The Hidden Cost of Customer Service’s Second Conversation

A customer contacts an airline about a missing refund. The support agent can see the reservation, the payment history, and the original request. What the agent cannot determine is why the money has not arrived.

The answer may require a finance specialist, a supervisor authorized to approve an exception, or someone familiar with an obscure payment rule.

The customer sees one unresolved problem. Inside the airline, solving it may require several conversations, multiple systems, and a sequence of decisions involving people who do not work together directly.

This is the second conversation of customer service: the communication that happens inside an organization before it can answer someone outside it.

Companies have spent decades improving the first conversation. They have invested in contact centers, customer relationship management systems, ticketing platforms, knowledge bases, and, increasingly, artificial intelligence.

The second conversation has received investment too. Modern service platforms offer internal notes, specialist collaboration, escalation workflows, and integrations with workplace messaging systems.

Yet the underlying challenge remains: getting the right information, expertise, and authority to the person responsible for resolving a customer’s problem.

That challenge is becoming more important as companies spend heavily on automation.

The Automation Paradox

Customer service executives face an unusual financial equation.

According to an August 2026 Gartner survey of 199 service and support leaders, spending on artificial intelligence increased 38%, while overall service and support budgets grew just 2%.

The imbalance reflects an industry investing aggressively in technology without comparable growth in its total resources.

The financial results have been less convincing than the investment figures. Gartner reported in July 2026 that only 24% of surveyed service and support leaders had demonstrated positive financial returns across their AI use cases.

Meanwhile, customers have not abandoned their expectations of human assistance.

A separate Gartner survey of 3,566 customers found that 87% considered access to a human agent essential when companies use generative AI for customer service.

These findings point toward a strategic question.

As automated systems become more capable of handling routine requests, what happens to the work that remains?

One possibility is that human support teams increasingly handle exceptions, disputes, ambiguous requests, and problems requiring decisions across organizational boundaries.

If that happens, the productivity of human agents may depend less on how quickly they can retrieve a standard answer and more on how effectively they can coordinate complex work.

That outcome is not guaranteed. AI may also improve coordination or eliminate some of the work that previously required specialists.

But it suggests that measuring automation rates alone may provide an incomplete picture of customer service productivity.

The Cost of an Escalation

Consider a hypothetical customer service operation employing 2,000 agents.

Each agent handles 30 customer interactions daily, across 220 working days. The organization processes approximately 13.2 million interactions annually.

Suppose 20% require internal assistance, and each of those interactions involves three minutes of avoidable coordination work by the original agent.

That amounts to 132,000 hours annually.

At an illustrative fully loaded labor cost of $35 per hour, the organization is consuming $4.62 million in labor capacity on that activity.

This is not an industry benchmark or an estimate of typical waste. It is a sensitivity calculation based on explicit assumptions.

Nor does eliminating the activity automatically produce $4.62 million in savings. The financial outcome depends on whether recovered capacity reduces overtime, avoids additional hiring, accommodates growth, or improves service.

But the example illustrates why seemingly small operational inefficiencies deserve executive attention.

It also excludes the time spent by colleagues responding to requests.

An agent who spends three minutes seeking assistance may interrupt a specialist who needs additional time to reconstruct the problem, examine the evidence, and provide an answer.

The total cost of the interaction may therefore extend beyond the employee whose ticket is being measured.

Service-desk economics illustrate the cumulative nature of escalation costs. When a support incident moves across multiple tiers or departments, it accumulates handling costs at each stage rather than simply transferring the original expense from one employee to another.

Additional work by additional people consumes additional resources.

The objective should not be to eliminate escalation. Complex problems legitimately require specialized expertise.

It should be to distinguish necessary collaboration from avoidable coordination.

Three Clocks Inside Every Difficult Ticket

Most customer service organizations measure response times, handling times, resolution rates, and customer satisfaction.

These metrics are useful, but they can conceal different operational problems.

A more revealing approach is to separate three kinds of time.

Resolution time is the active work required to investigate and solve the underlying problem.

Coordination time is the effort spent identifying the right person, transferring information, requesting assistance, and managing handoffs.

Waiting time is the elapsed period during which progress depends on someone else providing information, authorization, or action.

These categories are analytically distinct, although real-world workflows can overlap.

Consider a billing dispute.

An agent spends four minutes investigating the account, three minutes preparing a request for finance, and then waits two hours for authorization.

The organization has not spent two hours continuously working on the ticket.

But the customer has waited two hours, and the unresolved case may create follow-up contacts or additional administrative work.

A better knowledge base might reduce investigation time.

A more effective collaboration process might reduce the time needed to prepare and route the request.

Clearer decision authority might eliminate the authorization delay.

Three different problems require three different interventions.

This distinction matters when executives evaluate technology investments.

A faster interface may improve individual productivity without addressing organizational delays. A messaging system may accelerate communication without changing who has authority to act. AI may produce an accurate recommendation that still requires human approval.

Improving one clock does not necessarily improve the others.

How to Measure the Second Conversation

For executives attempting to improve customer service productivity, the first step may not require purchasing another technology platform. It may require measuring work that existing performance dashboards fail to distinguish.

Consider an organization processing several million support requests annually. Management knows its average resolution time, customer satisfaction score, and cost per contact. But how much of that cost comes from solving customer problems, and how much comes from finding the people authorized or equipped to solve them?

Answering that question requires examining the operational dependencies behind individual tickets.

A useful starting point is the internal assistance rate: the proportion of customer interactions that require another employee’s involvement. A high rate is not inherently undesirable. Complex cases often require specialized expertise. The question is whether assistance is necessary and whether the process is efficient.

The second measure is coordination effort: the active employee time consumed by identifying specialists, preparing requests, transferring context, and following up. This should include the work performed by colleagues receiving those requests, not merely the original support agent.

The third is dependency waiting time: how long a case remains unresolved while awaiting information, approval, or action from another person or department. Unlike active coordination effort, this is elapsed time rather than continuous labor consumption. It should not be converted directly into payroll savings.

Finally, organizations should examine repeat-contact rates after internal escalation. If a customer must contact the company again because an internal request remains unresolved, the original coordination problem may generate additional work and customer dissatisfaction.

Together, these measures can help distinguish four different opportunities: reducing unnecessary escalations, making necessary collaboration more efficient, shortening approval delays, and preventing unresolved cases from generating additional contacts.

The distinction matters financially.

A reduction in coordination effort can release employee capacity. A reduction in waiting time can improve the customer experience without necessarily reducing labor expense. A reduction in repeat contacts may accomplish both.

For technology executives, the same measurements can help identify whether the bottleneck is primarily organizational or technical.

If agents repeatedly search for the correct specialist, better routing and clearer ownership may help. If specialists must reconstruct customer histories, improvements in context sharing and system integration may be more valuable. If requests wait for approval despite complete information, the problem may concern decision authority rather than communication speed.

None of these interventions requires assuming that additional software is always the answer.

A practical evaluation would begin with a representative sample of escalated cases, classify the sources of delay, and establish a baseline. Management could then test a specific process or technology change against comparable cases, measuring whether it reduces effort, waiting time, repeat contacts, or some combination of the three.

The objective is not to maximize the number of conversations employees can conduct.

It is to determine which conversations create value, which merely transfer information, and which should never have been necessary.

The Problem Isn’t a Lack of Software

Enterprise customer service platforms already contain substantial collaboration functionality.

Zendesk, for example, supports internal notes, ticket followers, light agents, and side conversations. Its collaboration features can involve email, Slack, Microsoft Teams, and related tickets.

Other enterprise platforms also offer mechanisms for routing work, preserving context, and coordinating employees.

The existence of these capabilities makes the problem more interesting, not less.

If organizations have tools for internal communication, why can resolving a customer problem still require repeated explanations and multiple handoffs?

Part of the answer lies in organizational structure.

A support agent may need approval from finance, information from engineering, or action from a logistics provider.

Those groups operate under different priorities, permissions, and systems.

The support team may be measured on resolution speed. Finance may prioritize payment controls. Engineering may focus on product reliability.

An urgent request for one department may be routine work competing with other obligations in another.

Software can make these dependencies visible. It cannot automatically eliminate conflicting priorities or assign decision-making authority.

There is also a technical dimension.

Customer information may reside in one system, operational records in another, and internal discussions in a third.

Moving information between them introduces opportunities for missing context, inconsistent records, and duplicated work.

Integrations can reduce those problems, but they introduce their own requirements around permissions, security, data quality, and system maintenance.

For technology executives, the central challenge is therefore not simply selecting a communication platform.

It is designing an operating environment in which information and decisions can move reliably across organizational boundaries.

When the Customer Has to Start Over

The consequences of fragmented information are visible to customers.

Zendesk’s 2026 Customer Experience Trends research, based on surveys involving more than 11,000 consumers and business respondents, found that 74% of surveyed consumers were frustrated when they had to repeat information.

Another 81% wanted agents to continue conversations without retracing previous steps.

These findings do not establish how much of that frustration results specifically from internal employee collaboration.

But they illustrate why continuity matters.

When a customer explains the same problem repeatedly, the failure may involve missing records, inadequate handoffs, disconnected channels, or employees who cannot access the relevant information.

The customer rarely distinguishes among these causes.

From the outside, they all appear as an organization that cannot remember what it has already been told.

This is where internal coordination becomes more than an operating-cost issue.

It becomes part of the customer experience.

A company may respond quickly to every message and still deliver poor service if each interaction begins without the context of the previous one.

Speed and continuity are not interchangeable.

What AI Changes—and What It Doesn’t

AI introduces another layer to this problem.

It can summarize conversations, retrieve documentation, classify requests, recommend actions, and identify relevant information across systems.

In some cases, it can resolve problems without human involvement.

In others, it may help employees coordinate more effectively by reducing the time needed to reconstruct a case or locate an appropriate specialist.

But the distinction between information and authority remains important.

An AI system may identify why a refund was rejected. That does not necessarily mean it should be authorized to reverse the decision.

It may recommend a technical remedy without having permission to modify a production system.

It may retrieve a policy correctly while failing to account for an unusual exception.

The challenge is not merely giving AI access to more information. It is establishing when the system can act, when it must defer, and how its recommendations fit into accountable business processes.

Research from Intercom’s 2026 Customer Service Transformation Report suggests that many organizations are still working through these questions.

The survey of 2,470 support professionals found that 82% of senior leaders reported AI investment over the preceding year, but only 10% of respondents described their organizations as having reached mature deployment.

The future of customer support will undoubtedly involve more automation. But even highly automated companies will encounter problems that require human judgment, institutional knowledge and cooperation.

The economic opportunity is not merely to make the first conversation faster.

It is to make the second conversation less costly—and, wherever possible, unnecessary.