We use AI and machine learning to help build and enrich our own dataset of UK, German and Irish private companies.
We've been using machine learning for years, and many of our processes involve a blend of automation and manual curation - the skill is knowing which parts of a task suit each.
We use traditional machine learning and natural language techniques, simple statistical methods, and generative AI, depending on the problem at hand. We're well-placed to take advantage of these technologies because of our years of experience understanding the intricacies of company data and producing hand-curated datasets. Automation speeds up the parts of the process that suit it, while our Data team continues to handle the many parts that need human judgement.
We use AI and ML in two main ways:
1. Background data enrichment
We apply machine learning and language models to our own source data - public filings, company websites, news and registry information – to classify companies, structure and extract information, and keep the platform accurate and current. This runs entirely on Beauhurst's side, on data about companies. Examples of this include:
Record linking. We use machine learning to match and link records. For example, finding the best website for a company, judging whether two filings referring to people with similar names are, in fact, the same person, or matching a news article to the right company record. These matches are then checked and refined by our Data team rather than taken at face value.
Automated data extraction and enrichment. We use machine learning to extract or generate structured information. For example, categorising companies with an industry or buzzword, finding the right trading address from a company website, categorising news articles into general topics, or writing a description of a business.
Decision support for our Data team. We use machine learning and AI to help our domain experts with curation and quality assurance. For example, we use AI to discard irrelevant news articles from our announced transaction curation process, freeing our analysts to focus on the transactions that matter.
2. In-platform AI features
The platform offers AI-assisted features that a user can use to help them discover and understand companies more easily.
For example, Natural Language Search turns a plain-English question into a structured search of our data, and AI Lenses produce evidence-backed summaries of a company in which every point is traceable to its source. Preset lenses (Highlights, Growth Trajectory, Financial Health, Innovation & Impact, People) are available out of the box, or you can create a Custom Lens by typing any company-research question and saving it for repeated use, including sharing it across your team.
Our Claude MCP connector, allows your team to query Beauhurst data directly within your own AI tools and systems (for example Claude, ChatGPT, or Copilot). This is entirely set up and controlled by you; our AI is not deployed in your environment.
Data protection and security
Our AI works on our data, not yours. If you choose to connect our API, MCP server or CRM integration, data you pass through those connections is processed solely to operate them, and where Beauhurst sits inside a workflow of yours, it is because you have connected it there. In every case, your information is never used, in any form that could identify you, your users or your organisation, to train, fine-tune or improve any model (ours or a third party’s).
Where platform features rely on a third-party model, that provider is contractually prevented from retaining your inputs or outputs or using them to train or improve any model, and zero-data retention is enforced for these requests, so your prompts and the results are not stored by the provider.
Your Confidential Information, and the content your team creates or uploads to the platform, are never used to train, fine-tune or improve any AI or machine-learning model – ours or a third party’s – in any form, including anonymised. Usage data, such as the queries your users run, may be used to evaluate and improve our features and models only once it has been anonymised and aggregated so that it cannot identify you.