GTM Workflow Optimization Spec

Analysis and redesign of GTM Agent workflows from Miro Board: design (uXjVH7iuOYc=)

Open Board in Miro
28 → 3
Research Nodes Reduced
3 → 1
HITL Loops Consolidated
4 Families
Collapsed Scenarios (~20)
3 Packs
Standardized Deliverables

Operational Shift: Monolithic vs Orchestrated

The core optimization in v2 moves from a high-overhead, redundant grid research model to a streamlined "Configure Once, Research Once, Deliver Phase-Specific Packs" pipeline. This drastically lowers token cost, eliminates human review fatigue, and introduces a first-class post-call workflow.

Workflow Gaps & Friction in v1

Redundant Research Grid (8x4 Layout) An exhaustive query matrix (8 scenario categories x 4 research topics) triggers 32 concurrent requests per run. Results overlap 70% of the time, leading to major latency (3+ mins) and token waste.
Repetitive Human-In-The-Loop Checkpoints Representatives must approve three consecutive review gates (Trends ➔ Insights ➔ Layouts) before any deliverable is built. This causes severe friction and screen-hopping.
Option Explosion on Run Execution Reps are presented with 27 individual output checkboxes when triggering a pipeline, creating severe choice fatigue.

Redesign Principles in v2

Adaptive Research Gating Dynamically maps scenario complexity to research depth (Light: 2-3 queries, Standard: 4-5, Deep: 6-8), bypassing heavy runs for simple cold-intro tasks.
Core Shared Research Engine Splits research outputs into immutable, structured JSON packages (Account Pack, Persona Pack, Opp Pack). These caches are queried once and reused in downstream tasks.
Single Consolidated Checkpoint Unifies brief edits, custom objection overrides, and format overrides into a single, comprehensive Review Panel before finalizing the run assets.