AI development and integration
Useful AI features added to real products where they solve a real problem, and an honest answer when they don’t.
Overview
Large language models can do things that were hard for software a few years ago. They can read messy text, summarise long documents, draft replies and answer questions about your own information.
They can also be slow, expensive and wrong in ways that are hard to predict. So we treat AI as one tool among many. We use it when it’s clearly the best way to solve the problem, and we design around its weaknesses.
This might be you if…
- Your team reads and sorts a lot of unstructured text: emails, forms, documents or notes.
- People spend time searching for answers that already exist somewhere in your documents.
- You want AI in your product but aren’t sure where it would genuinely help.
- You have an AI prototype that works in demos but not reliably in real use.
Typical projects
Document and text processing
Extracting, classifying and summarising information from emails, PDFs and forms.
Search over your own information
Letting people ask questions about your documents and get answers that show their sources.
Drafting and assistance
Features that prepare a first draft, such as a reply or a summary, for a person to check.
AI inside existing products
Adding an AI capability to software you already have, through LLM APIs.
Our approach
- 01
Start without AI
We first ask whether rules or a simpler approach would solve the problem. Often they do.
- 02
Test on real examples
We measure the feature against real inputs from your business before building around it.
- 03
Keep a person in the loop
Where mistakes matter, the software suggests and a person decides.
- 04
Plan for cost and data
We plan for usage costs, response times and exactly what data is sent to which provider.
Worth knowing
When this isn’t the right choice
We’d rather tell you now than build something you don’t need.
- The answer must be exactly right every time and can be written as rules. Plain code is cheaper and more predictable.
- The main goal is to say the product has AI. People notice when a feature doesn’t help them.
- There are no real examples to test with yet.
From our work
Talk2UsNow
In Talk2UsNow we deliberately left AI out. Simple keyword rules were enough to sort customer feedback, and they kept running costs close to zero.
Read the case studyCommon questions
Which AI models do you work with?
We build on LLM APIs from established providers and choose based on the task, cost, speed and data requirements.
Is our data safe if we use AI?
It depends on the provider and on how the feature is built. We check what data is sent, how the provider handles it and whether sensitive details should be removed first.
Can AI replace our customer support?
Usually not completely, and we’d be careful about promising that. AI can answer common questions and prepare drafts, with people handling the rest.
Do you build AI chatbots?
We can, when a chat interface is the right way to solve the problem. Often a simpler feature, like better search or an automatic summary, is more useful.
Tell us what you’re trying to build.
Start with a few lines. We’ll reply by email.