AI Development
Logic first.
AI last.
Most AI projects start with a model and go looking for a problem. The result is a chat box on top of a business that still runs on the same spreadsheets and judgement calls as before.
We do AI development the other way round. We map the decisions your business makes, write down the rules behind them, and only then decide where a model belongs. The AI works inside that logic, so its output can be checked, explained and trusted.
Agents on an explicit logic layer
The rules decide. The model assists.
An agent that runs on free-form prompting will eventually say something you didn't approve. We build agents that sit on top of a documented logic layer: the rules, statuses and allowed actions are defined in code, and the model works within them. When the logic says stop or hand to a person, it does.
- Rules and allowed actions defined in code
- Clear handoff points to a person
- Every decision logged and traceable
- Behaviour you can change without retraining
Judgement-heavy steps and extraction
The reading and sorting your team does by hand.
A lot of skilled time goes into reading documents, pulling out the details that matter and deciding what happens next. We automate those steps: extracting structured data from emails, PDFs and forms, classifying what comes in, and routing it. Where the input is unclear, it goes to a person instead of being guessed.
- Document and email data extraction
- Classification and triage of incoming work
- Drafting responses for review
- Confidence checks with human review
AI inside your existing system
Added to the software you already run.
You don't need a new platform to use AI. We add AI features to an existing application through its data and APIs, so they work with your current records and permissions rather than beside them. If the system needs a cleaner logic layer first, we build that before adding the model.
- AI features in existing applications
- Connected to your data and permissions
- API-first so other systems can use it
- Full code and repository ownership
Deliverables
We can build
- AI agents that guide customers through a process
- Intake that replaces static forms
- Document and invoice data extraction
- Email triage and routing
- Drafted quotes and replies for review
- Classification of incoming requests
- AI features in an existing web application
- Internal assistants over your own records
How we work
Same method, every build.
- 01Discovery
- 02Logic Map
- 03Data Model
- 04Build
- 05Handover
Case study
FormOS
Our own AI product, built the same way. FormOS replaces static forms with an AI agent that runs on our logic engine, so a prospect arrives on your site and walks away with a clear diagnosis of their problem, no form submission required. It's in early access now.
See FormOS →FAQ
Common questions
What can AI be trusted with, and what can't it?
AI is good at reading, sorting, extracting and drafting. It is not reliable as the final word on decisions with real consequences, such as pricing, approvals or anything legal or financial. We design for that: the rules live in code, the model handles the reading and drafting, and anything uncertain or high-stakes goes to a person.
What happens to our data?
We agree how data is handled before we build, as part of the logic map. That includes which data the AI is allowed to see, which model provider is used, and where your data and logs are stored. You own the data and the accounts it lives in, and the AI only gets access to what the process actually needs.
How much does AI development cost?
It depends on how many steps are being automated, the systems involved and the volume the AI has to handle, which also drives ongoing model usage costs. We scope every project after a discovery session and give you a fixed quote for each stage, so you know the build cost before it starts. Start a project and we will tell you what yours involves.
How do you test AI features?
Against your real work. We build a test set from actual examples, including the awkward ones, and check the output against what your team would have done. The rules in the logic layer are tested like any other code. After launch, decisions are logged so they can be reviewed and the logic adjusted.
Got a step that needs judgement?
Tell us which part of your process takes the most reading, checking or deciding. We'll tell you where AI fits and where it doesn't.
Related services