Five anonymized stories. Five workflows that stopped breaking.

A two-sided marketplace was processing hundreds of publisher submissions weekly through manual review and routing. The ops team was the bottleneck for every transaction, quality was inconsistent across reviewers, and onboarding new reviewers took weeks. Growth was capped by how many submissions one human could read in a day.

A Zoom → n8n → Claude pipeline that captured intake calls, transcribed and structured them, scored submissions against a multi-dimensional rubric, routed approved publishers into the right marketplace tier, and flagged edge cases for human review. Wrapped in an eval harness so scoring drift could be monitored and the rubric tuned over time without retraining.

A multi-brand publisher was running editorial intake through a single overloaded editor inbox. Every brief came in via Typeform, got manually triaged, manually tagged, and manually routed to the right desk — averaging 45 minutes per piece before a writer was even briefed. Misroutes were common, writers got dropped acks, and capacity planning was guesswork.

A form → n8n → Claude pipeline that classifies briefs against a brand-fit rubric, auto-routes to the right desk based on confidence threshold, escalates ambiguous fits to the editor with reasoning, and sends acknowledgement to the writer. Human-in-the-loop kicks in only when confidence drops below 0.85, and a weekly digest surfaces drift patterns to the editor.

A multi-team services company knew they were behind on AI but had no internal capability, no roadmap, and no clear first wins. Leadership wanted real production deployments, not a pilot graveyard, and needed an outside lens to prioritize across competing internal asks.

90-day audit and roadmap covering 12 candidate workflows, scored on impact and feasibility. Prioritized three for immediate build: client deliverable QA, capacity planning, and contract intake. Shipped all three to production with adoption tracking, change management, and team training so the framework outlives the engagement.

A team handling thousands of customer calls per week had no scalable way to monitor quality. Manual QA covered ~5% of calls; the other 95% were a black box. Coaching was reactive, at-risk accounts surfaced too late, and team leads were spending more time pulling samples than coaching.

An automated call scoring system using Claude to evaluate every transcript against a multi-dimensional rubric, surface coaching moments in real time, and flag at-risk accounts before they churned. Dashboard for team leads, weekly trend reports for leadership, eval harness to catch scoring drift across the rubric.

A high-volume support operation was staffing reactively. Demand forecasts were built in spreadsheets once a quarter, ignored mid-month, and consistently off by 20–30%. Team leads were either over-staffed and burning margin or under-staffed and burning out the team. No early warning when volume was about to spike.

A capacity planning system that ingests historical volume, seasonality patterns, and live signal feeds, runs daily forecasts through a structured model, and surfaces staffing recommendations to team leads with confidence intervals. Alert layer flags volume spikes 48 hours before they hit. Dashboard for ops, weekly digest for leadership.