Call wrap-up with AI automation

March 22, 2024 | 8 minute read
Oleg Kobchenko
Principal Applications Engineer, Oracle Utilities Management Solutions for Energy and Water
Text Size 100%:

AI is making a significant impact on utilities, of which customer service is a key focus.  In particular, AI is proving to be successful in automating repetitive customer inquiries and enhancing the overall customer experience. Utilities have already deployed AI to manage and optimize their customer support operations, using chatbots to streamline customer engagement and provide quick answers to frequently asked questions, which provides a positive customer experience. AI can also be deployed to provide on-demand, self-service, further reducing wait times.

The value of AI can be applied to other areas of importance to utilities, like managing internal processes and back-office tasks. For example, a quick generative AI service can summarize call transcripts, freeing up human resources. AI can also automate call wrap-up functions, helping operators create internal service call records and provide links and concise descriptions to explain the benefits of certain programs—both of which reduce the need to use conventional multi-step workflows and manual searches.

Call wrap-up automation

Oracle Utilities Customer Cloud Service helps utilities of all sizes provide customer care, service order, metering, and billing. An important yet time-consuming part of customer care is addressing customer concerns in direct communication through phone calls, live chats, and text chats, and in following up the customer calls to various automated systems, such as chat bots, and social media feeds, support forums, and so on.

In this blog post, we show how AI improves a typical call wrap-up process based on live calls and chat transcripts. During the wrap-up stage, the operator must ensure that the call is accurately documented in the system and any follow-up service calls and communications are initiated. Any program enrollments discussed in the call were either signed up for or marked as declined to avoid repeat offers in subsequent calls.

Overall, applying AI to the call wrap-up procedures plays an important role in both enhancing the customer experience and optimizing the efficiency of utilities customer care personnel.

Customer Service Edge App interface

The user interface hub for the customer operator workflow is the Customer 360 portal, part of Oracle Utilities Customer Cloud Service suite of products, with all the critical information and links to all the details. On the right is the Chat Wrap-Up panel, where you can review the transcript of the call and further actions available with the Wrap-Up button.

An overview of the Customer 360 portal, showing the Chat Wrap-Up panel.


The Wrap-Up button leads to the Chat Wrap-Up portal where the call is summarized, and related actions, such as service calls and program sign-ups, are created or documented with the help of AI Automation.

The following figure shows an example of a typical outage call scenario with corresponding AI automation calls:

Example of a typical outage call scenario with corresponding AI automation calls


AI functions use the service call details as input and AI prompts. Several actions are available as buttons, including Analyze, Associate SR, Create SR, and Program. The Analyze button performs the initial AI call, and the other action buttons perform follow-up actions and act as indicators of features discussed during the call. The AI call determines which features are present in the call, such as service issue, appointment discussed, program, and service changes. Related objects, such as customer contact record, outage calls, and service calls, are either automatically created or found if they already exist in the system. The AI automation uses these objects to extract descriptions of specific features, such as service issue, appointment details, or programs.

In the following figure, a customer is discussing a potential program sign-up. If they decline the program, it’s recorded in the system, so that later, an operator has a chance to discuss a different program.

customer is discussing a potential program sign-up


The AI function extracts the program description from the call transcript and uses it to characterize and identify the program. It uses program identification to find the specific program in the system. An embedding API passes a list of available program descriptions and the extracted program from the call. It determines which program is the most similar and the similarity confidence. Based on the insight determination algorithms specific to the programs, the automation function determines the customer’s enrollment status and shows the Enroll or Go To buttons. User Declined is a special determination extracted by prompt engineering, which guides customer interactions in future calls.

Use cases

We used the following use cases to demonstrate the automated call wrap-up. Each use case represents a problem about which customers contacted the customer service and how the call wrap-up system helped identify the solutions, facilitate documenting the issues, and create appropriate follow up actions quickly.

The descriptions of the use cases were automatically generated using the summarization features of Generative AI. The summaries are created using well-formed natural language queries. After the summary, the role of AI in the call wrap-up process is provided.

Loss of electricity: Schedule outage call

A customer reaches out to report an issue with their electricity. After confirming that the problem has been present since this morning and checking the circuit breakers, the operator arranges for a repair crew to attend tomorrow at 10 a.m. The customer confirms the appointment.

AI role: Summarize the call, identify the outage call, and associate with the customer call record.

High electric bill: Smart Time of Use program

The customer’s electric bill has increased by 25% compared to the same period last year. The conversation reveals that the increase is likely because of the installation of a new, less economical air-conditioning unit and an increase in the area cooled. The operator recommends the customer enroll in the Smart Time of Use program, which offers flexible rates to run appliances during off-peak hours to reduce the electrical bill. The operator helps the customer sign up for the program, which takes effect in the next billing cycle.

AI role: Summarize the call, extract program description, and use it to identify the specific program in the system.

Water leak: Send out a crew for a service call

The customer reports a water leak starting a few houses down the road and draining further down the street. The operator schedules a crew to assess the situation.

AI role: Summarize the call, extract issue description, and use it to create a service call.

Increased gas consumption: Energy Saving Insulation program

The customer contacts the gas utility support service because their gas bill has increased because of higher consumption this season. They explain that they have an extra room in their house because of their son moving in, and they’re looking for recommendations to help insulate the house and reduce their gas bill. The operator suggests the Energy Saving Insulation program and the option to purchase an energy it, which the customer declines. The customer is working on their own insulation project.

AI role: Summarize the call, extract program description, and use it to identify the specific program in the system. Identify customer intent to decline the program to mark it in the system.

Extra charge on energy bill: Sign up for automatic payments to avoid late fees

The customer discusses a late fee on their bill from last month with the operator because they were on vacation and missed the payment deadline. The operator agrees to waive the fee as a one-time exception and recommends enrolling in the Auto-Pay program to avoid late fees in the future. The customer agrees and signs up for the Auto-Pay program, and the operator gathers their payment information to process the enrollment.

AI role: Summarize the call, extract program description, and use it to identify the specific program in the system and help associate the program with the customer call record.

Calling AI from edge app

The implementation of AI Automation uses the following system topology, which represents the API calls and data flow between the edge app and Oracle AI services.

The key to the AI Automation system is the intermediate AI communication server, which provides the abstraction layer between the edge app and the Oracle AI services. The communication server is responsible for the AI call configuration and parameters, prompt engineering, and data formatting and conversions.

Workflow of the applications and services in the deployment, from Oracle Utilities Application Framework to AI communication to Oracle AI

Prompt engineering

The overall strategy that AI Prompt Engineering uses is to favor multistage processing with simpler specialized calls, instead of trying to compose a single large complicated prompt with multiple tasks and goals. This method also provides opportunities for logic branching and user overrides. The AI prompts used the following organization principles:

  • From general to specific
  • Allow for evaluating intermediate responses to choose the next actions
  • Follow up with more specific calls

The same AI automation system is used for both automated processing and documenting with human readable descriptions. To facilitate both these aspects of AI interaction, we used the following approaches:

  • For automated processing of AI responses, consistency of terminology is important, such as always referring to the call participants as “customer” and “operator.” Specific keywords to look for and return in extraction and determining the outcomes must be defined, such as Is Program Discussed and Is Program Declined. Both prompt input and result examples used strict structure and data formatting.
  • To improve readability of generated documentation, generative AI parameters allow specifying the level of variability of the responses. You can configure these parameters in the application to improve the quality of the responses and select the most appropriate generated text variants.

Identification with embedding

Besides generative AI, we used embedding, another AI style, to identify which specific programs were discussed in the transcript. Identification consists of two stages: Program description extraction from the call transcript and program identification from the list of existing program descriptions, embedded in the AI call. The result of identification is the index of the most similar program from the list and the similarity confidence. The operator can then override the choice of the matched program using the confidence level: If the confidence is acceptable, then they can proceed with the found item. Otherwise, they review and decide as appropriate.


Using AI in utilities is an investment with the potential to help enhance customer service, and customer satisfaction. AI can also help utilities more efficiently manage and optimize their operations. AI algorithms can analyze data to identify issues and provide solutions, such as predicting customer concerns and offering personalized energy plans and other programs. AI can also automate service calls and other call wrap-up tasks, optimizing and facilitating customers contacting human support or automated customer service. Furthermore, it can streamline customer-facing processes, such as new service connections, billing, and program adoption.

For more information, see the following resources:

Oleg Kobchenko

Principal Applications Engineer, Oracle Utilities Management Solutions for Energy and Water

Previous Post

An AI application that can chat with any service

Andrii Kovalov | 9 min read

Next Post

Enhancing OCI Data Science: Unveiling the New Autoscaling Feature for Model Deployment

Abhinav Gupta AG | 13 min read