Taras Tymoshchuk is the founder and CEO of Geniusee.
A 2025 World Economic Forum and Cambridge Center for Alternative Finance survey of 240 fintech firms across six regions found that 80% were implementing AI across multiple business domains. Among adopters, 83% reported improved customer experience, 75% higher profitability and 75% lower costs.
These gains explain why AI-powered personalization has become a priority for fintech leaders. Banking apps can recommend cards, investment platforms can tailor guidance to individual goals and payment systems can identify risk earlier. Yet even a highly relevant recommendation can rely on outdated data, exceed a user’s permissions or produce a decision nobody can defend.
In my work with fintech product teams, I keep seeing attention concentrated on the customer-facing feature. The harder work lies beneath it: establishing whether the data is current, whether the system is authorized to use it and whether someone can reconstruct how the result was produced.
Treat The Data Supply Chain As Part Of The Product
Financial AI rarely works from one clean source. Inputs may come from core banking systems, payment gateways, KYC providers, market feeds, CRMs and third-party APIs. Each source carries different formats, timestamps, ownership rules and failure modes.
This is already a recognized industry concern. In a Bank of England and FCA survey, four of the five most significant current AI risks identified by financial firms were data-related: privacy and protection, data quality, data security and bias and representativeness.
While working with market data platforms, I have seen how confidence breaks down when prices, identifiers and timestamps arrive from multiple providers at different intervals. A model can produce a fluent answer from inconsistent inputs. The business still carries the consequences.
Validated datasets, lineage, freshness checks and permission rules belong in the product design. When AI flags a transaction or recommends a financial product, the team should be able to identify which records informed the result and whether they were complete and current.
Make Important Decisions Reconstructable
AI is already influencing financial decisions. The same research by the Bank of England and the FCA found that 55% of reported AI use cases involved some degree of automated decision-making. At the same time, 46% of firms reported having only a partial understanding of the AI technologies they used.
That gap deserves executive attention.
Role-based access, consent controls, encryption, audit logs and human approval for high-impact actions should be designed into the workflow. If an assistant retrieves payment information, recommends a credit-related action or changes a process, the organization should be able to reconstruct what it accessed, how the recommendation was formed, who approved it and which system was affected.
Compliance added near launch usually creates redesign work. Bringing legal, security and risk specialists into the design stage gives product and engineering teams clearer boundaries from the beginning.
Limit What The System Can Access And Change
Personalization increases the financial, behavioral and identity data available to an AI system. It also expands the consequences of prompt injection, data leakage, insecure integrations and excessive permissions.
Retrieval, recommendation and action should remain distinct responsibilities. Access should be restricted by role, tenant and use case. High-impact changes should require explicit approval, and model behavior should be monitored alongside application and security events.
This matters in payment environments, where an assistant may need sufficient context to explain a transaction without authority to move money. Credibility grows when permissions match the precise task assigned.
Measure Whether Personalization Serves The Customer
AI can help customers optimize card rewards, understand investment progress or select an appropriate payment method. Those capabilities can also steer people toward products that generate better economics for the provider.
Fintech leaders should define customer-aligned personalization before selecting the model or interface. I would assess production readiness through three signals:
• Traceability: Can the team identify the data and rules behind a recommendation?
• Boundaries: Can the system access and change only what its role requires?
• Challengeability: Can a customer, employee or reviewer question the result and trigger human review?
Teams should also track override rates, false positives, data-quality failures, unexplained outputs, access violations and customer complaints. These indicators reveal whether trust survives beyond the demo.
AI adoption in finance is accelerating because the business case is visible. Sustaining that value requires systems whose data, permissions and decisions can withstand scrutiny. Personalization earns its place when the organization can explain, control and prove that it serves the customer.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
