Enterprise AI Agents for Regulated Industries

What is OpenDialog?

OpenDialog is an enterprise AI agent platform developed by OpenDialog AI Limited; it currently focuses on serving regulated industries with high levels of interaction, such as the insurance sector. It is used for designing, deploying, and managing trackable customer conversations by combining large language models, structured business processes, enterprise systems, and human services.

In the database, “Enterprise AI Agents for Regulated Industries” is a descriptive title that was used for this product and still reflects its brand; the actual name of the product is OpenDialog. The current focus of this product lies on specific areas such as insurance quotes, policy services, re-engagement, and subsequent claims processing.

Current positioning and product boundaries

OpenDialog is not a general chatbot with only Q&A capabilities, nor is it a standalone basic model for sale. It functions more as an interactive layer within the insurance customer journey, organizing the behavior of intelligent agents through controlled scenarios, knowledge retrieval, business actions, human intervention, and audit records.

• Targeted at insurance companies, brokers, agencies, and service operation teams.
• It is important to move concept validation forward into actual customer journeys, rather than staying at the stage of demonstrations.
• It supports predefined insurance agents, and it is also possible to customize them for other business processes.
• It does not replace policy documents, quotes, CRM systems, or core claims processing systems; instead, it interacts with them through actions and APIs.
• The model’s output may still contain errors, and the responsibility for compliance does not shift to the platform as a result of using an agent.

Current agents and their online status

Michaela and Clark are clearly indicated as features that will be released soon, and they cannot be considered as functions that all customers can purchase at present. Even for the agents that have been demonstrated, the available channels, languages, system connections, and capabilities for automatic execution depend on the scope of the project.

Selma offers intelligent sales agents

Selma is designed for the insurance quotation and purchase process; it provides product information around the clock, collects the data needed for quotations, and offers personalized yet controlled support based on real-time customer behavior. Its goal is to prevent customers from abandoning their process due to complicated forms or unclear information.

1. Customers can request quotes through websites, chats, emails, or other designated channels.
2. The agent identifies the stages of the journey and the customer’s objectives, and poses questions in accordance with the approval process.
3. The necessary data is collected through structured messages or forms and transmitted to the quoting or business system.
4. The system explains the guarantees, conditions, and next steps, but must not exceed the established compliance boundaries.
5. When encountering complex, sensitive, or low-confidence issues, forward them to authorized human personnel.
6. The team evaluates performance by examining conversation trajectories and conversion data, and continues to optimize it.

The increase in conversion rates resulting from product displays is a consequence of specific implementations or brand presentations, and it is not a guarantee for every company. The actual benefits depend on factors such as product prices, the quality of traffic, underwriting rules, channel design, and the responsiveness of the customer service team.

Jamie Policy Inquiry Bot

Jamie uses customer policies and approved knowledge to answer specific questions, and it can handle requests via web chat, email, voice, and other channels; it forwards those requests that cannot be handled securely to human agents. It is suitable for service scenarios involving a high volume of requests whose answers can be found in policies and knowledge materials.

• Explain the coverage, deductible, limitations, dates, and general policy process.
• Generate more relevant responses by taking into account the current customer and policy context.
• Identify complex, complaint-related, or high-risk requests that are not suitable for automatic response.
• Submit the entire conversation context to a human agent, thereby reducing the need for customers to repeat their statements.
• Log retrieval, response, and decision-making steps to facilitate quality review and auditing.

OpenDialog has previously disclosed data on the proportion of cases handled autonomously and on the reduction in associated costs, but the specific figures come from the relevant projects. Insurance companies should measure the rate of cases resolved outright, the rate of incorrect responses, the rate of cases forwarded to human agents, the complaint rate, and any compliance issues, rather than focusing solely on the proportion of cases handled automatically.

Rhea contacts the agent again.

Rhea is used to get in touch again with customers who started but did not complete a quotation process, allowing the selection of the appropriate timing and content for the communication based on the status of that process. Its value lies in ensuring that the follow-up efforts are consistent with previous interactions with the customer, rather than sending identical mass notifications.

• Identify the breakpoints as well as the collected quote information.
• Determine whether it is appropriate to contact again based on business rules.
• Provide next-step guidance for unresolved questions or missing steps.
• Allow the customer to continue with their original journey or be connected to a human sales representative.
• Record consent, cancellation, and contact results to avoid unnecessary repeated outreach.

Reaching out to customers again is related to rules on electronic marketing, privacy, and customer consent. Companies must determine the legitimate basis for contacting customers based on factors such as region, channel, and customer relationship, and they must also implement mechanisms for allowing customers to opt out, express objections, and limit the frequency of such contacts.

SAFER: Safety and Compliance Assessment

OpenDialog evaluates and monitors agents using the SAFER AI Benchmark, and is designed for regulated industries such as insurance. It aims to take security, retrieval accuracy, compliance rules, and knowledge boundaries into account during both pre-deployment testing and ongoing operation.

This framework is said to support automatic monitoring, continuous improvement, and alignment with the regulatory requirements set by the UK’s Financial Conduct Authority; however, it does not equate to certification by regulatory authorities. Companies still need to carry out independent compliance checks using their own rules regarding products, consumer responsibilities, record-keeping, and complaint handling.

Code-free agent design

OpenDialog offers a code-free interface that enables business and dialogue designers to organize scenarios, themes, conversations, nodes, intents, messages, and conditions. The technical team can focus on system connectivity and security aspects, while business professionals are responsible for maintaining the language and rules used in these interactions.

• Define a complete application or customer journey using scenarios.
• Use dialogues and nodes to break down the objectives and transition conditions for each stage.
• Save customer inputs and values returned by business systems using structured attributes.
• Conditions and filters are used to determine which messages, actions, or paths can be executed.
• Release after testing in the preview environment, then make improvements based on the analysis records.

The absence of a code-based approach does not mean that no engineering effort is required. Authentication, handling of sensitive data, payment processes, writes to core systems, error recovery, and handling of high-concurrency scenarios still require the involvement of developers, security specialists, and operations staff.

Structured dialogue and message components

The platform not only outputs text, but also supports various message types such as buttons, lists, date selection, forms, file uploads, automatic address completion, location data, electronic signatures, progress indicators, and full pages. Structured components help to reduce ambiguity in free-text input and convert it into fields that can be processed by business systems.

• The quotation for a journey can be generated by using forms and date components to collect the necessary data.
• The buttons and lists available for policy services limit the possible actions.
• Data submission can be done by uploading files to proceed with the subsequent verification steps.
• For steps that require formal approval, a confirmation process or electronic signing procedure can be designed.
• When chatting cannot resolve the issue, a message for manual transfer can be displayed while preserving the context.

When using payment, signing, and document handling components, it is necessary to clarify the boundaries of responsibility between those components and the actual third-party systems. The status indicated in the platform cannot serve as a substitute for the final confirmation provided by the payment service providers, signing services, or the core business systems.

RAG, Models, and Language Processing

OpenDialog enables the vectorization of corporate documents and the generation of responses through retrieval enhancements; it can also make use of semantic classifiers to determine the intent behind a query. The document lists language services such as OpenAI, Azure OpenAI, and Google Gemini, while also preserving access to various traditional NLU interpreters.

1. Choose insurance policies and service materials that are approved and have clear versions specified.
2. Organize knowledge topics by product, region, customer type, or expiration date.
3. Configure embedding, retrieval, and language models to restrict the materials accessible to the agent.
4. Use real questions and adversarial questions to test recall, refusal to answer, and escalation to a human agent.
5. After going live, check the search records, incorrect answers, and any changes to the materials.
6. The index is cleared simultaneously when older policy versions or expired rules are removed.

The model providers and configurations affect the data pathways, costs, latency, and performance of the outputs. Companies should determine, during the ordering and technical design phase, which models are available, in which locations they will be hosted, how long the data will be retained, and whether the providers are allowed to train the models.

Third-party actions and manual takeover

Agents can send or retrieve data from external systems through actions, save the input and resulting outputs as attributes, and continue the conversation. Common uses include requesting quotes, creating service tasks, sending emails, using date and time tools, or connecting with human customer service representatives.

• Webhooks can be connected to internal APIs, decision engines, and business processes.
• The action library includes example capabilities such as email, Freshdesk, Chatwoot, etc.
• Dialog takeover is used to route high-risk or complex requests to human agents.
• The execution log helps in troubleshooting external calls, conditions, and return fields.
• Secret properties can be used to store values that should not be exposed directly in the conversation.

When allowing agents to write to the core system, it is necessary to employ principles such as minimum permissions, idempotency, approval processes, timeouts, retry mechanisms, and rollback capabilities. High-impact actions such as refunds, cancellation of policies, changes to beneficiaries, or decisions regarding claims should not be left to the generation models to decide on their own.

Distribution channels and interface

OpenDialog primarily builds applications using web workspaces, and it is possible to embed WebChat within websites or browser applications. The documentation also describes various communication channels such as web chat, email, and voice; however, the actual channels used by each insurance agent need to be determined based on the specific project.

API and WebChat SDK

The platform offers a Chat API as well as Public APIs designed for use in workspaces. The Chat API enables reading configuration and history data as well as sending messages, while the Public APIs allow for managing knowledge services, exporting user interactions and user data, and generating WebChat deep links with predefined context.

• Public APIs use Bearer Tokens, which inherit the permissions of the creator and have a specified validity period.
• The knowledge service API enables the management of topics, text, files, and web page content.
• The user interaction API allows for the retrieval of conversations within a specific time range, making it suitable for analysis and compliance-related exports.
• The WebChat SDK can be embedded in websites or browser applications to monitor events and custom components.
• Some endpoints may have rate limits; specific details can be found in the response headers or in the account settings.

Version 3 updates the Chat API endpoints and requires user, scenario, and authorization information; the old endpoints are no longer supported. Custom frontends must be developed in accordance with the current upgrade guidelines, while projects that use the official embeddable components or SDKs are less affected by this migration.

API integration process

1. Create dedicated users with minimal permissions in the workspace, and avoid reusing the administrator’s personal account.
2. Create time-limited API tokens from the Identity and Security page.
3. Distinguish the permissions for chat calls, knowledge management, interactive export, and user data export.
4. Verify the user ID, scenario ID, authorization headers, and error responses in the testing environment.
5. Set up secure storage, rotation, revocation, and alerts for abnormal access for tokens.
6. Timeout, idempotency, auditing, and manual upgrade rules are added for external actions.
7. Complete data protection, stress, failure, and privilege escalation testing before production release.

Monitoring, analysis, and auditing

OpenDialog logs the interactions and decision points within a dialogue, allowing one to see how the agent identifies intents, selects paths, carries out actions, and generates responses. This capability helps in identifying errors, optimizing the workflow, and preparing materials for internal reviews.

• Track where users are and what questions they ask.
• Check the execution records of language services, LLM actions, and external actions.
• Compare the manual transfer, completion, exit, and failure nodes.
• Export user interactions and conversation history for a specified time period.
• Identify issues with conditions or messages through scenario previews and visual positioning.

Being auditable does not mean that logs automatically meet all regulatory documentation requirements. Companies should verify whether the integrity of the logs, access controls, time zones, immutability, export formats, and retention period satisfy their own obligations.

Deployment and usage process

1. Schedule a demonstration by submitting the target insurance journey, channels, systems, and regulatory scope.
2. Decide whether to use a pre-defined agent or create a new one, and define the success metrics.
3. Outline the customer processes, manual tasks, business rules, exceptions, and automated actions that are prohibited.
4. Connect the approved knowledge materials, models, and testing systems to first create constrained scenarios.
5. Test using normal, boundary, malicious, complaint, and unanswered questions.
6. Configure settings for switching to manual handling, auditing, data retention, alerts, and shutdown.
7. Start with a low volume of traffic, and continuously review the accuracy rate, compliance issues, and business outcomes.

Price and procurement methods

There is currently no available page with public pricing information; the products are mainly available through scheduled demonstrations or corporate orders. The designation of “free plus premium features” in the database cannot be used as a basis for any current free offering. No permanent free version, public trial quota, or uniform price per session was identified in this review.

The order will specify the license period, the number of sessions allowed, the level of support provided, and the costs involved; it may also include professional services. The terms of service state that the costs are non-cancellable and non-refundable, and they do not include any applicable taxes. Before making a purchase, it is necessary to check whether there are additional charges for implementation, model usage, channels, additional sessions, and support.

• The agreement is automatically renewed after the initial period, unless either party provides written notice two months in advance.
• A material breach may be terminated if it is not remedied within thirty days following the notification.
• Companies should include pilot acceptance, go-live of production, as well as capacity and service levels in the order.
• Exceptions regarding refunds, data export, migration assistance, and early termination fees require separate confirmation.

Privacy and data retention

According to the terms of service, customer interaction data and conversation histories are retained for 90 consecutive days in each OpenDialog instance, unless otherwise specified in the order. Once this period has elapsed, the oldest daily dataset is removed on a daily basis.

• Website analytics data may include IP address, time of visit, country, device, browser, and pages viewed.
• Website analytics data is stated to be stored indefinitely for the purpose of monitoring usage trends.
• Contact forms and lead data can be used for communication, customer relationship management, and marketing.
• The privacy policy lists processing services such as Framer, Pipedrive, Google Workspace, and Slack.
• Users can request access, correction, deletion, and portability, with a commitment to respond within 30 days.
• The form uses SSL, along with access restrictions for trained employees, but this does not provide absolute security.

The website’s privacy policy explains information related to the company’s website, marketing activities, and customer relationship data; it does not provide a complete list of all the subcontractors involved in the creation of intelligent agents, nor of the model providers, locations where data is stored, or the schedules for data backup and deletion. Insurance companies should obtain separate data processing agreements and security attachments.

Safety and compliance considerations

• Set roles, minimum permissions, strong authentication, and procedures for revoking access upon departure for workspaces.
• API tokens have the same permissions as the creator, and they should not be exposed in browser code; they need to be rotated regularly.
• The knowledge base should have its versions controlled by insurance product, region, and expiration date.
• Implement stricter data minimization for personal health, financial, identity, and claims information.
• Ongoing testing is conducted for model prompt injection, unauthorized data retrieval, data leakage, and inappropriate recommendations.
• Human oversight is retained for decisions with a high impact on areas such as denial of coverage, pricing, claims processing, and complaints.
• It should be noted that FCA alignment statements, SAFER results, and customer cases are not equivalent to regulatory certifications.

Copyright, Commercial Use, and Licensing

OpenDialog retains the rights to its core software and user manuals; customers obtain the right to use these services in accordance with the order placed. Costs, the number of sessions, support, and other licensing details are all determined by the order.

When using policy documents, brand materials, customer data, and third-party content in the knowledge base, customers must have the appropriate permissions for processing and commercial use. Answers, scripts, and dialogue flows still require verification with regard to facts, copyrights, trademarks, and regulatory requirements.

GitHub and the open-source status

When assessing the open-source status, it is necessary to examine each repository individually to check its owner, archival status, and license. The presence of repositories in an official organization only indicates that the corresponding code or materials are available publicly; it does not mean that the entire commercial platform can be self-hosted, modified, or redistributed.

Which users are it suitable for

• Insurance companies and brokers that need to handle high-frequency quotes and policy issues automatically.
• I hope to recover the unfinished quotes, while also managing a sales team that adheres to marketing regulations.
• Insurance digitalization teams that need to combine generative AI with deterministic business processes.
• Customer service organizations that place emphasis on conversation patterns, manual intervention, and regulatory oversight.
• Technical teams that need to customize web chat, APIs, and actions in enterprise systems.
• Regulated companies are prepared to gradually introduce production agents through pilot projects on a small scale.

Advantages and limitations

Main advantages

• The current focus is on the actual insurance process, rather than merely providing a generic agent builder.
• The predefined agents cover sales, inquiries, and re-engagement, and it is possible to customize additional journeys.
• The combination of structured dialogue and generative responses facilitates the control of high-risk steps.
• SAFER takes into account malicious requests, query scope, retrieval accuracy, compliance, and knowledge boundaries in its evaluations.
• Public APIs, WebChat SDK, actions, and rich messaging components enable deep integration.
• Interaction and decision pathways facilitate quality optimization and audit preparation.

Main limitations

• The public pricing page is no longer available; costs, trials, and quotas require inquiry.
• The two subsequent agents are still in the process of being released.
• The business performance figures are based on specific deployments and cannot be regarded as a guarantee of general results.
• The website’s privacy policy provides limited information regarding the data flow in production services and the subcontractors involved in their processing.
• The default retention period of 90 days for interactions may not meet all requirements regarding insurance records.
• The core platform is not an open-source product that has been officially recognized; public repositories mainly contain documents, demonstrations, or forks.

Summary

OpenDialog is suitable for companies that need to integrate AI into real-world processes such as insurance quoting, policy administration, and customer re-engagement, while still maintaining structured control mechanisms, the ability for human intervention, and audit capabilities. Its advantages lie in its specialized insurance-related intelligent agents and SAFER assessment tools, rather than simply in the performance of basic modeling functions.

Before official deployment, it is necessary to verify accuracy, compliance, the process of transferring requests to human agents, system responses, and business-related metrics through demonstrations and pilot tests with limited traffic. Price, conversation limits, models, hosting services, data processing, and responsibilities related to renewal must all be specified in the order; free or premium tags from databases cannot be used in this context.

Frequently Asked Questions

What is “Enterprise AI Agents for Regulated Industries”?

It points to OpenDialog. OpenDialog is an enterprise AI agent platform designed for industries that are subject to regulatory requirements; its current focus areas include insurance quotes, policy services, and follow-up outreach processes.

Are OpenDialog and New Dialogue AI the same product?

No. The two use different brands, companies, and product systems; the database records point to OpenDialog, so it is not possible to include information regarding the features or prices of New Dialogue AI in this entry.

Which insurance bots are already available for viewing?

Currently, Selma is available for quote sales, Jamie for policy inquiries, and Rhea for smart re-engagement. Michaela for policy changes and Clark for claims processing are marked as upcoming.

Can OpenDialog decide completely automatically whether to approve a claim or provide coverage?

It should not be understood in that way. The platform can organize conversations, retrieve knowledge, and invoke systems, but decisions related to underwriting, pricing, claims processing, and complaints that have a significant impact should still be governed by clear rules and involve human oversight.

Is SAFER a regulatory certification?

It is not a recognized regulatory certification. It is OpenDialog’s security and compliance assessment framework, which can assist with testing and monitoring, but companies still need to meet their own regulatory, risk, and audit requirements.

Is there a public version or a free version of OpenDialog?

The current page showing public prices is not available, and there is no information available regarding a permanently free version or a standard trial quota. The cost for this service is determined based on quotes, taking into account the number of sessions used, and it can be acquired through demonstrations or corporate orders.

Does OpenDialog provide APIs?

Chat API and Public APIs are provided, enabling functions such as chatting, customizing the user interface, managing knowledge services, exporting interactions, exporting user data, and generating deep links. These APIs require a token as well as the request information specified by the current version.

Is OpenDialog open-source software?

The core product is not software that is already open source. The official GitHub repository includes public documentation, demonstrations, and third-party fork repositories, but this does not mean that the entire commercial platform has its source code made available publicly.

How long are conversation data stored?

According to the terms of service, the default retention period for interaction data and conversation history is 90 days in a row, unless otherwise specified in the order. If a longer period of storage is required, this must be clearly stated in the contract.

Can the contract be canceled at any time and a refund obtained?

The terms state that the agreement is automatically renewed, and written notice two months in advance is usually required; the listed fees are non-cancellable and non-refundable. The specific initial duration and any exceptions shall be confirmed in the order.