Discovery & Planning
Week 1Key activities
- Business objectives
- MES functional areas
- Stakeholders mapped
- Discovery approach planned
Deliverables
- MES Functional Map
- Stakeholder List
- Discovery Plan
- Project Charter
Our Methodology
The MES + AI integration framework built on real manufacturing operations.
10 phases. 10 weeks. One trusted AI intelligence layer.
Why This Methodology Exists
The biggest mistake in AI + MES projects is assuming AI will automatically understand your manufacturing database and business logic.
It will not.
Here’s what usually happens: consultants connect AI to raw MES tables. AI returns answers based on table names and column patterns. Answers don’t match how your engineers know the system actually works. Engineers stop trusting the system. The project dies.
Manufacturing databases were not built for AI. They were built for MES applications, with business logic hidden in stored procedures and years of undocumented customizations. Raw tables contain the data. They don’t contain context.
Before AI answers manufacturing questions correctly, it needs to understand your database semantically: what each table represents, which joins are valid, what business logic is hidden in procedures, and which answers matter.
This methodology builds that understanding. Here’s how:
The 10-Phase Framework
From unknown database to trusted intelligence layer.
Building a trusted AI foundation for MES using SQL Server and Application Server. Ten phased weeks from discovery through pilot rollout.
MES + AI Integration Roadmap
Scroll horizontally · all ten phases
Stakeholder Discovery
Weeks 1-2Key activities
- SMEs, developers, DBAs
- Key reports & KPIs
- Business rules documented
Deliverables
- Interview Notes
- Process flows
- KPI Catalog
- Business Rules List
Database Inventory
Weeks 2-3Key activities
- Tables, views, procedures
- Triggers, jobs, indexes
- Classification by role
- Identify key databases
Deliverables
- Database Inventory
- Object Catalog
- Table Classification
- Data Sources List
Data Model & Relationships
Weeks 3-4Key activities
- Keys & relationships
- Documented vs inferred links
- ER diagrams & genealogy
Deliverables
- ER Diagram
- Relationship Matrix
- Key Tables
- Model documentation
Application Server Analysis
Weeks 4-5Key activities
- App architecture review
- APIs & services
- Jobs, logs, configuration
Deliverables
- Architecture diagram
- Job & Process List
- Hidden logic doc
- API inventory
AI Knowledge Preparation
Weeks 5-6Key activities
- Data dictionary
- Columns, values, definitions
- Joins, filters, rules
- Sensitive data flagged
Deliverables
- Data Dictionary
- Business Definitions
- Join & Filter Guide
- Valid Values Catalog
Trusted Data & Semantic Layer
Weeks 6-7Key activities
- AI-approved views only
- Trusted query definitions
- Performance tuned
Deliverables
- Approved views
- Semantic layer spec
- Query catalog
- Performance notes
Security & Governance
Weeks 7-8Key activities
- Read-only access model
- Roles & permissions
- Privacy & audit logging
- Approval workflow
Deliverables
- Security model
- Access matrix
- Audit plan
- Governance policy
AI Integration & Pilot Build
Weeks 8-9Key activities
- MCP / API bridge
- Semantic layer wired
- Approved data only
- Pilot use cases live
Deliverables
- Integration architecture
- MCP / API setup
- Pilot configuration
- Prompt library
Validation & Pilot Rollout
Weeks 9-10Key activities
- Test question library
- Answers vs trusted reports
- SME sign-off gates
Deliverables
- Validation report
- Accuracy metrics
- Lessons learned
- Pilot rollout plan
Cross-cutting foundations, every phase
-
Stakeholder collaboration
Continuous SMEs, DBAs, developers & users.
-
Documentation
Central repository for every finding & decision.
-
Quality assurance
Review deliverables against standards.
-
Security by design
Privacy & access from day one.
-
Performance
Models & views tuned for production scale.
-
Continuous improvement
Iterate with feedback loops.
Outcome
A trusted AI foundation for MES.
Semantic understanding of business meaning, structure, relationships, and logic.
Ready for accurate, reliable insights.
Phase Detail
Inside every phase.
What we do. Why it matters. What you walk away with.
What we do
Before a single database query, we map business problems you need solved (yield visibility, equipment downtime, cycle time), which MES functional areas matter most, who the stakeholders are, and how discovery will run.
Why it matters
Most AI projects jump straight to connecting the database. That fails because nobody asked what questions your team actually needs answered. This is the only phase where we are not touching the database. We understand your business first.
Deliverables
- MES Functional Map
- Stakeholder List
- Discovery Plan
- Project Charter
What we do
We interview SMEs, developers, DBAs, and business users. We identify key reports, KPIs, and business rules. We capture the tribal knowledge that lives in people’s heads, not in any documentation.
Why it matters
The most important information about your MES is not in the database. It is in the people who use it every day. We find them and document them.
Deliverables
- Interview Notes
- Business Process Flows
- KPI Catalog
- Business Rules List
What we do
Using MCP-connected AI discovery, we inventory every table, view, stored procedure, trigger, and SQL job. We classify table types (master, transaction, history, summary) and identify key databases. No manual queries required.
Why it matters
Most MES databases have 400+ tables with no documentation. On typical engagements we fully inventory schemas in weeks, not months. This phase replaces slow manual discovery with a systematic, governed process.
Deliverables
- Database Inventory
- Object Catalog
- Table Classification
- Data Sources List
What we do
We identify primary and foreign keys, detect documented and undocumented relationships, build ER diagrams, and map lot genealogy and WIP flow. Many MES databases have no formal foreign keys, so hidden relationships must be inferred from stored procedures, naming conventions, and application traces.
Why it matters
AI cannot join tables it does not understand. This phase gives the AI the relational map it needs to answer multi-table questions correctly.
Deliverables
- ER Diagram
- Relationship Matrix
- Key Table List
- Data Model Documentation
What we do
We review application architecture, analyze APIs and services, review configuration files, analyze scheduled jobs, and review application logs. We identify hidden calculations and status transitions that the database alone does not reveal.
Why it matters
MES logic does not live entirely in the database. Nightly jobs transform data. Middleware recalculates values. If AI does not know this, it will answer questions about stale or transformed data incorrectly.
Deliverables
- App Architecture Diagram
- Job & Process List
- Hidden Logic Document
- API Inventory
What we do
We create the AI data dictionary, documenting table purpose, columns, valid values, business definitions, joins, filters, and rules. We identify sensitive data and build the trusted query catalog that governs every AI interaction.
Why it matters
This is the semantic layer between raw data and AI understanding. Without it, AI answers are unreliable. With it, AI consistently answers manufacturing questions correctly.
Deliverables
- AI Data Dictionary
- Business Definitions
- Join & Filter Guide
- Valid Values Catalog
What we do
We design AI-approved views, create the semantic layer, define trusted queries, build the query catalog, and optimize for performance. No AI query ever touches raw MES tables directly.
Why it matters
This is the architectural decision that separates AI implementations that work from ones that fail. Governed views mean governed answers. Raw table access means hallucinations and trust collapse.
Deliverables
- Approved Views (vw_AI_WIP_STATUS, vw_AI_YIELD_SUMMARY, etc.)
- Semantic Layer Definition
- Query Catalog
- Performance Notes
What we do
We define read-only access models, create SQL roles and permissions, define data privacy rules, set up audit logging, and establish the approval workflow for new AI use cases.
Why it matters
Manufacturing data is sensitive. Yield data, equipment performance, and production schedules are competitive intelligence. Governance is not an afterthought. It is built from day one.
Deliverables
- Security Model
- Access Matrix
- Audit Plan
- Governance Policy
What we do
We build the MCP/API layer, connect the AI assistant to the semantic layer, configure prompts and tools, restrict AI to approved data, and build the first pilot use cases.
Why it matters
This is where the methodology becomes a working product. Engineers ask questions in plain English. They get answers they trust, with full lineage back to the source data.
Deliverables
- Integration Architecture
- MCP/API Setup
- Pilot Use Case Configuration
- Prompt Library
What we do
We create test questions, validate AI answers against known reports, measure accuracy, refine views and rules, and get SME sign-off. No AI goes to manufacturing users without human expert validation.
Why it matters
One wrong yield number causes a bad production decision. Trust, once lost in manufacturing, is almost impossible to rebuild. Validation is not optional.
Deliverables
- Validation Report
- Accuracy Metrics
- Lessons Learned
- Pilot Rollout Plan
Outcome
A trusted AI foundation for MES.
Semantic understanding of business meaning, structure, relationships, and logic.
Ready for accurate, reliable insights.