organisation needs to be able to track the history of a customer's interactions, and ensure that the AI system is not providing inconsistent or contradictory answers.

A good audit trail should include information such as the user who asked the question, the question itself, the answer returned, and the data sources used to generate the answer, and should be easily accessible to authorised personnel, such as the customer service manager, or the data protection officer, and should be stored in a secure and tamper-proof manner, such as a write-only database, or a secure logging system.

Dependent data

The effectiveness of an enterprise AI system in customer service depends on the quality and completeness of the underlying data, including ticket history, knowledge base articles, product and order records, and escalation history, and the system should be able to integrate with existing systems, such as the customer relationship management system, or the enterprise resource planning system, and should be able to handle the complexity of multiple data sources, and provide a unified view of the customer's information.

The category manager, for example, may need to compare the answers provided by the AI system with the information in the knowledge base, to ensure that the system is providing accurate and up-to-date information, and the shift supervisor may need to review the ticket history to ensure that the AI system is not providing answers that are inconsistent with the customer's previous interactions.

Changes to customer service

The use of an enterprise AI system in customer service can have a significant impact on the way that customer service is delivered, as it allows the organisation to provide answers to complex queries quickly and accurately, without the need for human intervention, and can help to reduce the time to answer, and improve the first contact resolution rate, and can also help to identify root causes of problems, and feed this information back to the product development team, to improve the overall quality of the product.

For example, the customer service representative may be able to use the AI system to answer a question about a product's compatibility with a particular operating system, and the system may be able to provide a detailed and accurate answer, based on the information in the knowledge base, and the product and order records, and the representative may be able to use this information to resolve the customer's issue quickly and efficiently, and the organisation may be able to use the information provided by the AI system to identify a trend in customer queries, and make changes to the product or the documentation to reduce the number of queries.

Measuring success

The success of an enterprise AI system in customer service can be measured using a variety of metrics, including first contact resolution, time to answer, and repeat contact rate, and the organisation should be able to track these metrics over time, to see the impact of the AI system on customer service, and should be able to use this information to make adjustments to the system, and improve its performance, and should also be able to compare the performance of the AI system with that of human customer service representatives, to see where the system is providing the most value.

For more information on how to measure the success of an enterprise AI system, and how to use the metrics to improve the system's performance, the reader can visit the operations page, and for more information on how to get started with an enterprise AI system, the reader can visit the contact us page.

Limitations

An enterprise AI system is not a replacement for human customer service representatives, and should be used to augment and support the existing customer service team, rather than replace it, and the organisation should be aware of the limitations of the system, and should not rely solely on the system to provide customer service, and should have a plan in place for handling complex or sensitive queries, that require human intervention, and should also have a plan in place for handling queries that are outside the scope of the system's knowledge, or that require a level of empathy or understanding that the system is not capable of providing.