METHODOLOGY
Six phases. Repeatable. Measurable.
We apply a structured six-phase agent engineering methodology. Each phase has an objective, a deliverable, a typical duration. No ambiguity.
Identification
Diagnose friction and high-impact workflows.
Objective
Identify the ERP friction that costs your organisation most, and rank by potential ROI.
Deliverables
- ›Workflow map
- ›Top 10 prioritised friction points
- ›Rough ROI estimate
Typical duration
1 to 2 weeks
Roles
Executive sponsor, key users, KHMERP architect
"We discovered that 60% of our accounting team's time went into 3 identifiable friction points."
Definition
Scope priority use cases and quantify target ROI.
Objective
Select 1 to 3 use cases to attack first, and compute their precise target ROI.
Deliverables
- ›Signable scoping note
- ›Target ROI per use case
- ›Success criteria
Typical duration
1 week
Roles
CFO / CHRO / CIO, executive sponsor
"The costed scoping unlocked our budget in two weeks."
Representation
Model processes and data flows.
Objective
Visually represent existing and target processes, and data flows between ERP, agents and users.
Deliverables
- ›BPMN as-is / to-be diagrams
- ›Data flow diagrams
- ›Integration map
Typical duration
1 to 2 weeks
Roles
Solution architect, key users
"Seeing the process drawn aligned the whole team in one meeting."
Specification
Specify agents — skills, memory, prompts.
Objective
Design each agent in detail: capabilities, memory, prompts, guardrails, integrations.
Deliverables
- ›Agent design docs
- ›Versioned prompt catalog
- ›Guardrail policy
Typical duration
1 to 2 weeks per agent
Roles
Prompt engineer, agent architect
"The agent design doc is as detailed as an API spec."
Configuration
Configure and integrate in the ERP, user testing.
Objective
Build the agent, integrate it into the ERP, test it with real users in sandbox.
Deliverables
- ›Integrated and tested agent
- ›UAT test plan
- ›User documentation
Typical duration
2 to 4 weeks per agent
Roles
Developers, key users, QA
"Users validated the agent on the first test session."
Instantiation
Deploy, train users, continuously monitor.
Objective
Go to production, train teams, monitor usage and quality, improve continuously.
Deliverables
- ›Agent in production
- ›Monitoring dashboard
- ›Continuous improvement plan
Typical duration
1 week then monthly Run
Roles
Operators, support, key users
"In three months, the agent handles 80% of cases without human intervention."