kgbuilder.dev
in
INDUSTRIES · CROP PROTECTION

Knowledge graphs for crop protection companies

A crop protection portfolio is a web of connections: products, active substances, authorised crops and pests, doses and timings, country by country, each changing on its own schedule. Most of it lives in registers, labels and dossiers. A knowledge graph connects it, so a question that takes a person a day of cross-checking takes one query, with the label or register entry behind every answer.

THE PROBLEM

  • A substance review hits many products at once. Active substances are approved at EU level under Regulation (EC) No 1107/2009, while products are authorised country by country. When an approval is not renewed, the useful questions are about the portfolio: which of our products contain the substance, in which countries, for which crop-and-pest uses, and what is left to recommend instead.
  • The same product means different things in different countries. Authorised crops, pests, doses, timings and intervals differ by market and change with each label update. Answers for agronomists and sales teams have to come from the current authorised use, not from a brochure.
  • The knowledge sits in documents. Registration dossiers, labels, safety data sheets, trial reports and technical notes each hold part of the answer. Finding it depends on who knows where to look.
  • Resistance advice needs the chemistry. Recommending a rotation means knowing each product's mode-of-action group, which is rarely in the same system as the authorised uses.

WHAT A GRAPH ADDS

A graph stores each product with its active substances, and each authorised use as a connection to a crop, a pest, a country, a dose and a timing window, with the source document and date on it. Crops and pests are linked to EPPO codes, so names match across countries and languages. Mode-of-action groups and hazard classes sit on the substances. With that in place, questions follow the connections:

  • Exposure: which uses depend on a substance under review, and which of them have no alternative.
  • Substitution: for a crop and pest, which authorised products remain, grouped by mode of action.
  • Change: what changed in a national register since the last update, for your products and for competitors'.

Every answer can cite the register entry or label it came from.

WORKING DEMO

The Polish plant-protection register as a Neo4j graph: registered products, active substances, crops, pests, application windows and doses, linked to EPPO codes and mode-of-action groups, with an assistant that answers questions in plain language.

When the EU decided not to renew the approval of flufenacet (Regulation (EU) 2025/910), one query on this graph showed that it was in 45 products registered in Poland and that 36 crop-and-pest uses would be left with no other registered product, all in cereals.

Open the demo - sor.kgbuilder.dev ↗

WHAT A PILOT WOULD ANSWER

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

  1. Which of our products contain an active substance whose approval expires in the next 24 months, and which crop-and-pest uses would lose their only authorised product?
  2. For crop X and pest Y in country Z, which of our products are authorised, at what dose and timing, according to the current label?
  3. Which competitor products are authorised for the same uses as our product P?
  4. Which uses in our portfolio rely on products from a single mode-of-action group?
  5. What changed in the national register for our products since the last quarter?

Each answer names the register entry or label behind it. A domain expert on your side checks the answers against those sources, and that check is the result of the pilot.

WHEN A GRAPH IS NOT THE ANSWER

If most questions are about one product's label, good document search is enough. If they are about sales volumes and margins, that is a job for the data warehouse. A graph earns its cost when questions cross products, substances, uses and countries at once. When not to build a knowledge graph covers the test.

READ MORE

Start with a 2-3 day review↗Check whether your use case is ready↗