Backbase Conversational Banking

Agents interpreting chat and voice requests are the product, answering from live account context, executing payments and card actions, and carrying disputes through investigation. Kasisto's financial services intelligence, now Backbase's, sits underneath them.

Permissions and policy checks apply before any agent action executes, payments and card actions carry customer confirmation, and exceptions route to an authorized employee with the case context intact. Thresholds that turn an action into an exception are left to each bank's configuration.

Every answer is checked before it reaches the customer and grounded in account data and the bank's approved product information, which gives a validator something to test. Fabrication rates, containment quality and escalation precision are unpublished, and no model risk framework is mapped.

More than 120 banks on the Banking OS and I&M Bank's tenfold onboarding gain belong to digital banking rather than the conversational product. North American customers of Kasisto are cited without one named institution or a containment or resolution figure.

Frontline outcomes feed a continuous learning loop that refines the banking models, and whether outcomes at one bank inform the models serving another is not stated. Banks on the same platform compete for the same customers.

Balances, cards, payments and disputes run through these conversations, yet the product pages offer no data protection terms, subprocessor list or retention period for conversation records. A bank answers to its own customers for how long those records persist.

Governed execution and bank controlled access are described, but attestations, certifications and a trust center are not enumerated on the product pages. For an agent that executes payments, that leaves assurance asserted rather than shown.

Governance and regulatory controls are said to be built into the Banking OS, with no supervisor, statute or instrument named. An agent that moves money and handles payment disputes works inside electronic transfer error resolution rules and limits on deceptive customer communication.

Dialect, accent and literacy shape how well a conversational system understands people, and voice now runs inbound and outbound with multilingual support. Accuracy across languages and customer groups is unpublished, along with the logic that decides which customers see savings and product prompts.

Evidence behind each decision is retained with its confirmed outcome, so a bank can reconstruct and correct what an agent did. Behind the output stands no published guarantee or indemnity, and a customer given wrong information by an agent has no described route.

Reasoning native AI and Backbase's own banking models describe the platform, with no model provider, speech service or hosting arrangement named. Whether a frontier model sits between a bank's customers and the answer is not stated.

Nexus maps records across core, CRM and payments systems into a shared customer model that agents read and update. Partner integrations against those systems of record are listed in a Backbase Marketplace.

Managed hosting appears among services, but no provider, region or residency commitment is stated for the agents and their conversation records. Customers such as Techcombank in Vietnam, Judo Bank in Australia and Navy Federal Credit Union in the United States each face their own rules on where conversations may be processed.

A 90 day blueprint takes one AI use case to production, and the conversational product can be bought on its own beside an existing digital banking environment. That describes the buying path, while price, package and basis of charge stay unpublished.

Retail, commercial and private banking each carry maintained segment material, with sub pages for functions such as loan origination and advisor performance. The agents serve both customers and employees across that range.