Compressing a multi-month innovation workflow into under five minutes

The client's corporate innovation process was fragmented across tools, teams, and handoffs. Ideas waited in queues, context was rebuilt at every stage, and the elapsed time from intake to a structured decision ran into months — most of it spent waiting rather than thinking.

• Led agentic AI workflow design for a Big Four consulting company as Project Manager at Illinois Business Consulting.

• Mapped the existing fragmented corporate innovation process end to end, including the handoffs that were never written down.

• Specified the individual AI agents, their responsibilities, and the boundaries between them.

• Defined automation logic, data movement between stages, and where a human must stay in the loop.

• Built the working workflow in n8n.

• Separately conducted early-stage M&A screening for a Fortune 500 manufacturing company, evaluating acquisition targets across market opportunity, technical capability, and strategic fit.

1. 01

   Mapped the process as it actually runs, not as the org chart describes it — the delays lived in the handoffs, not the work.

2. 02

   Decomposed the workflow into discrete agent responsibilities narrow enough to be reliable, rather than one prompt asked to do everything.

3. 03

   Made data movement explicit between stages so context carries forward instead of being reconstructed by the next human.

4. 04

   Placed human-in-the-loop checkpoints where judgement genuinely matters and removed them where they were only queue-forming.

5. 05

   Built it in n8n so the client owns a running system they can inspect and modify — not a demo.

• The bottleneck was organisational, not computational: most elapsed time was waiting, so automating the thinking alone would have changed little.

• Agent boundaries had to be drawn where they stay reliable — too broad and quality collapses, too narrow and the orchestration overhead exceeds the benefit.

• Human checkpoints had to be justified individually; every one reintroduces latency, and the default in a consulting org is to add more.

• The output had to survive handoff to a client team that would maintain it without me.

Turning ambiguity into an executable system. The value wasn't 'using AI' — it was being willing to map a messy process honestly, decide where judgement actually belongs, and ship something the client can run and change themselves.