Named Entity Recognition

Named Entity Recognition (NER) is a machine learning technique that automatically recognises and categorises specific information from unstructured text. The system identifies names of people, companies, locations, dates, monetary amounts and other predefined entities. For small and medium-sized companies, this means you can search through large volumes of customer communications, contracts or reviews without having to read every piece of text manually.

How Named Entity Recognition performs text analysis

A NER system scans text word by word and compares patterns against a trained database. The model recognises contextual cues such as capitalisation, word order and grammatical structure. A sentence such as “Jan Jansen from Bakkerij De Korenbloem in Utrecht ordered 500 pounds’ worth of ingredients” yields four entities: a person’s name, a company name, a location and a monetary amount. The system automatically labels each fragment with the correct category. United Kingdom NER models are specifically trained on our language, because ‘de’ can be an entity marker in English but is an article in United Kingdom. Accuracy depends on the quality of the training data and the complexity of your texts. For standard business communication, modern systems achieve an accuracy of 85 to 95 per cent.

Why Named Entity Recognition is now practically usable

NER has been around since the 1990s, but for a long time it was too expensive and too complex for SMEs. The breakthrough came with open-source libraries such as spaCy and Hugging Face Transformers, which offer ready-to-use Dutch models. You no longer need a data centre: a standard cloud server is sufficient. Companies now use NER for customer segmentation, contract analysis and trend detection in feedback. An online shop with thousands of reviews can automatically identify which product names, brands or locations customers mention. An accountancy firm extracts company names and VAT numbers from scanned invoices. The technology works best with structured texts such as emails, forms or reports, and less well with informal chat or dialect.

What Named Entity Recognition delivers in combination with other AI tools

NER is rarely a stand-alone solution. You combine it with API integrations to load extracted data directly into your CRM or ERP. An insurer can automatically extract policy numbers, names and claim amounts from claims and forward them to the relevant system. A recruitment agency extracts job titles, company names and locations from application letters to match candidates with vacancies. In practice, we see SMEs using NER as the first step in a wider automation process. Monkey Vision assists in the design of such workflows through bespoke development and AI integrations, linking NER to n8n workflows or bespoke dashboards. The investment only pays off if you process at least 100 documents per month; otherwise, manual work remains more efficient.

Named Entity Recognition is used in practice whenever you need to search through large amounts of unstructured text for specific information. Think of customer communications, contracts, news articles or internal reports. Below are four specific situations in which small and medium-sized companies use NER to save time and gain better insights from existing data.

Automatic customer segmentation based on email traffic

A B2B service provider receives hundreds of enquiries per month via email and contact forms. With NER, you can automatically extract company names, job titles, locations and industry indicators from those messages. Your CRM system immediately tags each lead with the correct labels, so your sales team knows whether it’s an SME client in the construction sector or a multinational in the financial sector. An accountancy firm in Arnhem uses this to separate enquiries from self-employed professionals from those of limited companies, which has reduced the turnaround time for quotes by 40 per cent. All you need is an API connection between your email inbox and your CRM. The initial setup takes around 12 hours of development time; after that, the system runs autonomously. Please note: Dutch models work better than English-language ones for typically Dutch entities such as Companies House numbers or town names.

Contract analysis and compliance checks

Legal departments and compliance teams often need to extract the same information from dozens of contracts: party names, contract terms, termination clauses, penalty amounts. NER scans PDFs or Word documents and automatically extracts these fields into a spreadsheet. A property manager with 200 tenancy agreements can thus compile an overview of all expiry dates and indexation clauses in a single afternoon, a task that would take two weeks to complete manually. The technology works well with standard contracts, but less so with highly non-standard wording or handwritten notes. You often combine NER with optical character recognition (OCR) if you have scanned paper contracts. Accuracy is around 90 per cent for typed text, which means you always need to manually check a sample. For companies with fewer than 50 contracts a year, the investment is usually not cost-effective.

Trend detection in client feedback and reviews

An online shop with 5,000 reviews per quarter wants to know which products, brands or product features customers mention spontaneously. NER automatically extracts product names, colour descriptions, material types and brand names from free-text reviews. You can immediately see that ‘waterproof’ appears 120 times in connection with a specific rucksack model, or that customers from Limburg mention ‘fast delivery’ more often than those from Friesland. A cycle shop in Utrecht discovered via NER that 30 per cent of negative reviews concerned a specific lock brand, which led to a change of supplier. This application also works for internal feedback: HR departments scan exit interviews for department names, management names and reasons for leaving. You can link NER to a content marketing strategy by identifying which search terms customers use in reviews and incorporating those terms into your product pages.

When is Named Entity Recognition the right choice, and when is it not?

NER is worthwhile if you process at least 100 similar documents per month and are looking for recurring entities such as names, amounts or dates. It does not work well with highly variable texts, dialects or creative content such as blogs. A sole trader who writes 10 quotes a month does not need NER. A recruitment agency that receives 500 CVs a week, however, does. Also important: NER recognises entities, but does not understand intent or sentiment. It sees “Apple” as a company name, but does not know whether the writer is being positive or negative. You should therefore combine NER with sentiment analysis or chatbots to get a complete picture. According to the Dutch Data Protection Authority, when processing personal data via NER, you must always enter into a data processing agreement with your AI supplier.