Teardown: Six Weeks to Two

Client names are withheld under confidentiality. The structure, the numbers and the mistakes are reported as they happened.

Problem

A services business with a six-week turnaround on a document-heavy customer workflow. Three teams touched every case, each rebuilding context the previous one already had. Two prior AI pilots had been built and shelved: both worked in demo, neither had an owner outside the innovation function.

Intervention

• Baseline measured on 90 real cases before any build began
• Ownership moved to the operations lead, with a weekly written decision memo
• Evaluation harness built in weeks one and two, release threshold agreed in advance
• One workflow automated end to end, with human review only on flagged exceptions
• The manual route switched off at week seven

Result

Turnaround moved from six weeks to two. Exception volume settled at roughly one case in eight, which the same team absorbed without added headcount. The measured saving over the following year was reported internally as cost avoidance rather than headcount reduction — the team was redeployed onto backlog.

What nearly went wrong

The release threshold was set too high in week three and would have kept the workflow in review indefinitely. It was renegotiated against the human baseline — not against perfection — which is the single change that made the switch-off possible.