Definition. The “Frozen Middle” describes a change that stalls between a leadership decision and daily practice because the managers who run the work have not been given a workable path to implement it.
Leadership can approve an AI initiative and a project team can build a promising pilot, yet the people responsible for staffing, service quality and exceptions may still be unable to use it. Calling that resistance too early hides useful information. The manager may see a real problem in the data, have no capacity to train the team or lack authority to change an approval step.
What to look for
- The pilot has a technical owner but nobody responsible for changing the working process.
- Managers are asked to adopt a tool while still being measured against targets built for the old workflow.
- The team has not agreed who handles exceptions, customer complaints or mistakes after the change.
- Approval requests circulate between functions without a decision owner or a date for resolving them.
- Training is complete, but people have no time or safe place to practise on their actual tasks.
These signs can coexist. Ask the manager and the people doing the work to walk through one real case. Where would the AI output appear? Who would check it? What would happen if it were wrong? The answers often reveal a missing operating decision more clearly than a survey about attitudes to AI.
Understand the constraint
Some concerns are legitimate boundaries. A data source may be unusable, an approval may be required, or a customer promise may be at risk. Verify the concern and assign work to resolve it. Other delays come from a mismatch between responsibilities and incentives: a manager is held accountable for the result but has not been allowed to change staffing, measures or procedures. In either case, an instruction from above is not a substitute for an operating plan.
The pattern can also reflect fear of losing a valued role. Be direct about how the role changes. A manager may spend less time assigning routine tasks and more time checking exceptions, improving rules and supporting the team. That work needs authority, training and recognition rather than a vague promise that AI will “free up time”.
Move the work forward
Name a process owner with authority to decide how the workflow changes. Bring the relevant managers into the design of the new flow, including its limits and escalation route. Agree on measures for the outcome and for service quality, then revisit any target that rewards the old way of working while penalising the new one.
Use a Dual Run where it helps people see the system's proposed decisions before they affect customers. Record disagreements and act on them. A Champions Network can help colleagues practise and return specific problems to the owner; it should support managers, not secretly bypass them.
If the pilot keeps being extended, hold an explicit exit review. Decide whether the issue is model quality, process design, capacity or a permission that has not been granted. Each calls for different work. A cautious manager who names a real risk and a way to test it is helping the project; the concern becomes a blocker only when no one can resolve it or decide how to proceed.
