Artificial Intelligence
Systems that decide, read and follow through. One rule underneath: arithmetic runs in a deterministic engine, the AI handles judgement, and every output is measured.
Request a short diagnosticDecision Intelligence
How much to produce, what to charge and which job is profitable is usually decided from the past and from instinct.
Knowing how much to produce, what to buy and what to charge — from measurement rather than instinct.
- Demand and sales forecasting
- Order book and capacity planning
- Cash flow and collections forecasting
- Stock levels and reorder points
- Lead time and delivery date prediction
- Quotation and pricing engine
- True profitability by product, customer and order
- Cost and margin modelling
- Pricing scenario analysis
Enterprise AI
Reading invoices, contracts and specs — and finding the company's own knowledge — eats most of the time; that knowledge lives in a few heads.
Assistants that read the company's documents and knowledge and cite their source.
- Reading invoices, specs and contracts
- Internal knowledge assistant (with source citation)
- Document verification and reconciliation
- Enterprise search and knowledge retrieval
- Workflow agents and flow automation
Industrial AI
Quality issues, faults and deviations are usually noticed too late — by eye or by complaint.
Intelligence that reads image and signal on the floor and is built into production.
- Visual quality inspection — using methods that work from few examples
- Predictive maintenance
- Process deviation and anomaly detection
- Counting, measurement and recognition automation
- Sensor-data intelligence
AI Assurance
Whether a model works is usually understood only when a user complains.
The layer that proves a model works — not an add-on sold later.
- Accuracy gate and threshold management
- Regression suite and back-testing
- Model monitoring and drift alerts
- Model validation and evaluation
- Audit trail and governance
How much should we make next month, and how much should we buy?
Who keys in what the incoming invoice and the spec actually say?
Can this defect be caught before the part ships?
How do we know the model is still working today?
At a frozen-food producer, order history, the promotion calendar and shipment records feed one forecasting model. The model returns a weekly demand range, and the production plan is built on that range. The same system reads supplier invoices, pulls out the fields, and routes anything doubtful to a named reviewer. Forecast error is re-measured every week and shown on the management dashboard. Anything below the threshold stays out of the plan and goes to a person instead.
At least two years of order or sales records — a spreadsheet is enough.
A product and customer list with unique codes.
Dates of events such as promotions, price changes and stoppages.
A few dozen sample documents to be read; scans of paper are fine.
The acceptance measure is forecast error (MAPE): the threshold is written down before the build and checked by measurement before the system goes live.
Not just what is delivered — that it works is proven
The assurance layer is not an add-on sold later; it ships inside every solution. Concreteness comes from mechanism, not numbers.
How we work