Natural Language Processing (NLP)

Natural language processing is the work of getting software to handle human language — which is harder than it sounds, because human language is a mess.

"Can you get the door?" is a request, not a question about ability. "That's just great" might be delight or fury. "Book a flight to Newcastle" means Newcastle upon Tyne to almost everyone and Newcastle-under-Lyme to somebody. People navigate this without noticing. Software has to be taught.

NLP covers everything involved in that: working out what words mean in context, what someone is asking for, how they feel about it, and what to say back.

What NLP actually does

The umbrella covers several distinct jobs:

• Understanding intent — turning "my order still hasn't turned up" into intent: delivery_chase, order_status: late
• Extracting entities — pulling names, dates, amounts and addresses out of an email or a PDF
• Sentiment analysis — is this customer annoyed, and how annoyed
• Translation — across languages, in real time
• Speech to text — turning what someone said into something searchable

NLP and LLMs

Large language models transformed NLP so thoroughly that people now use the words interchangeably. They aren't the same thing. NLP is the problem; LLMs are the current best tool — and often an expensive one.

This matters commercially. If you need to route enquiries into five categories, a small classifier trained on your own tickets will be faster, cheaper and more predictable than calling an LLM API for every message. Reach for the LLM when the task genuinely needs open-ended language, not because it's the technology in the headlines.

Why it matters for your business

Most of what your business knows is trapped in language — emails, tickets, call recordings, contracts, reviews, PDFs. It's the largest asset most companies can't query. NLP is how that becomes searchable, sortable, and countable.

The practical wins are usually unglamorous: triage that used to eat an hour every morning, contract data that used to be typed in by hand, a support queue that sorts itself before anyone opens it.

How we use NLP

We built Audico, a generative AI voice platform working across 40+ languages, from concept through to international commercialisation — freeing up 12 hours per week per staff member at RAFA care homes. Igloo runs custom NLP pipelines over journaling data to surface mental health insights for practitioners. BaaBaa, currently in development, does real-time multilingual translation across 80+ languages.

The common thread: language people were already producing, turned into something the business can act on.