Argument in brief

Trust is usually discussed in the wrong place. Organisations put it in brand campaigns, values, tone of voice, annual reports and sentiment trackers. But customers do not decide whether to trust an organisation by reading its promises. They decide when they need the system to behave properly and cannot see the machinery underneath.

• Trust is decided in operating moments, not brand language.
• At scale, trust travels through architecture: data, rules, controls, permissions, channels and recovery loops.
• Consistency, explanation, reliability and recovery are stronger trust signals than promises.

Diagnostic questionWhere does the customer depend on your system without being able to see, question or repair the machinery underneath?

Argument in brief

A complaint is not just a service event. It is evidence that something inside the organisation has reached the customer in a form they can feel. The customer was asked to repeat information. The app said one thing and the call centre said another. A case was closed while the cause remained open. A published process was followed and still produced failure.

• Complaints are evidence of structural failure, not just moments of customer dissatisfaction.
• Silence is not satisfaction; many customers absorb the cost without complaining.
• Customers often pay the integration tax created by disconnected teams, channels and systems.

Diagnostic questionWhich recurring complaint points to a structure the organisation has learned to process rather than repair?

Argument in brief

Every team can now build an agent. That is useful, and dangerous. Sales, support, finance, operations and engineering can all create local agents that make sense inside their own work. The risk is that useful local systems create a broken whole.

• Useful local agents can still create a bad system.
• The critical design problem is the join between agents, systems, teams and accountabilities.
• Identity, context, memory and escalation are not implementation details. They are control surfaces.

Diagnostic questionWho owns the seam when an agent crosses a system, a team, a permission boundary or a customer consequence?

Argument in brief

AI can make tasks faster without making the business better. A model can summarise the case, draft the response, prepare the pack or triage the document. Then the work can still wait in the same queue, cross the same handoff, hit the same approval or fail at the same ownership boundary.

• Task productivity is not the same as enterprise value.
• Process debt is the inherited work an organisation would not design again but still pays people, customers and systems to carry.
• Some friction is debt and some friction is protection; the distinction has to be made before automation.

Diagnostic questionWhat part of the operating model are you about to teach AI to preserve?