Alejandro Murcia, Director of Specialized Sales at Sinch
Banks are increasingly using artificial intelligence to support customer communication. From explaining payments and supporting identity checks, to responding when fraud is suspected, AI can help make interactions quicker, more accessible and tailored to individual needs. However, reliability is vital when these tools are handling financial communications. An incorrect response or notification can have serious consequences for a customer trying to access their money, prevent fraud or receive the financial support they need.
As a result, preventing these failures has become a priority for financial services firms. However, evidence suggests that investment in governance alone cannot guarantee reliable customer interactions.
Governance alone cannot guarantee reliable AI
Financial institutions already understand the importance of protecting customer communications. Recent research from Sinch found that 78% of financial institutions consider trust, security and compliance a leading investment priority for their AI communications programmes.
Despite this focus, 69% of financial institutions have had to roll back an AI agent. This suggests that high levels of investment and attention to governance do not always prevent problems once an agent reaches customers.
Rolling back an agent can be a necessary response when monitoring identifies poor performance or incorrect behaviour. The underlying issue, however, may sit in different parts of the technology stack. Data exposure, missing context or incomplete audit trails can emerge from the communications infrastructure, while inaccurate or inappropriate responses may be linked to the model, its training data or prompts.
Model guardrails can shape how an agent responds, while controls built into the communications platform can provide protections such as personal data masking. Understanding where a problem originates also allows engineering teams to address its cause rather than repeatedly correcting individual failures.
Performance after launch matters
The real test for an AI communications agent comes once customers begin relying on it, particularly during periods of high demand. If mobile payments are disrupted, for example, customers may immediately turn to their bank for information and support.
Failures during these periods can create pressure elsewhere as customers move towards contact centres and increase the workload for human customer service agents precisely when demand is already high. Engineering teams are then diverted towards diagnosing and rebuilding the agent, while inaccurate or delayed interactions create wider reputational and regulatory risks.
Financial institutions therefore need visibility into how agents behave throughout their time in production. Monitoring should help teams identify deteriorating performance early and intervene before customers are affected.
The infrastructure behind AI shapes customer trust
The infrastructure supporting an AI agent is fundamental to maintaining trust. Sinch’s research found that the reliability of communications infrastructure was more closely linked to successful AI deployment than levels of investment, the guardrails in place or how long an organisation had been using AI. Yet, 42% of organisations said their current communications provider did not offer sufficient reliability to operate AI at scale and 37% cited limited multichannel capabilities. These findings highlight that even well governed AI agents can struggle to deliver reliable customer experiences if the infrastructure supporting them is not fit for purpose.
For financial institutions, limitations around multichannel communications can become visible as customers move between agents and channels such as messaging, email and voice. Someone who raises a suspicious payment through chat should be able to continue that conversation when transferred to a fraud specialist by phone. Similarly, a customer discussing a vulnerable financial situation such as struggling to keep up mortgage repayments, should not have to repeatedly provide sensitive information because context cannot move securely between systems.
More than half of surveyed organisations are building their own capabilities to preserve this context across channels because these functions are unavailable natively through their platform. As AI expands across messaging, email and voice, building these capabilities for individual deployments can place additional demands on engineering teams and complicate governance.
However, the correct platforms should provide a reliable foundation of native security, monitoring and orchestration capabilities, while financial institutions add safeguards reflecting their products, customers, risk profiles and regulatory obligations.
Scaling AI communications with confidence
AI has great potential to make financial support faster, more conversational and more responsive, particularly at moments when customers need clear information quickly. Achieving that at scale requires complete oversight of the journey behind an AI interaction, including how information reaches an agent, how the agent communicates with the customer and how conversations are handed to human teams when necessary.
As financial institutions expand their use of AI, maintaining accurate, secure and connected communications will be central to preserving customer confidence. The quality of the underlying communications infrastructure will play a major role in determining whether AI can deliver that experience consistently at scale.