Rules where rules work. AI where they don’t.
Some problems need a rule that gives the same answer every time and can be checked line by line. Others need a model that reads what no rule can: free text, voices, images, patterns across millions of events. We choose per problem, in what we build, how we guard, how we advise and what we teach.
A rule or AI?
Algorithms are good, and AI is good. Each belongs in its right place.
| Use… | When… | Example | What it asks of you |
|---|---|---|---|
| A rule or fixed algorithm | The logic is known, the answer must be the same every time, and an auditor must be able to trace it. | Fee calculations; access rules; our incident page’s decision to show the number. | Clear logic and tests. |
| A machine-learning model | The rules are too many or too fuzzy to write by hand, and there are enough labelled examples. | Ranking which alerts are likely real; flagging unusual transactions. | Enough labelled examples, and ongoing review. |
| Generative AI (a large language model) | The material is language: reading, summarising, sorting or drafting text. | Sorting thousands of open survey answers; summarising an incident timeline; answering from your own documents. | Every output checked against its source. |
| An AI agent | The task has many steps that cannot be scripted in advance, and each step can be limited and logged. | Gathering evidence across logs; drafting a response plan for approval. | Limits on what it may do, and a log of what it did. |
| A person | The decision has legal, financial or human consequences, or the data is too thin to trust a model. | Declaring an incident; signing an audit-readiness finding; advising a board. | Time and judgement. |
We start with the simplest thing that works. Often that is a rule.
AI across our work
Build
We build ordinary software and AI features side by side. AI goes where a rule cannot do the job; the rest stays code you can test line by line.
Software & AIGuard
Threat actors now use AI to write convincing phishing, imitate voices and faces, and probe systems at machine speed. Countering that takes AI too: Red Crow uses AI agents in detection and response. Ask us what they do on their own, what needs your approval, and where your data is processed.
Red CrowGuide
Bridgeway uses AI to read the whole data set, not a sample: every interview, every open survey answer, every complaint and transaction record. We read the text and the numbers together, and our consultants turn what we find into advice your board can act on.
Train
AI adoption training by role: what your staff may put into an AI tool and what they may not, how to check an answer before using it, and how to recognise AI-made phishing and a cloned voice on a call.
Boundaries
We publish no accuracy, speed or savings figures for AI that we have not measured with you.
FAQ
Where is our data handled?
Agreed with you in writing before the engagement starts: where each deliverable is stored, in which region, who can open it, and how it is handed back or deleted when we finish.
Will AI replace our staff?
It takes the reading and sorting off people’s desks. We propose which decisions stay with your people, and you approve that split.
Why start with a rule?
A rule is cheaper to run, easier to test and easier to audit. When it cannot do the job, we say which kind of AI we propose and why.