Oversight01
Human in the loop
AI suggests, a person approves. You set the threshold.
Ctrl+Enter
AI in product and process
Solutions built on your own data: computer vision, prediction and optimisation.
Ctrl+Enter — history on the left, prediction on the righthorizon: 8 periods
Application types
Pick a technology, see what it requires.
A camera that sees what an inspector sees
Recognising defects or completeness. The model doesn't replace quality control, but supports it heavily.
Makes sense when the defect is visible in an image and you can gather examples of good and bad units.What you need to get started
- Photos or video from a repeatable angle
- Examples of defects — the more types, the better
- A point in the process where a signal appears: rejection or alert
Principles
AI is part of the system, not an attraction.
We treat AI like any production component: it needs an owner, tests, monitoring and a plan for the day it starts making mistakes.
Harness02
A harness around AI
Rules, tests and constraints within which AI operates.
Measurability03
A criterion before the pilot
We agree the result that will mean the solution works.
Infrastructure04
Local, when required
Solutions can run on our servers or yours.
Maintenance05
Tuning after launch
Monitoring and tuning remain part of the service.
Honesty06
The simpler solution wins
When a normal rule is enough, we say so directly.
Questions
Five frequently asked questions.
ALT
Not seeing your question? ALT answers directly — including when the honest answer is “this is not a task for AI”.
Ask ALT01We don't have a data team or anyone in AI. Can we still start?+
Yes. We run the project end to end. On your side, one person needs to know the process and settle substantive questions.
02How do we know the model isn't making things up?+
Assistants cite sources. For prediction and classification we report accuracy on data the model never saw during training.
03Will our data go to an external model?+
We settle that at the start and put it in the contract. Some solutions keep data entirely within your infrastructure.
04What happens if the model starts getting it wrong after a year?+
We monitor quality and tune the model as part of maintenance. When retraining is needed, we say so directly.
05When do you advise against AI?+
When the process is rare, the data is missing or inconsistent, or a simple rule achieves the same result more cheaply.
Tell us what you want to predict or recognise.
We'll tell you what data it requires and whether you already have it. If not, we'll explain where to start.
Other services