INDUSTRIES · SUPPLY CHAIN & MANUFACTURING

Knowledge graphs for supply chains and manufacturing

Products depend on parts, parts on materials, materials on plants, and plants sit in countries. Each system in a company usually sees one step of that chain: a supplier list, a bill of materials, a quality record. The risk and the answers live in how the steps connect. A knowledge graph puts those steps back into paths you can query.

THE PROBLEM

The supplier that stops production is rarely a direct one. Tier-1 suppliers are visible. The fragile node usually sits several tiers upstream: a single plant, a single material, a single country that many parts route through.
Traceability stops at one step. EU food law, for example, requires each business to know its direct supplier and direct customer (Regulation (EC) No 178/2002, Article 18). In the 2013 horsemeat case, rebuilding the full chain from those one-step records took investigators months.
Where-used and substitution questions cross systems. "Which finished products use this component, at any depth?" and "what can replace it?" need the bill of materials, approved alternates and supplier data together.
Not every record is equally true. A supplier's declaration, a scan and an audited certificate often end up as the same field. When something goes wrong, you need to know which links were checked.

WHAT A GRAPH ADDS

A graph models the chain as what it is: parts, materials, suppliers, plants and countries as nodes, with dependencies, bills of materials and approved alternates as relationships. Questions of unknown depth become one pattern instead of recursive joins rewritten for every new tier.

Chokepoints: graph algorithms such as betweenness centrality and articulation points find the nodes most paths route through, and the ones whose loss disconnects part of the network.
Exposure: which products contain material from a given source, however far upstream.
Substitution: which approved alternates exist for a part, and which of them avoid the same upstream dependency.
Evidence: each link can carry its source and status (declared, observed, verified), so risk can be ranked by how well it is evidenced.

WORKED EXAMPLES

There is no live supply-chain demo on this site yet. The closest working graph is the SEC 10-K filings graph, which extracts declared dependencies from annual reports: you can ask it which companies depend on TSMC and get the filing and section behind each answer. The patterns above are worked through on synthetic networks in the articles below. Bills of materials follow the same where-used and substitution patterns as the substitution engine for crop protection, but I have not built a production BOM graph, and a pilot would start by testing that fit on your data.

WHAT A PILOT WOULD ANSWER

Examples of control questions. In a real pilot they are replaced by yours during scoping.

Which finished products use component C, at any level of the bill of materials?
If supplier S or plant P stops delivering, which products are affected, and which of them have no approved alternate?
Which three nodes do most of our critical supply paths route through?
For part X, which approved alternates avoid the same upstream material or country?
Which links in the chain for product Y rest only on a supplier's declaration?

A domain expert on your side checks each answer against the source systems, and that check is the result of the pilot.

WHEN A GRAPH IS NOT THE ANSWER

If your questions stop at the first tier, your ERP already answers them. If supplier data is thin or has no shared identifiers, a graph will show gaps rather than risks, and the first job is data collection. Algorithms on an incomplete network produce confident-looking noise. When not to build a knowledge graph covers the test.

Start with a 2-3 day review↗

Check whether your use case is ready↗