Why are customers still struggling?

A closed conversation can leave an open problem.

Containment records whether a conversation stayed inside automation. It cannot tell you whether the customer completed a suitable purchase, resolved a problem, tried again elsewhere or gave up.

An increase in escalations can reflect better recognition of a problem that needs human judgment. Examine the type of work reaching people, its difficulty and the eventual resolution before treating that increase as a failure.

From my writing: Count What Reaches the Human.

Look at what the employee can do next.

A handoff should preserve the customer’s request, what the AI attempted and the reason it could not finish. The receiving employee needs enough context and authority to take the next decision. Requiring the customer to reconstruct the interaction transfers work to them.

When AI absorbs routine questions, the human queue can contain more difficult work. Staffing, coaching and performance expectations need to recognize that change. A longer conversation may be doing necessary recovery work that the old measures overlook.

From my writing: Make Self-Service Earn Its Name.

Choose the fix that addresses the unfinished work.

A different model will not, by itself, give an employee missing account information or permission to resolve an exception. Establish whether the failure sits in the answer, an attempted action, a policy or the handoff. Leadership can then decide whether to improve the AI, change the operating rules or put the task with someone who can complete it.