Method and apparatus for enhancing an integrated voice response system with generative ai functionality

Abstract

An Interactive Voice Response (IVR) system, comprising a user interface configured to receive an incoming user inquiry, a response generation module configured to process the inquiry and generate a response, a Generative AI agent interface, wherein the response generation module is configured to direct at least partially the user inquiry through the Generative AI agent interface to a Generative AI agent for processing, and wherein the response generation module is responsive to an output elicited from the Generative AI agent in response to the user inquiry and presented via the Generative AI agent interface to generate the response in response to the output and output the modified response via the user interface.

Description

METHOD AND APPARATUS FOR ENHANCING AN INTEGRATED VOICE RESPONSE SYSTEM WITH GENERATIVE AI FUNCTIONALITY Field of the invention The invention pertains to the field of Integrated Voice Response (IVR) systems, specifically to methods and systems designed to enhance the functionality of IVR with Generative AI capabilities, thereby delivering an improved user experience. IVR systems are telephony technology that allow automated interaction with callers, utilizing pre-recorded voice statements and menu options via a touch-tone keypad or speech recognition. These systems are traditionally employed for routing calls, providing information, and conducting transactions without human intervention. This invention also provides a method and associated technologies for converting a virtual agent, originally based on IVR technology, to one employing Generative AI to oƯer services to end-users. Notably, the invention encompasses a process for progressively transitioning from an IVR system to one leveraging Generative AI, thereby ensuring a seamless upgrade in service provision. The proposed enhancements aim to address limitations in conventional IVR systems, such as inflexible interaction patterns and limited context awareness, by integrating advanced AI techniques. These improvements enable a more dynamic and responsive interaction, allowing the system to better understand and fulfill user requests, ultimately leading to a superior user experience. 1 Background of the invention Traditional Interactive Voice Response (IVR) systems have been a cornerstone of customer service for many years. These systems utilize pre-recorded prompts and touch-tone or basic voice recognition inputs to navigate callers through a menu. They are cost-eƯective for managing predictable and repetitive inquiries, such as payment questions or account balances, and can eƯiciently direct calls to appropriate departments. However, these systems often frustrate customers due to their inflexible menus and limited adaptability, resulting in high rates of call abandonment. Traditional IVR systems are characterized by one or more of the following characteristics: Rule-based logic: Operate on explicitly defined rules, heuristics or decision trees; Deterministic behavior: Produces the same output for the same input, no learning over time; Limited context awareness – Context must be explicitly encoded into the system, limiting its ability to infer nuanced user needs. Human dependent Updates: Any changes or optimization to the system require manual intervention by developers or administrators. On the other hand, AI voice or text assistants leverage machine learning and natural language processing (NLP) to engage callers in a conversational, human-like way. Customers can ask questions naturally, and the AI interprets their needs and responds accurately. AI voice or text assistants oƯer natural, conversational interactions, contextual understanding, and scalability, making 2 them more adaptable and eƯicient for complex inquiries. AI agents are characterized by one or more of the following characteristics: a) Adaptive Behavior: Learns or improves over time based on data, interactions and feedback. b) Contextual Understanding: Uses advanced algorithms to understand context, user preferences, and nuanced inputs, such as tone, intent, etc. c) Autonomous Decision Making: Capable of making decisions in uncertain environments by leveraging probabilistic model and reinforcement learning. d) Real-Time learning: Adjusts to new patterns or trends dynamically, often without human intervention. e) Broad Functionality: Integrates across various domains, oƯering multi- tasking and scalability. It is generally accepted that that AI agents perform better than IVR counterparts, however there is a large installed base of IVR systems in the industry and transitioning from an IVR system that is currently running to an AI agent is a complex costly task. A typical user concern which slows the migration to AI agents, is that while the currently installed IVR system has deficiencies, they are well understood and still delivers some level of user service. While an AI agent is expected to perform better, a switch from one system to another still represents a level of uncertainty, in particular for large scale organizations that cannot aƯord a massive disruption in client services. 3 It is generally accepted that AI agents exhibit superior performance compared to Interactive Voice Response (IVR) systems. However, the industry retains a substantial installed base of IVR systems. Transitioning from an operational IVR system to an AI agent constitutes a complex and costly undertaking. A common user concern, which impedes the migration to AI agents, lies in the predictability and familiarity of the existing IVR system despite its deficiencies. This system is understood to a degree and continues to deliver a certain level of service. Conversely, although an AI agent is anticipated to enhance performance, the transition from one system to another introduces an element of uncertainty. This is particularly pertinent for large-scale organizations that cannot risk significant disruptions in client service continuity. Summary of the invention As embodied and broadly described herein the invention provides an Interactive Voice Response (IVR) system, comprising: a) a user interface configured to receive an incoming user inquiry; b) a response generation module configured to process the inquiry and generate a response; c) a Generative AI agent interface; d) wherein the response generation module is configured to direct at least partially the user inquiry through the Generative AI agent interface to a Generative AI agent for processing, and e) wherein the response generation module is responsive to an output elicited from the Generative AI agent in response to the user inquiry and presented via the Generative AI agent interface to generate the response 4  in response to the output and output the modified response via the user interface. In a specific and non-limiting example of implementation, the IVR system is voice based, that is to say it interacts with the user via voice. This is typical of most IVR systems where the user interaction is done via a telephone network, either a cellular network or wireline network. In those applications, the system is configured to process user spoken utterances and respond to the user using speech. In an exemplary implementation, the Interactive Voice Response (IVR) system is designed to interact with the user through voice communication. Typically, this involves utilizing a telephone network, which may be either a cellular network or a wireline network. In such applications, the system is configured to process spoken utterances from the user and provide responses using synthesized speech. This configuration ensures that users can engage with the system via voice, receiving real-time feedback and instructions through voice interaction. Alternatively, the Interactive Voice Response (IVR) system may be implemented as a text-based system, wherein user inquiries are received via text inputs, such as SMS messages or any other text-based messaging protocols. The IVR system is configured to process these text-based inquiries and generate responses using synthesized text outputs. In yet another embodiment, the IVR system comprises an omnichannel communication capability. This configuration enables the IVR system to interact with users through multiple communication channels, including voice 5 channels, text channels, or a combination thereof. The omnichannel capability allows for seamless transitions between voice and text interactions, thereby enhancing user experience and ensuring consistent service delivery across various communication modes. The inquiry processing module comprises logic designed to manage the user inquiry. In certain embodiments, this logic is referred to as legacy logic, which is engineered to formulate a response to the user inquiry autonomously, without any intervention from an AI agent. Traditionally, this legacy logic is predicated on predetermined rules, wherein a typical user interaction with the IVR system involves presenting the user with a predefined list of options to select from, where each selection may lead to further sub-options, and so forth. Essentially, the legacy logic is configured to generate responses independently, without the involvement of external entities, particularly those based on Generative AI. The inquiry processing module interfaces with the Generative AI system via the Generative AI interface, facilitating the transmission of the user inquiry or specific elements thereof to the Generative AI system, and conversely, the reception of the AI-generated response. Moreover, the inquiry processing module may include an AI integration module that determines the extent of interoperability between the legacy logic and the Generative AI system. In one embodiment, the AI integration module is configured such that the Generative AI agent intervenes in one or more instances to augment the operation of the legacy logic. For example, the AI integration module intercepts inbound events that would normally process by the legacy logic to enhance 6 decision-making and prevent incorrect actions. Specifically, in scenarios where the user articulates a request, the legacy logic, utilizing conventional speech recognition technology, may fail to map the utterance to a predefined option. In such instances, the Generative AI agent can more accurately perform this task and return a precise menu selection, thus ensuring the user interaction proceeds correctly. In this specific case, the inbound event is passed via the interface to the Generative AI agent, while the legacy logic is paused. The output of the Generative AI agent, once available, is passed to the legacy logic that can then process it to generate a response. Alternatively, instead of pausing the execution by the legacy logic, the legacy logic is allowed to continue the processing of the inbound event, while the Generative AI agent performs the same processing in parallel. The results of both processing threads can then be combined, compared or a selection can be made between them to select the one to present to the end-user. In one embodiment, the AI integration module is configured such that the Generative AI agent selectively intervenes during certain instances to augment the operation of the legacy logic. For instance, the AI integration module may intercept inbound events ordinarily processed by the legacy logic to enhance decision-making and prevent erroneous actions by the legacy logic. Specifically, in scenarios where the user articulates a request, the legacy logic, utilizing conventional speech recognition technology, may fail to accurately map the utterance to a predefined menu option in an option tree. In such instances, the Generative AI agent can more accurately perform this task and return a precise menu selection, thus ensuring the user interaction proceeds correctly downstream. 7 In this specific embodiment, the inbound event is routed via the interface to the Generative AI agent, while the execution by the legacy logic is temporarily suspended. The output from the Generative AI agent is then received and transmitted back to the legacy logic, which subsequently processes it to generate a response. Alternatively, execution by the legacy logic may proceed concurrently with the processing by the Generative AI agent. In this parallel processing scenario, the results from both the legacy logic and the Generative AI agent are collated, compared, or selected such that the most accurate and contextually appropriate response is presented to the end-user. In the first instance where the Generative AI agent is configured to intervene in a selected fashion and provide inputs in areas of the process which may be more challenging for the legacy logic to handle on its, the installation may be of permanent nature, in other words, the Generative AI agent remains as an assist while the legacy logic continues to handle the main processing responsibilities. From a practical perspective, the upgrade to the IVR system would involve a software upgrade to add the AI integration module with an interface to the Generative AI system that runs the LLM in the background. That upgrade can be performed relatively seamlessly with little risk of a major service disruption. In the second instance, where the Generative AI agent generates a full response to the inquiry, which requires access to end-user data and thus connection to enterprise databases, the installation is more complex and makes practical sense in instances where longer term it makes more sense to phase out the legacy logic in favor of the Generative AI agent. Accordingly, in this application, both the legacy logic and the Generative AI agent could be run for a time 8 suƯicient to confirm the full functionality of the Generative AI agent, at which time the legacy logic can be decommissioned. In the first embodiment wherein the Generative AI agent is configured to selectively intervene and provide inputs in areas where the legacy logic may encounter diƯiculties, the implementation involves a permanent installation. This approach designates the Generative AI agent as an auxiliary component, while the legacy logic retains primary processing responsibilities. The system upgrade entails integrating a software module that encompasses the AI integration component and an interface for the Generative AI system, which operates the Large Language Model (LLM) in the background. This upgrade can be executed with minimal risk of significant service disruption. In the second embodiment, where the Generative AI agent produces a comprehensive response to the user inquiry necessitating access to user data and enterprise databases, the implementation is more intricate. This method is practical when a long-term transition from legacy logic to the Generative AI agent is anticipated. In this scenario, both the legacy logic and the Generative AI agent would concurrently process user inquiries for a suƯicient period to validate the Generative AI agent's full functionality. Upon successful validation, the legacy logic can be phased out, thereby establishing the Generative AI agent as the primary processing entity. As embodied and broadly described herein, the invention further provides a process for upgrading a functionality of an IVR system which includes: a) a user interface configured to receive an incoming user inquiry; 9  b) a response generation module configured to process the inquiry and generate a response; c) wherein the process includes: a. interfacing the response generation module with a Generative AI agent; b. directing at least a portion of the inquiry to the Generative AI agent to elicit form the Generative AI agent an output; c. producing a response to the inquiry by the response generation module at least in part on the basis of the output. Brief description of the drawings Figure 1 illustrates a block diagram of an Interactive Voice Response (IVR) system enhanced with a Generative AI agent for functional augmentation. Figure 2 is a flowchart depicting the process executed by the IVR system, as illustrated in Figure 1, to handle an inquiry. The flowchart outlines the steps where the response generation module interfaces with a Generative AI agent, directs a portion of the inquiry to this agent, and subsequently produces a response based on the output received from the Generative AI agent. Description of a detailed example Figure 1 depicts a block diagram of an Interactive Voice Response (IVR) system interfaced with a Generative AI agent. In this configuration, the Generative AI agent is integrated as a permanent solution to augment the operational capabilities of the IVR system. The Generative AI agent is designed to provide input to the IVR system, thereby facilitating and enhancing its functionality. The legacy logic of the IVR system oversees the overall system management, while 10 the Generative AI agent supports decision-making processes at specified points within the workflow. The Interactive Voice Response (IVR) system, herein designated as reference numeral 12, is operatively connected to users, herein designated as reference numeral 10 (only one being depicted for simplicity), via a network, designated as reference numeral 32. The network 32 facilitates bi-directional communication between the IVR system 12 and the users 10. Typically, such communication is conducted digitally and can utilize various mediums, including but not limited to voice, text, and potentially video. Therefore, the communication is characterized as omnichannel, enabling the user 10 to formulate an inquiry received by the IVR system 12 via a single medium or a combination of multiple media and subsequently receive a response from the IVR system 12. In one exemplary embodiment, involving multi-media communication, the user 10 may initiate an interaction with the IVR system 12 through the cellular component of the network 32. Concurrently, the user 10 can engage with the IVR system 12 via a web interface over the Internet component of the network 32. This dual interaction allows the user 10 to utilize voice communication and, when preferable, input additional information through text via the web interface. Accordingly, the terminology "user inquiry" and "response" should be construed broadly to encompass communications that may involve a singular medium, such as voice, text, or video, as well as combinations of multiple media, such as voice and text, among various other permutations. 11 The IVR system 12 comprises an interface 14, which is connected to the network 32. This interface is responsible for receiving user inquiries and sending back responses. The structure of the interface 14 includes several components that facilitate communication between the users 10 and the IVR system 12. It typically contains telecommunication hardware, software for processing inquiries, and connectivity modules that enable interaction over various media such as voice, text, and video. The interface's function is to eƯiciently manage and route incoming inquiries to the appropriate modules within the IVR system, ensuring that responses generated by the Generative AI agent are delivered promptly back to the users. The interface 14 communicates with a response generation module 16, which is configured to receive user inquiries, process the inquiries, and generate responses. The response generation module 16 is software-implemented and operates on a computing platform. The computing platform comprises a data processor configured to execute computer-readable code. When executed, this code provides a range of functions suitable for the overall operation of the response generation module 16. The computing platform includes, but is not limited to, a central processing unit (CPU), memory storage units such as RAM and ROM and input / output interfaces among other components. The CPU is configured to process data and execute instructions stored in the memory units. The memory storage units are configured to store the computer-readable code, in a non-transitory fashion, and data necessary for executing various functions of the response 12 generation module 16. The input / output interfaces facilitate communication between the CPU and peripheral devices. The data processor within the computing platform is designed to handle high volumes of data and algorithms required for generating accurate and contextually appropriate responses. This processor operates in conjunction with the memory storage units to retrieve and store data dynamically, ensuring the eƯicient execution of the response generation module's functions. The response generation module 16 includes a manager module 18 that has the responsibility to provide an overall management of the operations of the response generation module 16. The manager module 18 is responsible for orchestrating and managing how the IVR system 16 formulates responses to user inquiries. In a specific example, its functions include: 1. Context management – receives user inquiries and establishes conversation sessions with individual users including tracking ongoing interactions. 2. Speech and text management - manages speech and text by converting voice to text via speech recognition and text to speech for user-friendly voice responses. 3. Integration with back-end services – retrieve necessary data from enterprise databases, such as database 30, Customer Management Systems (CRM) or Application Programming Interfaces (APIs) to access enterprise data which is necessary to provide accurate responses. Also, manages user authentication. 13  4. Legacy Logic and Generative AI agent process management – triggers the response generation process by utilizing both the legacy logic and the Generative AI agent. It ensures that each processing channel is provided with the requisite input data, such as the end-user inquiry or specific elements thereof, necessary for generating accurate and contextually relevant responses. 5. Error handling and fallback – detects incomplete, ambiguous or incorrect inputs and decides on clarification voice or text prompts to the user, triggers fallback responses or decides to escalate to a live agent. 6. Logging and Monitoring – Keeps logs of interactions for analytics, compliance and future improvements, including monitoring system performance. As indicated above, the manager module 18 controls the response generation process performed by the legacy logic module 20 and also by the Generative AI agent. The legacy logic module 20 is configured to generate a response to the user inquiry using traditional techniques, characterized by one or more of the following; 1. The legacy logic does not rely on AI, rather uses fixed rules to generate a response. 2. It employs structured database queries to extract data from the database 30 or associated data sources which the user needs, in contrast to vector based searches which are used with a Generative AI agent. 14  3. It is a fixed system, such that for a given input it will produce generally the same output. In a typical operational flow, the legacy logic module 20 is configured to receive the user inquiry, which, in the event of voice inputs, is converted into text format by the manager module 18. Additionally, the manager module 18 accommodates DTMF tones (keypad entries) or other input modalities. Subsequent to the receipt of the user inquiry, the legacy logic module 20 performs a matching operation against Interactive Voice Response (IVR) menus within a preset decision tree. If data retrieval is necessitated, the legacy logic module 20 initiates a request to the manager module 18 to perform a structured query on the database 30 or associated data sources, thereby extracting the requisite data. Upon successful data extraction, the legacy logic module 20 is further configured to formulate a response. This response may be derived from pre- recorded messages or dynamically generated text responses, contingent upon the context of the user inquiry. The formulated response is subsequently transmitted to the manager module 18, which facilitates the delivery of the response to the user. For text-based responses, the manager module 18 may convert the text response into a synthesized voice output, thereby ensuring seamless user interaction. The legacy logic module 20, as delineated, operates independently of AI methodologies, relying instead on established fixed rules to generate responses. This traditional approach contrasts with the functionalities of the 15 Generative AI agent, which employs vector-based searches and advanced AI techniques to generate contextually relevant responses. The Generative AI agent includes a Generative AI integration module 22 which is in eƯect the functional link between the IVR system 12 and a Generative AI service 26, which in most cases will be remotely located, such as it resides in the cloud and the IVR system 12 communicates with it through a data network 27, typically the Internet. The Generative AI integration module 22 communicates with the manager module 18 to receive user inquiries and prompt the Generative AI service 26 to elicit an output which is used to generate a response to the user. The Generative AI integration module 22 communicates with the Generative AI service 26 via a suitable interface 24. The Generative AI service 26 is preferably based on a transformer architecture and uses a Large Language Model (LLM) to generate outputs in response to inputs. The Generative AI agent is comprised of a Generative AI integration module (22) which functions as an operational interface between the Interactive Voice Response (IVR) system (12) and a remotely located Generative AI service (26), typically residing in the cloud and engaging in communication with the IVR system (12) through a data network (27), preferably the Internet. The Generative AI integration module (22) interfaces with the manager module (18) to receive user inquiries and subsequently prompt the Generative AI service (26) to generate an output, which is then utilized to construct a response for the user inquiry. 16 The Generative AI integration module (22) engages with the Generative AI service (26) via an appropriate interface (24) which forms part of the Generative AI agent. The Generative AI service (26) is preferably implemented using a transformer architecture and leverages a Large Language Model (LLM) to produce outputs in response to the received inputs. In one example of implementation, the Generative AI agent is aa permanent component of the IVR system 12, and it is used to provide focused assist to the legacy logic module 20. In this form of implementation, the legacy logic module 20 is configured to provide the Generative AI integration module 22 at predetermined point in the workflow with the system state and the associated data, such that the Generative AI agent can return a response which is then injected into the legacy logic workflow. This approach is helpful in instances where it is desired to improve the functionality and performance of the IVR system 12 without replacing entirely with a new version based on Generative AI. In other words, the legacy logic module 20 maintains overall control of the response generation process and invokes the Generative AI agent at certain points of that process, which may be predetermined or can be triggered when specific events occur. In one possible example, the Generative AI agent is invoked in order to more accurately identify a selection made by the user in a certain menu, in particular in instances where the user is conversing with the system via voice. In this instance, the legacy logic module sends to the Generative AI integration module 22 the converted voice to text data received from the user which conveys the user input along with a decision tree, namely 17 the list of options that are available to the user to select from. In response to this input, the Generative AI integration module 22 will construct a prompt, which includes the decision tree providing the list of options, the utterance of the user converted to text and the instruction to match the user text to one of the selectable options. The prompt is transmitted via the interface 24 and the network 28 to the Generative AI service 26, which returns a response which indicates what selectable option best matches the user utterance. That option is then conveyed to the legacy logic module 20 which uses this input to continue its workflow. In this embodiment, the Generative AI agent is thus permanently integrated and provides targeted assistance to the legacy logic module 20. In this embodiment, the legacy logic module 20 is configured to supply the Generative AI integration module 22 at predetermined points in the operational workflow with the system state and associated data, thereby enabling the Generative AI agent to generate a response that is subsequently integrated into the legacy logic workflow. This method is advantageous for enhancing the functionality and performance of the IVR system 12 without necessitating a complete replacement with a new version based solely on Generative AI technology. In essence, the legacy logic module 20 retains overall control of the response generation process and invokes the Generative AI agent at specific junctures within this process, which may be either predetermined or event-triggered. By way of example, the Generative AI agent may be invoked to more accurately interpret user selections from a given menu, particularly in instances where the user is interacting with the system via voice inputs. 18 In such an instance, the legacy logic module dispatches the converted voice- to-text data received from the user, which encapsulates the user's input, along with the decision tree—representing the list of options available to the user— to the Generative AI integration module 22. In response to this input, the Generative AI integration module 22 formulates a prompt, incorporating the list of options, the user's text-converted utterance, and the instruction to match this text to one of the selectable options. This prompt is transmitted via the interface 24 and the network 28 to the Generative AI service 26, which generates a response indicating the selectable option that best matches the user's utterance. The identified option is then conveyed to the legacy logic module 20, which utilizes this input to proceed with its operational workflow. In another embodiment, the Generative AI agent is interfaced with the legacy logic in a transitional process aimed at phasing out the legacy logic and permanently replacing it with the Generative AI agent. During this decommissioning process, the legacy logic module 20 and the Generative AI agent operate in a parallel configuration. The legacy logic module 20 functions as a fallback mechanism, ensuring system resilience and preventing total operational interruption in the event of a Generative AI agent failure. Additionally, throughout the decommissioning phase, the legacy logic module 20 serves as a benchmark for evaluating and calibrating the outputs generated by the Generative AI agent, thereby facilitating fine-tuning of the Generative AI responses. 19 It should be appreciated that while the legacy logic module 20 has usability issues for users, the responses generated by the legacy logic module 20 are fact-based, which presents a significant advantage. For instance, if a user requests the dollar amount in an account, the response is generated by formulating a database query executed on the data within the database 30. Thus, the output is predictable and factually accurate. In contrast, the responses generated by the Generative AI agent are not necessarily fact-based. Large Language Models (LLMs) are statistical models of knowledge bases rather than actual repositories of factual data. As such, there is no assurance that input data, such as data extracted from the database 30 and provided to the LLM for response generation, will consistently form the basis of the LLM's response, particularly if the input data conflicts with the LLM's internal data. Consequently, while the LLM can enhance user interaction by making it more user-friendly and seamless, a potential drawback is that it may not generate responses grounded in factual accuracy to the same extent as the legacy logic module 20. This discrepancy could lead to system performance issues if an immediate transition to the Generative AI agent were attempted. Therefore, the ability to progressively phase in the Generative AI agent and evaluate it against a source of accurate knowledge, namely the legacy logic module 20, provides a distinct advantage. This phased integration allows the Generative AI agent to be introduced in a controlled manner while maintaining a baseline of system performance. This embodiment is illustrated by the flowchart at Figure 2, which depicts the main steps of the process. 20 The method starts at step 36 wherein the IVR system 12 is initialized under the control of the manager module 18. At step 38, a user inquiry is received at the interface 14 and communicated to the manager module 18, which performs preliminary processing of the user inquiry. This preliminary processing may include converting the user's speech input into text format via speech recognition technology when the user interacts with the IVR system 12 through voice commands. Subsequently, the manager module 18 transmits the text- converted user inquiry to two parallel processing channels: one utilizing the legacy logic module 20 and the other employing the Generative AI agent. In the initial phase, both processing channels try to ascertain the user's intent. Within the legacy logic channel, this is accomplished by parsing the user input as depicted at step 40. Specifically, the text string is segmented into individual components and analyzed based on predefined rules, such as based on keywords to determine the user's intent, subsequently mapping the identified intent to a specific option within a menu of possible action choices. At step 44, the legacy logic constructs the requisite data queries based on the selected option. At step 48 the pipeline is orchestrated, which includes extracting from the database 30 the necessary data items on which the queries will be run. Concurrently, the Generative AI agent processes the user inquiry through its own distinct channel. This process includes analyzing the query at step 42 to identify a workflow to be performed in response to the user's inquiry. The identification of the workflow may necessitate submitting the user inquiry to the Generative AI service, which returns a workflow identification to be 21 performed to generate a response. At step 46, the Generative AI integration module 22 constructs a prompt in accordance with the workflow requirements, and at step 49 composes the pipeline. This pipeline includes obtaining necessary data elements from the database 30 or other data sources, embedding those data elements, or creating a link to the data elements in the prompts to allow the Generative AI service to access them. At steps 50, 54, and 58, the legacy logic channel runs the structured query on the data, parses the results, and prepares them. In parallel, at step 52, the Generative AI channel submits the prompts to the Generative AI service 26, receives the results, re-ranks the results at step 56, and at step 60 provides a summary. The outputs from both channels are processed by the manager module 18, which constructs a consolidated response based on both inputs. Typically, the responses will overlap; however, some variations may occur. One approach is for the manager module 18 to merge the responses while eliminating duplications. In such cases, the response includes the traditional legacy logic output, enhanced by the output of the Generative AI channel. The manager module 18 may conduct this processing based on logic rules, though another option is to submit both responses to the Generative AI service 26 with instructions to merge them, compare them, summarize them, or otherwise process them to generate the desired output. In this instance, both responses are sent to the Generative AI integration module, which constructs a prompt that embeds both responses along with the necessary instructions to combine the responses in a semantically meaningful manner, while also comparing 22 them to flag any significant discrepancies to the manager module 18, which will log the discrepancy as it indicates a system fault. At step 64, the results are communicated to the user by the manager module 18, which may involve converting the text to speech format. At step 66, the system gathers feedback based on the delivered results and updates the conversation status at step 68. In the event that the conversation is determined to be complete at decision step 70, the final feedback, if available, is collected from the user at step 74, and the process is concluded at step 76. Conversely, if decision step 70 yields a negative outcome and the user issues further requests, the process reverts to step 78, initiating an additional iteration. As previously mentioned, the manager module 18 maintains a log of system performance. If this log indicates that there are an excessive number of errors originating from the Generative AI agent, steps may be taken to fine-tune the LLM model for improved accuracy, or alternatively, a diƯerent model may be implemented. This iterative process is repeated as many times as necessary until the error rate is reduced to an acceptable level. Once this objective is achieved, the legacy logic module 20 can be decommissioned, allowing only the Generative AI agent processing channel to produce responses. 23

Claims

Claims:

1. An Interactive Voice Response (IVR) system, comprising: a. a user interface configured to receive an incoming user inquiry; b. a response generation module configured to process the inquiry and generate a response; c. a Generative AI agent interface; d. wherein the response generation module is configured to direct at least partially the user inquiry through the Generative AI agent interface to a Generative AI agent for processing, and e. wherein the response generation module is responsive to an output elicited from the Generative AI agent in response to the user inquiry and presented via the Generative AI agent interface to generate the response in response to the output and output the modified response via the user interface.

2. A process for upgrading a functionality of an IVR system which includes: a. a user interface configured to receive an incoming user inquiry; b. a response generation module configured to process the inquiry and generate a response; i. wherein the process includes: ii. interfacing the response generation module with a Generative AI agent; iii. directing at least a portion of the inquiry to the Generative AI agent to elicit form the Generative AI agent an output; 24 iv. producing a response to the inquiry by the response generation module at least in part on the basis of the output. 25