The Missing Middle: Where AI-Shaped CX Is Really Decided

By the time a customer reaches a human employee, the organisation may already have routed, ranked, scored or restricted what can happen next. An AI system may have decided which queue the customer enters, what the employee sees, which answer is recommended and whether an exception appears risky. The conversation feels human. Much of the decision architecture behind it is not.

This is why customer experience leaders should stop asking only, “Where can we automate?” A more important question is: “Who is now making the service decision, and whose judgement still counts when the system is wrong, incomplete or blind to context?”

In contact centres and other service environments, the AI conversation usually concentrates on two groups. Executives approve the investment; frontline employees and customers use the resulting tools. Between them sits the middle-management layer, often absent from the transformation story yet central to what happens every day.

Middle managers turn strategy into operating practice. They interpret performance targets, coach employees, handle exceptions, approve escalations and explain why one outcome is considered efficient, fair or responsible. When AI enters the workflow, their role becomes more demanding. They must decide when a recommendation should be followed, when human judgement should override it, and when a recurring exception signals a deeper design flaw.

That is leadership in practice. It is not visible only in a job title or steering committee. It appears in the small, repeated decisions through which work is coordinated and accountability is assigned.

In South African service environments, this layer also carries contextual knowledge that is difficult to encode in a generic model or rule set. Language, affordability, digital access, vulnerability and regulatory obligations can change what a fair service outcome looks like. The middle manager is often where these realities meet standardised process. If that judgement is removed, ignored or measured only as a deviation from the script, AI may produce consistency without appropriateness. That is not a technical defect alone. It is an operating-model and leadership choice.

A useful example comes from IKEA. Reuters reported in 2023 that its Billie chatbot had handled 47% of customer enquiries over two years, while 8,500 call-centre employees had been trained as interior design advisers. The compelling lesson is not simply that a bot absorbed routine queries. It is that the organisation reconsidered what human contribution could become when routine work shifted. Technology changed the task; people and leadership had to reshape the work.

What should leaders do differently?

First, map decisions, not only touchpoints. Most journey maps show what the customer does, sees and feels. They should also show where an algorithm routes, prioritises, recommends, approves, declines or suppresses an option. For every decision point, identify the accountable human and the available route for challenge.

Second, equip middle managers for judgement, not only adoption. AI literacy is necessary, but knowing how to operate a tool is not enough. Managers need to understand its purpose, limits, data dependencies and likely failure modes. They also need explicit authority to question outputs without being treated as resistant to change. Gallup’s 2026 State of the Global Workplace report reinforces this point: manager-led adoption is among the leading drivers of frequent AI use, yet active managerial support remains limited.

Third, treat exceptions as intelligence. A transfer, override or complaint is not automatically a failure. It may reveal a missing customer circumstance, an unclear rule, a weak knowledge base or a policy that is efficient for the organisation but unreasonable for the customer. If exceptions are merely closed, the organisation loses the opportunity to learn.

Fourth, rebalance the scorecard. Speed, cost and containment matter, but they do not tell us whether the right decision was made. Add measures for repeat contact, avoidable escalation, override quality, customer effort, vulnerable-customer outcomes and whether recurring problems change the process.

This is where a 360-degree assessment becomes useful. AI-shaped customer experience cannot be understood by inspecting the technology alone. Leaders need to examine the surrounding service system: strategy, people, processes, technology, customer interactions, data, quality, risk and the operating environment. A strong chatbot inside a weak system can simply make a poor journey move faster.

AI is not merely a technology project. It is a human project because it redistributes discretion, voice, responsibility and learning. The decisive question is not only whether AI can serve the customer. It is whether the people closest to the work remain able, equipped and authorised to challenge the decision when it cannot.

BIOGRAPHY

Iemraan Kara is Founder and CEO of The Strategic Thinking Partner, a CX and BPO/GBS strategist, and a PhD Leadership researcher examining AI-shaped work. MBA, MPhil: Management Leadership in Emerging Countries.
He is part of the team behind ASSESS 360, an independent contact-centre assessment and benchmarking proposition focused on maturity, customer impact, operational risk and practical improvement.