/ Digital Transformation · Case 01 · F&B Enterprise Software
F2Tech Asia
AI-Native Sales Information System replacing legacy ERP via Semantic Modeling + Knowledge Graphs in a SOA architecture.
The brief.
F2Tech Asia rebuilt enterprise software for F&B from an AI-first foundation. Instead of bolting AI on top of a traditional ERP, we developed and deployed an AI-Native Sales Information System — replacing the ERP layer entirely through Semantic Modeling and Knowledge Graph integration inside a Service-Oriented Architecture (SOA). As Head of AI & Digital Transformation since mid-2023, I owned the architectural roadmap, the AI integration model, and the financial measurement framework that proved every shipped feature against a P&L outcome. This technological disruption and execution capability were key drivers in securing $30M of seed funding and positioning the company as a pioneer in AI-native enterprise systems.
What was on the table.
Traditional ERP systems blocked F&B companies from real digital transformation. Legacy stacks throttled the speed and intelligence the modern F&B market demands — the only honest fix was to replace the ERP layer, not retrofit it.
How it shipped.
Designed an AI-driven digital transformation roadmap with built-in financial ROI tracking.
Replaced traditional ERP with an AI-Native Sales Information System — Semantic Modeling + Knowledge Graph integration on a SOA architecture.
Built a semantic platform enabling enterprise-wide AI adoption across product, sales, and operations.
Embedded RAG (Retrieval-Augmented Generation) on top of structured knowledge graphs for context-aware, cost-efficient querying.
Established ROI and financial measurement frameworks so every AI feature ships with a P&L story.
AI-native, not AI-retrofitted
The conceptual frame was simple and unfashionable: AI is a downstream beneficiary of governance. An ERP cannot host AI without paying a translation tax on every query. The only honest path is a semantic spine the AI sits on top of, not next to.
Semantics first, model second; the LLM is a feature of the semantic platform.
Every AI feature ships with a P&L story or it does not ship.
A clear glossary is the cheapest performance optimisation in the stack.
Semantic spine + Knowledge Graph + RAG retrieval
A canonical entity catalog (customer, order, product, outlet) anchored the logical model. A Knowledge Graph published the relationships between those entities and supported traversal queries that joins-and-views could not. RAG sat on top of curated semantic objects — never on raw documents — so retrieval returned governed answers, not chaos.
Conceptual model: 10–12 canonical entities owned by named stewards.
Logical model: Knowledge Graph with explicit relationships and contract tests.
Process logic: SOA services consume semantic objects, not raw tables; the spine is the API.
Retrieval: RAG indexes are built on curated entities, not the data lake.
SOA services, MDM spine, and an ROI framework
The physical implementation is a Service-Oriented Architecture that exposes the AI-Native Sales Information System as a set of stable services. An MDM spine governs master entities. A P&L measurement framework gates every shipped feature against a quarterly ROI review.
AI-Native Sales Information System replacing the legacy ERP layer.
SOA service catalog with versioned contracts.
MDM platform anchoring the canonical entity catalog.
RAG retrieval layer over the Knowledge Graph.
ROI / P&L framework reviewed quarterly with the leadership team.
“Retrofitting AI on legacy ERP is the most expensive way to ship slower. Native architecture is the only honest path.”
Numbers, verified.
- ROI achieved
- 75%
- Revenue per employee
- 2×
- Native AI integration
- 100%
- Digital investments
- Break-even
/ What we shipped
AI-driven transformation roadmap
AI-Native Sales Information System (ERP replacement)
Semantic Modeling + Knowledge Graph platform
SOA service catalog
RAG retrieval layer over the knowledge graph
ROI / P&L measurement framework
Enterprise data model + governance
What stuck.
The hardest sale was internal: convincing senior leadership to invest in architecture instead of features. Once the first business unit shipped on the new platform and reported 2× revenue per employee — and once investors funded the bet at seed stage — the roadmap stopped needing defense. Every subsequent quarter was a question of sequencing, not justification.