Frequently Asked Questions

This FAQ will provide you with a general understanding of the technologies used to build the Finn AI offering.

General Technology

What is Artificial Intelligence?

At the highest level, AI is the theory and development of software systems that mimic human intelligence. Unlike rules-based expert systems, AI systems do not rely on program logic explicitly set by humans. Instead, AI leverages existing data to construct its own internal mathematical algorithm to find unseen patterns, categorize information, and offer predictions. This makes AI suitable for tasks with vague or  hard-to-specify rules, such as visual perception, natural language understanding, and decision-making.

Finn AI specializes in conversational AI, using AI technology to understand what bank customers say and what they want to achieve, while recognizing information such as the name of bank products, accounts, or other details. With sophisticated and proprietary Finn AI technology and deep banking domain expertise, Finn AI allows banks to communicate with their customers using natural conversation.

What is NLP and NLU?

Natural Language Processing (NLP) is a branch of AI that enables computer systems to process and analyze natural human language by converting unstructured language utterances into discrete data.

Natural Language Understanding (NLU) is the science of comprehending the meaning of the sentences. This allows the system to translate the utterance of a bank customer into commands for the computer system.

Since Finn AI focuses solely on banking, the NLU models are pre-trained (in multiple languages) for banking conversations. This means Finn AI is capable of understanding most banking terms, processes, and product categories out of the box, for a faster time to market.

Because natural language can extend beyond the banking use case, Finn AI is also trained to support a range of small talk queries and other general topics that could occur with a customer during a conversation.

What is ML and DL?

Machine Learning (ML) is the science of constructing algorithms so that computers can learn and improve from  every experience without being explicitly programmed to do so.

Deep learning is currently one of the most advanced forms of machine learning techniques. It uses  artificial neural networks to understand very complex unstructured concepts, such as natural language processing.

Finn AI uses state-of-the-art deep learning technology to train AI models so they can accurately understand and respond to  banking conversations. These models are continuously refined and optimized through real world use – even after the virtual assistant has been deployed – for ongoing performance optimization.

What is conversational design?

Conversational design is more dynamic and interactive than traditional media and GUI-based systems. With conversational design, the model is open-ended and the user may be more emotionally involved.

The user experience extends across a range of categories that can include the structure of responses, the tone of the virtual assistant, the length of the conversation, and the complexity of the workflow.

It is usually the bank that determines how the virtual assistant will behave and interact with customers. To help support this work, Finn AI provides access to a dedicated conversational design expert to guide each bank through this process.

Hosting & Integration

How does Finn AI host its operations?

Finn AI operates on a subscription-based SaaS model using modern cloud infrastructure to develop, deploy, operate, and upgrade each virtual assistant across different global regions. There is no need for the bank to deploy anything, requiring fewer resources than a self-hosted model and lower costs overall.

If you would like additional security and cloud information for Finn AI, contact [email protected].

How does the Finn AI technology architecture work?

A detailed Finn AI Technical Architecture whitepaper is available upon request. Contact [email protected] or your Finn AI sales person for more information.

What is the Finn AI SDK model?

The Finn AI SDK allows banks and third-party vendors to develop their own virtual assistant user interfaces. It provides the system developer with:

  • Secured access between the end user device and the Finn AI backend
  • Secured access between the Finn AI backend and the bank’s data center
  • Full range of UI widgets and user interactions
  • Consistent handling of authenticated and unauthenticated interactions
  • A sample application to help the developer get started
  • Technical support

Finn AI SDKs are available for the following applications:

  • Native iOS
  • Native Android
  • Web

How does Finn AI integrate into the backend system of a bank?

Finn AI uses APIs to integrate seamlessly into a bank’s backend system. The recommended integration is through REST Api with OAuth2.

Other methods for integration include:

  • Direct REST APIs banks can expose on the internet, with authentication and/or IP whitelisting
  • Proprietary agents, including cloud-to-ground agents to deploy into the Finn AI infrastructure
  • The bank’s own SDKs

Which platforms/social channels does Finn AI support?

Finn AI can be easily deployed on the bank’s existing digital channels, including:

  • Website
  • Mobile banking apps (Android and iOS)

The technology can also be deployed on most social media platforms including WhatsApp, Facebook Messenger, and others.

For banks that use support agent systems (LivePerson LiveEngage, BoldChat, Sprinklr, etc.), Finn AI can be used in concert with these systems to answer end user questions more cost effectively, including seamless transfers between the virtual assistant and a human agent, when needed.

Deployment & Support

What is involved in implementing Finn AI?

Initial set-up of a virtual assistant with out-of-the-box features, languages, and a trained AI model requires little to no effort from the bank.

A Finn AI implementation fits the typical UAT production cycle and can be ready for review within a week. More time may be required depending on additional features and customizations requested by the bank.

Once implemented, the bank will collaborate with Finn AI to design and customize content, train the data, work on APIs, and add unique information about products and services offered by the bank. A standard start-to-first-phase production cycle can be as short as three months.

Once the production system goes live, the model continues to evolve and improve through ongoing usage data and subscription upgrades that will optimize performance.

Is Finn AI available in other languages?

Yes, supported languages include: English, French, Spanish, German. Additional languages can be easily added.

Finn AI can also accommodate subtle differences that occur regionally and from customer to customer. Native speakers and local testers are used to ensure the customer experience reflects the vernacular and unique requirements of each region.

Does the bank need to build the AI model from scratch?

No. The Finn AI banking-specific model already understands hundreds of banking terms and questions. This is because Finn AI aggregates banking utterances from all customer deployments around the world. These anonymized utterances from bank customers reflect real banking conversations and are used to continually train and optimize the Finn AI model. Banks automatically get the most current model ready to deploy, saving them significant time (months) and resources which would be required if they were to develop their own model on a generic platform. The Finn AI model provides tested performance and real-world usage out of the box.

Does the Finn AI NLP model improve after deployment?

Optimization is ongoing and never ending. It starts the moment we deploy the first testing environment. It continues to improve as real world utterances, content, and data is generated from usage. Throughout the customer subscription period, the Finn AI team will analyze the customer data, perform additional annotation, retrain the model, and evaluate model performance.

Additional information on Finn AI architecture and processes is available upon request. Contact [email protected] or your Finn AI sales person for more information.

Can the virtual assistant incorporate the bank brand?

Finn AI is a white-label technology that is fully customizable to support the brand and visual requirements of each bank. Additional customization can be achieved by using the SDKs to develop apps.

Security & Privacy

How does Finn AI handle PII data?

Personally Identifiable Information (PII) is redacted before the data is stored. Finn AI only uses this anonymized data for annotation and training. The Finn AI model is not trained to identify individual users.

How does Finn AI access personal financial data?

Finn AI does not store or use personal financial data received from the bank for any purpose other than answering an end user’s question. Once the question has been answered, the data is immediately discarded.

Here’s how it works:

  • An end user asks the virtual assistant a question about their personal finance (“what is my balance?”)
  • Finn AI makes a call to the bank’s backend system, with authentication, to retrieve data to answer the end user (access to the data must be permitted by the bank’s APIs)
  • Once the answer is provided to the end user, the data is discarded from the system

Does Finn AI know the identity of the user?

Finn AI does not require user identity for general FAQs, product information, and any conversations that a bank agent would have with an unauthenticated user.

For authenticated or tailored exchanges, such as account balances or money movement, the bank must issue a temporary token to Finn AI to enable the use of the bank’s APIs, following standard security practices. The bank maintains full authorization control throughout the entire process.

Finn AI does not need the identification or password of the user. Only the bank will know which user maps to the temporary token, and is able to revoke a token at any time.

Additional authentication implementation information is available upon request. Contact [email protected] or your Finn AI sales person for more information.

How does Finn AI store personal financial data?

Finn AI receives personal financial data from the bank in order to provide the end user with a response to a question that they’ve asked. For instance, when a user asks “What is my balance?”, Finn AI will retrieve the information from the bank with the user’s explicit permission. Finn AI does not keep a record of the user’s data and discards it once the response has been delivered to the end user.

The only data that may be partially retained is the text log within conversations. This is used for additional model  training and is a common AI practice. The training is focused on natural language understanding and uses no financial-related fragments within the textual data. It is also anonymized and cannot be attributed to any individual.

Does Finn AI share data with other customers?

No, banks do not have access to other bank’s data. Only authorized Finn AI personnel can access the pool of anonymized data used for annotation and training. The training model provides common improvements in understanding user utterances for general banking. Bank-specific information (product information, special terms, customized workflows and responses, etc.), is not shared and only exist within each bank’s virtual assistant.