Table of Contents
At CXFS, the question now is how, and for an industry this regulated, the answers are surprisingly far along.
I attended the Customer Experience for Financial Services Summit (CXFS) in Toronto this week expecting to hear thoughtful conversations about innovation, regulation, and the future of customer service.
What I did not expect was the degree of progress already being made.
The industry is already moving beyond isolated AI experiments and applying new technologies to meaningful customer journeys and doing so within some of the most demanding regulatory, privacy, and risk environments in the world.
My biggest takeaway was simple:
Financial services companies are proving that innovation, regulatory control, and customer adoption do not have to compete with one another.
1. Changes are coming across the entire experience
Agentic AI was an important part of the conversation, but it was only one part of a much broader transformation.
- Interactive Teller Machines are expanding the services customers can complete remotely while still providing access to a person when assistance or judgment is needed.
- Digital-payment experiences are becoming faster, more intuitive, and increasingly embedded into the broader customer journey.
- Identity, authentication, and fraud technologies are reducing friction while strengthening security.
There is also an emerging shift from payments initiated by customers to payments executed by AI agents, an evolution that will make trust, permission, transparency, and control even more important.
These innovations may appear unrelated, but they reflect the same fundamental change: financial institutions are rethinking where work should happen, who, or what, should perform it, and when a human should become involved. That is also the central opportunity presented by agentic AI.
2. Agentic AI is moving from assistance to ownership
Much of the early application of AI in customer experience focused on assisting employees. Those capabilities remain valuable, but agentic AI introduces a different operating model.
Instead of merely assisting someone else, an AI agent can own a customer request from beginning to end. It can understand the customer’s intent, gather information, follow business rules, complete actions across systems, communicate the outcome, and determine when additional guidance is necessary. For financial institutions, however, autonomy cannot mean the absence of control.
Banks and insurers operate in environments where a seemingly simple customer request may involve identity verification, product eligibility, disclosure requirements, fraud controls, policy interpretation, or regulatory obligations. This is where a generative-only approach shows its limits. An LLM can understand the request and carry the conversation, but it cannot guarantee a regulated process runs identically every time, and it doesn't know where its authority ends.
Owning a request end-to-end in this environment takes three things working together:
- Agentic intelligence to understand the situation, make decisions, and manage the interaction.
- Deterministic automation to ensure that defined processes, policies, and regulatory requirements are followed precisely.
- Prescriptive human guidance when the situation requires judgment, authority, empathy, or exception handling.
This changes what 'owning' a request means. The AI owns the interaction, but it doesn't have to produce the outcome by itself, and knowing when to ask is part of the job rather than a sign it failed.
3. Human guidance is not a failure of automation
One of the most important mindset shifts for CX leaders is how we define successful automation. That traditional definition, where an automation is only successful when a human never becomes involved, is too narrow for agentic AI, especially in financial services.
There will always be moments when seeking human guidance is the safest and most effective action an AI agent can take. The goal should be to use human expertise intentionally while preserving the continuity and efficiency of the AI-led experience.
This creates a powerful new model: one human expert can guide multiple simultaneous AI-led interactions rather than personally taking ownership of one customer conversation at a time.
That model also changes how success gets measured. Containment and cost per contact still matter, but they don't tell you whether the AI resolved the request correctly, followed the required process, or asked for help at the right moment.
An interaction where the AI consulted an expert and then finished the job is a success, even if a containment report could count it as a miss.
4. Regulation can become a design requirement, not an innovation barrier
The progress shared in Toronto challenged the assumption that highly regulated industries must always move slowly.
Many organizations are no longer treating compliance, security, privacy, and governance as reasons to delay innovation. They are incorporating those requirements into the design of the experience from the beginning. And you could hear the difference in the questions being asked:
- What decisions can AI make autonomously?
- Which processes must always follow deterministic rules?
- What evidence should be captured for auditability?
- When is human authorization required?
- How should the AI explain what it did?
- Who is accountable for monitoring and improving its performance?
- How will customers reach a person when they want or need one?
My colleague Max Black captured how these questions are answered in practice, starting with a line I keep coming back to: “The governance teams are actually our friends.”
5. Customer adoption may be stronger than many leaders expect
Another encouraging observation was the apparent willingness of customers to adopt well-designed technology. Customers are not opposed to automation simply because it is automation. They resist experiences that create friction, limit their choices, misunderstand their needs, or prevent them from reaching a resolution.
The question that matters is simpler: did the customer get what they needed? They ultimately judge the experience by the outcome, not by the technology modality.
My lasting impression from Toronto
I left the conference more optimistic about the future of financial services customer experience.
- Banks and insurance companies are not waiting for every uncertainty surrounding AI to disappear. They are learning how to innovate responsibly inside those uncertainties.
- Governance is showing up as a design input rather than a gate at the end of the process.
- Human guidance is becoming part of how AI does the job, not evidence that it couldn't.
- Customers will adopt what works. They are judging the outcome, not the technology.
The leaders getting this right are building a customer-experience model in which AI acts, deterministic processes provide control, and human experts guide the moments that matter most. The most regulated industry in the room may end up being the one that shows everyone else how agentic AI is supposed to be done.
Until next year, Toronto!
