Overview
We piloted human-in-the-loop automation for two workflows: (1) proposal drafting (template population + quality checks) and (2) reconciliation (data → reconciled schedule). The hybrid model reduced manual hours by ~45% while preserving decision quality.
Decision matrix: automate vs human
Automate when: tasks are repetitive, rules-based, and low-ambiguity (e.g., field mapping, formatting). Human when: tasks demand contextual judgment, stakeholder negotiation, or interpretation of messy inputs.
Pilot results
- Proposal drafting: template autopopulation + 1-pass human edit reduced time from 8h → 3.5h (56% reduction).
- Reconciliation: automated matching rules flagged exceptions that were resolved manually — 40% fewer man-hours for the reconciler and faster cycle time.
- User acceptance: staff reported that the hybrid flow preserved control while saving time on repetitive steps.
Implementation pattern
- Map the end-to-end workflow and mark tasks as A (automate), H (human), or H/A (hybrid).
- Build small automations for A tasks using scripts or RPA connectors; expose results in a review dashboard for human operators.
- Measure cycle time and error rates before/after and iterate.
Starter assets
Included: a decision matrix template, a proposal autopopulation script example, and a reconciler checklist.