The experience worked. The process didn't.
A refund had been stuck for nine days. When an AI agent picked up the case, it stopped: no system recorded who could approve the payment. So, an experienced advisor messaged a colleague in finance, and within the hour the refund was approved.
The customer got their money, and by every measure the business tracks, the experience was a success. Yet it succeeded only because one employee knew who to ask. The AI agent didn’t fail; it exposed a dependency nobody had ever written down.
That advisor performed a small act of heroics. Customer experience has depended on such acts for decades, across every kind of interaction, not only in the contact centre.
What are heroics in customer experience? Heroics are the unofficial actions employees take to compensate for weaknesses in a process: piecing together a customer's history from several systems, chasing approvals, coordinating between departments or making judgement calls that exist nowhere in documented workflows.
Customer experience was built on heroics
Most organisations believe their customer journeys run through systems. Many actually run through people. A customer who starts a mortgage application online and finishes it in a branch relies on an advisor who checks three systems to work out what has already been agreed. A business client renewing a contract relies on an account manager who remembers last year's billing dispute, because the CRM doesn't show it. In each case, the experience works because a person bridges the gap between channels, systems or teams.
The better the employee, the less visible the problem. That creates a counterintuitive blind spot: operational weaknesses are often hardest to see where the team is most experienced, because that is where they are fixed fastest and most quietly. Satisfaction scores record whether the customer got the outcome, not how much unseen effort it took. An organisation can post strong results while a handful of people hold its journeys together, and assume the process is working when it isn't.
AI agents expose where heroics are required
AI agents are becoming stress tests for the whole customer experience, wherever they are deployed. An agent greeting a returning customer knows nothing about them unless the CRM supplies the context. What an experienced employee compensates for without thinking, an agent exposes immediately.
Sometimes the agent stops. Sometimes, more dangerously, it fills the gap itself. In April 2025, users of the coding tool Cursor were logged out when switching devices, and the AI agent answering its customers told them this reflected a new login policy. No such policy existed; a bug was responsible, and users had begun cancelling subscriptions before the company corrected the record (CX Today). The agent had no reliable answer, so instead of handing over, it invented one. The same risk exists anywhere an agent speaks for a business, from sales enquiries to onboarding.
The issue is rarely AI capability alone. It is AI operating inside journeys designed around human rescue.
From heroics to repeatability
Good customer experience has long been defined by outcomes: fast resolution, high satisfaction and low effort. Those outcomes still matter. What is changing is what it takes to deliver them.
Repeatability means every customer receives the same standard of experience, whichever advisor, team, channel or AI agent handles their request. It does not mean uniformity. A first-time buyer, a long-standing business client and a customer in a vulnerable situation need very different journeys, and they should get them. What should not vary is the quality underneath: the same context at every touchpoint, the same discipline in how decisions are made and the same route to the right person when one is needed.
| Traditional CX | AI-ready CX | What changes |
|---|---|---|
| Heroics | Repeatability | Quality no longer depends on who handles the case |
| Individual expertise | Shared knowledge | What one employee knows is available to every employee and every agent |
| Human workarounds | Designed routing and escalation | Customers reach the right person or channel first time, with their context intact |
The capabilities that make this possible are familiar: intelligent routing that sends each customer to the right resource first time, CRM context that means nobody has to repeat themselves, cross-channel continuity that lets a journey move from app to chat to phone without starting again, and analytics that show where journeys still depend on heroics. AI agents sit on top of these foundations, and they are only as good as what lies underneath.
Repeatability also includes a reliable route to a person. In a 2026 Gartner survey, 87% of customers said companies using generative AI for service must provide access to a human agent (Gartner). Repeatable CX is not CX without people. It is CX in which people are deployed by design rather than by rescue.
Why many AI agent projects disappoint
Most organisations still approach AI agents the same way: select a use case, deploy an agent, measure automation. In doing so, they try to automate experiences that run on heroics they cannot see.
The consequences are showing. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls (Reuters, via The Star). Its analysts observe that AI handles routine problems well but often struggles with exceptions and high-risk situations (Gartner). Exceptions are exactly where heroics live.
AI agents as customer experience diagnostics
The first wave of AI adoption focused on replacing tasks. The next may prove more valuable, because it reveals how customer experience actually operates.
The organisations getting the most from AI agents treat them as diagnostics, not just automation tools. Every place an agent stops, escalates, loses context between channels or relies on information that does not exist marks a point where the journey depends on heroics. Running an agent in shadow mode, proposing actions on live cases while people still do the work, surfaces those points before any customer is affected.
What matters is how those findings are read. Treat each one as a technology problem and the response is to tune the prompt or try a better model. Treat it as a customer experience problem and the response is to fix the approval with no owner, connect the data that disagrees and record the knowledge only a few people hold. The employees who have been performing the heroics are the ones who can explain them, so they need to be partners in this work, not its casualties.
Most executives expect AI to automate. Its first and most valuable job may be to diagnose.
The advantage is knowledge, not models
Every company can license the same models, from OpenAI's GPT to Anthropic's Claude, Google's Gemini and Microsoft Copilot. What none can buy is an organisation's own operational knowledge: its escalation paths, its customer context and its judgement about when a rule should bend. Companies that capture that knowledge will deploy stronger agents with every fix. Companies that cut experienced staff before capturing it risk losing it for good. That makes the shift from heroics to repeatability a board-level decision, not a technology project.
The future of CX is repeatability, not heroics
For years, customer experience has been judged by outcomes. AI agents introduce a tougher test: can that experience be delivered consistently, across every channel and scenario, without relying on individual employees to bridge the gaps?
The organisations that pass will not necessarily have the most advanced models. They will be the ones that turn customer experience into knowledge, decisions and journeys that people and AI can deliver to the same standard. The answer is no longer heroics. It is repeatability.
So how do organisations move from heroics to repeatability? Next week, in Part 2, we set out three practical steps.
Sources
.jpg?width=40&height=40&name=Adam%20Bearder%20Originalv2%20(002).jpg){kind=small}
