Technology · Automation & AI
Automation you can explain
We start with clear rules and alerts, add analytics and AI only where the data supports them, and keep people in control of anything that acts on the physical world.
Overview
Useful automation is careful automation
Most operational value comes from simple, reliable automation: an alert when a cold room drifts out of range, a notice when a vehicle leaves a site after hours, a reminder before a service falls due. Rules like these are easy to understand, easy to test and easy to change.
More advanced methods, such as anomaly detection and forecasting, become useful once a system has collected enough clean history. We add them where the data supports them and where a better prediction leads to a better decision. We do not add AI to a system for its own sake.
When automation acts on equipment, safety comes first. The remote immobilizer in LIOMO Fleet is an example: the command is sent only when the vehicle is stationary with its ignition off, the condition is checked again when the tracker connects, and the command counts only once the tracker confirms it. If confirmation does not arrive, the command is taken back automatically; a release is retried with back-off until it is confirmed. Every step is logged.
Principles
How we approach automation and AI
01
Rules before models
Clear, testable rules cover most needs. Statistical and machine-learning methods come later, where they add something rules cannot.
02
People stay in the loop
Systems alert and recommend. Actions that are costly, hard to reverse or safety-related need a person’s decision or a strict interlock.
03
Explainable results
Every alert and prediction should show the reading, rule or trend behind it, so the person acting on it can check it.
04
Clean data first
Results are only as good as their inputs, so missing, stale and implausible readings are dealt with before anything is analysed.
Building blocks
From rules to prediction
- 01
Rules and alerts
Conditions checked against every incoming reading, with notifications sent to the right people.
- Thresholds and limits
- Geofences
- Schedules and working hours
- Notifications and escalation
- 02
Analytics and reporting
Summaries that turn raw readings into trends, scores and costs people can act on.
- Aggregates and trends
- Scores per asset or driver
- Consumption and cost metrics
- Scheduled reports
- 03
Anomaly detection
Methods that flag readings or patterns that differ from an asset’s normal behaviour.
- Statistical baselines
- Rate-of-change checks
- Daily and seasonal patterns
- Sudden drop and refill detection
- 04
Forecasting and machine learning
Models that estimate what is likely to happen next, used only where enough clean history exists.
- Time-series forecasting
- Demand and consumption forecasts
- Maintenance prediction
- Model performance monitoring
- 05
Control and safety
Safeguards for any automation that switches, stops or changes equipment.
- Precondition checks
- Confirmation from the device
- Automatic roll-back
- Retries with back-off
- Command logs
Engineering practices
Engineering practices
- Interlocks before actions
- A command to equipment is sent only when stated conditions are met, and those conditions are checked again at the moment of delivery.
- Confirmation, not assumption
- An action counts only when the device confirms it. If confirmation does not arrive, the system takes the command back or retries it, whichever is the safe outcome.
- Alerts without noise
- Thresholds, schedules and delays are set with the people who receive the alerts, because an alert that fires too often is soon ignored.
- Tried on history first
- New rules and models are run against past data before going live, to see what they would have flagged and how often.
- Every action on record
- Alerts, commands and the people who acted on them are logged, so any automated action can be traced and reviewed.
Applied in
Solutions
FAQ
Questions buyers ask
Do we need AI to get value from our data?
Usually not at first. Well-chosen rules, alerts and reports deliver most of the early value. AI methods become worth considering once there is enough clean history and a clear decision they would improve.
How much data is needed before forecasting is useful?
It depends on what is being predicted. A pattern with weekly or seasonal cycles needs history covering several of those cycles, with gaps and faulty readings dealt with. We assess the data you have before recommending a model.
Can automation control our equipment directly?
It can, where the equipment supports it and the risk is understood. We design such actions with preconditions, confirmation from the device, automatic roll-back and a full log, as in the remote immobilizer in LIOMO Fleet. Safety-critical functions stay with the people and safety systems responsible for them.
How do we know why the system raised an alert?
Each alert should record the rule and the readings that triggered it. For predictions, the aim is the same: show the inputs and the trend behind the result, not a score without an explanation.
Technology areas
- IoT & ConnectivityDevices, sensors, gateways and networks chosen for each site, so field data arrives complete and on time.Explore
- Software EngineeringPlatforms, dashboards, APIs and data pipelines built to run every day, tested before release and kept healthy after.Explore
- Systems IntegrationConnecting devices and new platforms to the ERP, accounting, warehouse and building systems you already run.Explore
Discuss the technical side of your project
Bring your engineers. We are happy to go into protocols, data models, security and integration details.