Sales Process Optimisation

How do you turn five undocumented sales workflows into a coordinated agent team that finds, validates, and converts the right buyers?

Problem Statement​

This PRD defines the orchestration layer that coordinates specialized sales agents (starting with the Sales Dev Agent) across the full sales lifecycle. It sits between the individual agents (Horizon 1) and the Trust Commerce protocol (Horizon 3).

When a business has built a CRM and RFP tool but has zero paying customers, the founder struggles to systematically find, validate, and convert prospects who match their ICP because sales processes are undocumented, unmeasured, and disconnected — each workflow operates in isolation without feedback loops, and ICP validation requires human judgment that doesn't scale.

This costs every week without a systematic sales process as a week of zero pipeline. The CRM has 23 contacts, 1 deal at $0, zero activities logged. Nine algorithms sit dormant. The Sales Dev Agent is a component within this system — it handles lead generation and outreach. This PRD orchestrates the full team.

Currently: Manual LinkedIn browsing, gut feeling for qualification, copy-paste cold emails, no tracking of what converts.

Classification: Complex — no single cause-effect. Requires experimentation with agent workflows across multiple sales stages.

Priority Score​

Three Horizons​

HORIZON 1: PROVE IT WORKS (Now → 90 days)
┌─────────────────────────────────────────────────┐
│  SALES DEV AGENT                                 │
│  Find → Score → Draft → Sequence → Learn         │
│  Target: 10 scored prospects/week                │
│  Proof: Outreach generates replies               │
│  Stack: CRM + Algorithms + Adapters              │
└───────────────┬─────────────────────────────────┘
                │ feeds into
                ▼
HORIZON 2: ORCHESTRATE THE TEAM (90 days → 6 months)   ← THIS PRD
┌─────────────────────────────────────────────────┐
│  SALES PROCESS OPTIMISATION                      │
│  Lead Gen → Qualification → Conversion → Lifecycle│
│  Target: Repeatable pipeline, multi-client        │
│  Proof: Sneakers Media HRV pilot, 2nd client     │
│  Stack: Agent orchestration + ICP engine +        │
│         Workflow state machines + Feedback loops   │
└───────────────┬─────────────────────────────────┘
                │ generates the data for
                ▼
HORIZON 3: MAKE IT A PROTOCOL (6 months → 18 months)
┌─────────────────────────────────────────────────┐
│  TRUST COMMERCE                                  │
│  Coach → Recommend → Convert → Reputation        │
│  Target: Trust network with on-chain settlement  │
│  Proof: >15% referral conversion, portable rep   │
│  Stack: Sui objects + PTBs + zkLogin + Privacy   │
└─────────────────────────────────────────────────┘

The Job​

When the CRM and RFP tools are built but dormant, help the business turn documented sales workflows into executable agent procedures and validate that prospects match ICP before investing human time — so discovery calls happen with the right people and the pipeline fills with qualified revenue.

The Core Hard Problem: ICP Validation​

Lead generation is automatable. The hard part is knowing which leads are worth pursuing. A prospect can match every demographic filter and still not be a buyer.

The gap between "looks like a fit on paper" and "will actually buy" is where most sales time is wasted. This PRD exists to close that gap. The 4-layer ICP validation model is the flagship capability — see ICP Validation Engine below.

ICP Validation Engine​

This is SPO's genuine edge. The 4-layer model is what other PRDs reference (the Sales Dev Agent uses this for lead scoring, Trust Commerce will use qualification outcomes as reputation seed data).

The system must score across all four layers. Firmographic scoring already exists. This PRD builds the behavioral, psycho-logic, and investment layers.

Scoring Algorithm​

Each layer produces a 0-100 score. Composite ICP fit = weighted blend:

Learning loop: Every qualification outcome (accepted call, rejected, ghosted, closed) updates the model weights. Minimum 50 decisions before trusting composite scores. The Sales Dev Agent's outreach data feeds this loop.

Demand-Side Jobs​

Job 1: Operationalize Sales Workflows​

Situation: Six sales workflows exist as knowledge base documentation. None are connected to the CRM. None have measurable outputs. An agent can't execute what isn't instrumented.

Features:

• Workflow state machine per sales process (lead gen → qualification → funnel → lifecycle)
• Trigger definitions in CRM (event-based, schedule-based, threshold-based)
• Stage transition rules with evidence requirements
• Metrics collection at each stage (conversion rates, duration, quality scores)
• Agent assignment per stage (which agent handles which workflow)
• Human handoff points defined and enforced
• Feedback loops between workflows (qualification insights → lead gen refinement)

Job 2: Validate Prospects Against ICP at Scale​

Situation: The founder has an ICP document with psycho-logic profiles. But matching real prospects to that archetype requires judgment that takes 30-60 minutes per prospect. With 20+ prospects per week, the bottleneck is validation, not generation.

Features:

• Multi-layer ICP scoring engine (4 layers: firmographic, behavioral, psycho-logic, investment)
• Firmographic scoring from company data (industry, size, geography, tech stack)
• Behavioral scoring from engagement signals (email opens, content downloads, response speed)
• Psycho-logic scoring from conversation analysis (stated vs actual needs, objection patterns)
• Investment scoring from interaction quality (do they invest time, share data, introduce stakeholders?)
• Composite ICP fit score with evidence trail
• Auto-routing: hot → discovery call, warm → nurture, cold → disqualify
• ICP refinement loop: every qualification outcome updates the model

Job 3: Coordinate Multiple Sales Agents​

Situation: The Sales Dev Agent handles outreach. The RFP agent handles bid response. But nobody coordinates the handoff between prospecting, qualification, and conversion. Leads fall through the cracks between agents.

Features:

• Agent role definitions mapped to Sales Work Chart AI Transformation Spectrum
• Shared CRM state: all agents read/write to the same contact, deal, and activity records
• A2A handoff protocol: structured hand-off between agents with context summary
• Orchestration layer: workflow engine dispatches tasks to appropriate agent based on stage
• Conflict resolution: when two agents want to act on the same prospect, priority rules apply
• Human override: any agent action can be paused, modified, or cancelled by the human

The Agent Team​

Each documented sales workflow becomes procedural memory for a specialized agent. The team follows the Sales Work Chart AI Transformation Spectrum.

LEAD GEN AGENT          QUALIFICATION AGENT        CONVERSION SUPPORT
────────────────        ────────────────────       ─────────────────
Research prospects      Score ICP fit              RFP auto-fill
Multi-channel outreach  Validate psycho-logic      Proposal drafting
Source enrichment       Route: pursue/nurture/cut  Deck generation
       │                        │                         │
       └────── CRM (shared state) ──── HUMAN ────────────┘
                    │              Discovery calls
                    │              Deal negotiation
                    │              Relationship + trust
                    ↓
            LIFECYCLE AGENT
            ────────────────
            Onboarding automation
            Health scoring
            Churn prediction
            Expansion signals

Composition Inventory​

What already exists vs what needs building:

Build ratio: ~75% composition, ~25% new code.

The new code is:

1. ICP validation engine (multi-layer scoring)
2. Workflow state machine (connecting KB to CRM stages)
3. Agent orchestration layer (dispatch, handoff, conflict resolution)
4. Process metrics collection (conversion rates per stage)

Success Criteria​

Functional​

Outcome​

Risk + Kill Signal​

Kill signal: If 50 agent-scored leads produce fewer than 3 discovery calls over 60 days, the ICP model is wrong. Diagnose: is it firmographic (wrong companies), behavioral (wrong signals), or psycho-logic (right company, wrong person)?

Build Sequence​

Phase 0: Foundation (depends on Sales CRM Sprint 0 completing)​

Wire the dormant algorithms that feed the agent team:

1. Sales Forecasting → lead scoring
2. Compound Rate → engagement velocity
3. Explore-Exploit → channel optimization

This is already in the Sales CRM PRD Sprint 0. Must complete first.

Phase 1: Single Agent (Lead Gen) — Minimum Viable Orchestration​

Build ONE agent executing ONE workflow (lead generation) end-to-end. The minimum viable orchestration is: Sales Dev Agent (Lead Gen) + manual human qualification step + CRM logging. This proves the workflow state machine works before adding more agents.

1. Wire Lead Generation workflow to CRM triggers
2. Build prospect research service (compose from DocumentIntelligence)
3. Build firmographic ICP scoring (layer 1 of 4)
4. Wire outreach drafting with human review gate
5. Track: prospects researched, outreach sent, response rate
6. Deploy for both internal dogfood (construction/solar) and Sneakers Media HRV pilot

Acceptance: Agent produces 20 scored prospects/week. Human reviews and sends outreach. Response rate tracked. Don't build the orchestra before proving the instrument.

Phase 2: ICP Validation Engine​

Add the hard capability:

1. Behavioral scoring layer (email opens, content engagement, response quality)
2. Psycho-logic scoring layer (conversation analysis, objection pattern matching)
3. Investment scoring layer (time invested, stakeholder introductions, data sharing)
4. Composite ICP fit score with routing logic (hot/warm/cold)
5. ICP refinement loop (qualification outcomes update scoring weights)

Acceptance: Multi-layer scoring produces ICP fit scores that correlate with actual qualification outcomes (r > 0.5 after 50 decisions).

Phase 3: Agent Team Orchestration​

Add qualification agent + lifecycle agent:

1. Qualification agent reads lead gen output, applies BANT+ framework
2. A2A handoff from lead gen → qualification → conversion support
3. Shared CRM state with conflict resolution
4. Lifecycle agent for post-close customer health monitoring
5. Process metrics dashboard (conversion rates per stage, agent performance)

Acceptance: Lead flows through full pipeline (gen → qualify → funnel → close) with agent handoffs logged. Process metrics visible on dashboard.

Phase 4: Optimization Loop​

1. Explore-Exploit optimizes channels and messaging
2. ICP model retrains on qualification outcomes
3. Workflow state machine adapts (skip stages for hot leads, add stages for complex deals)
4. Process maturity moves from "Draft" to "Approved" for all workflows

Acceptance: Agent team improves over time — week 8 outperforms week 1 on response rate, qualification accuracy, and pipeline value.

North Star​

First paying customer acquired through an agent-optimized sales process.

Not through personal network. Not through luck. Through a documented, measured, repeatable process that an agent team executed and a human closed.

Engineering Brief​

This PRD is a process intelligence capability. It sits between the Sales KB (how we plan to operate) and the engineering team (what to build).

The KB pages define WHAT the workflows are. This PRD defines HOW to implement them in the platform. The commissioning model tracks WHETHER what was built matches what was specced.

SALES KB (upgraded)                  THIS PRD (new)                       ENGINEERING
────────────────                     ──────────                           ──────────
Lead Generation workflow     →  Workflow state machine spec      →  CRM trigger definitions
Lead Qualification workflow  →  ICP validation engine spec       →  Multi-layer scoring
Funnel Engineering workflow  →  Pipeline metrics spec            →  Dashboard + analytics
Customer Lifecycle workflow  →  Health scoring spec              →  Churn prediction model
Advertising workflow         →  Channel optimization spec        →  Explore-exploit integration
Sales Activities index       →  Agent team orchestration spec    →  A2A handoff protocol

Key dependency: Sales CRM & RFP Sprint 0 must complete first (wire 3 dormant algorithms + Stripe payments). This PRD builds ON the wired algorithms, not alongside them.

Relationship to existing PRDs: See Relationship to Other PRDs for the full three-horizon architecture and data flow bridges.

External Validation: Sneakers Media HRV Pilot​

SPO isn't just internal tooling — it's the BOaaS product deployed for clients. The Sneakers Media HRV (Heat Recovery Ventilation) campaign is the first external deployment:

The Qualification Playbook concept — per-client ICP configuration — validates that the system generalizes beyond one vertical. If SPO can serve both construction EPC and residential ventilation with the same orchestration layer, it's a platform capability.

Agent Boundary Permissions​

The agent team shares CRM state, but structural boundaries prevent agents from overstepping. Following the Trust Architecture hexagonal principle: the orchestration layer defines ports, not trust in agent behavior.

Conflict resolution: When two agents want to act on the same prospect, the orchestration layer applies priority rules: Qualification trumps Lead Gen (don't outreach to a prospect mid-qualification), Conversion trumps Qualification (don't re-qualify a prospect in proposal stage).

Relationship to Other PRDs​

HORIZON 1: PROVE IT WORKS (Now → 90 days)
  Sales Dev Agent — the Lead Gen component
        │ feeds into
        ▼
HORIZON 2: ORCHESTRATE THE TEAM (90 days → 6 months)   ← THIS PRD
  Sales Process Optimisation — coordination + ICP engine
        │ generates data for
        ▼
HORIZON 3: MAKE IT A PROTOCOL (6 months → 18 months)
  Trust Commerce — on-chain reputation from SPO outcomes

Data flows between horizons:

• H1 → H2: The Sales Dev Agent generates outreach data, reply patterns, and qualification outcomes. SPO's Learning Agent consumes this to improve ICP scoring across all agents. Sales Dev Agent is the first test subject for SPO's orchestration — dogfooding the coordination layer.
• H2 → H3: SPO generates three things Trust Commerce needs: (1) qualification proof — structured evidence from the 4-layer ICP model, becoming the basis for on-chain reputation; (2) satisfaction outcomes — did the qualified lead buy and were they satisfied (Reputation Delta input); (3) relationship data — who recommended whom, through what channel, with what result (trust graph seed data).
• H3 → H1: Once Trust Commerce is live, warm referral leads flow back into the Sales Dev Agent pipeline, pre-qualified by the trust graph.

Direct dependencies:

• Sales Dev Agent — Component. The Lead Gen agent in this team.
• Sales CRM & RFP — The tool. This PRD makes it smart.
• Agent Platform — The infrastructure. This PRD is a consumer.
• Trust Commerce — The protocol. This PRD generates the data it needs.

Mycelium Capability​

Sales Process Optimisation is a process intelligence capability. It transforms documented sales workflows into executable agent procedures with measurable outcomes. Currently built for construction and solar EPC sales, with the Sneakers Media HRV pilot as the first external deployment. The pattern — workflow → state machine → agent → metrics → optimization — applies to any B2B sales function.

Currently Growing In: stackmates

Context​

• Sales Activities — The 6 documented sales workflows (KB source)
• Sales Work Chart — Human/AI split for every sales job
• Sales CRM & RFP — The tool that executes these workflows
• Sales Dev Agent — The first agent in the team
• ICP Framework — Target definition workflow
• Process Optimisation — The methodology for documenting and improving processes
• Agent Platform — Agent infrastructure (identity, memory, comms)
• Phygital Mycelium — The capability catalogue