IN Brief:
- Manufacturers are exploring AI across maintenance, inspection, forecasting, planning, and energy management.
- Core factory deployment remains limited by fragmented data, older equipment, skills shortages, and changing production conditions.
- Defined use cases, reliable baselines, and clear human accountability remain essential before systems are scaled.
Barclays Corporate Banking has published research examining how food and beverage manufacturers are approaching artificial intelligence across production, maintenance, quality, planning, and sustainability.
The study draws on responses from 100 industry leaders and indicates that investment intentions are rising, although implementation across core factory operations remains at an early stage for many businesses.
Applications under consideration include predictive maintenance, demand forecasting, production scheduling, visual inspection, workforce planning, energy management, waste reduction, and supply planning. These projects use plant or business data to identify patterns that would be difficult to assess consistently through manual analysis.
Fragmented information remains one of the principal barriers. Food factories frequently contain machinery from several suppliers and generations, with data divided between control systems, maintenance software, laboratory platforms, enterprise systems, spreadsheets, and paper records.
Production variability complicates the task further. Recipes, raw materials, pack formats, temperatures, cleaning cycles, promotions, and customer orders change regularly, meaning a model trained under one set of conditions may perform poorly when transferred to another line or product.
Skills shortages affect both development and continued operation. Engineering, production, quality, IT, and data teams need a shared understanding of the process, while external specialists must be able to interpret information generated within a food factory rather than treating it as a generic manufacturing dataset.
Recent processing investments already combine physical machinery with digital inspection and diagnostics. The modernisation of Mr Kipling processing equipment, for example, integrated AI-supported vision, robotics, sensor data, and remote engineering alongside retorts and material handling.
Reliable factory data comes before automation
Predictive maintenance is frequently prioritised because unplanned downtime has a measurable cost. Vibration, temperature, pressure, motor current, flow, and cycle data can identify changes before a bearing, seal, pump, gearbox, or drive fails.
Useful models require sufficient operating and failure history. A system trained only on normal production may recognise that behaviour has changed without identifying whether the cause is wear, a different product, a cleaning cycle, an altered speed, or a harmless process adjustment.
Food properties add another source of variation. Viscosity, particle size, temperature, fat content, aeration, and solids can affect equipment loading, so the same pump or mixer may produce different signals as recipes change.
Computer vision provides a more visible route into AI. Cameras and software can inspect seals, labels, date codes, fill levels, colour, shape, foreign material, and presentation at line speed, potentially identifying defects that are difficult to detect consistently through periodic manual checks.
Performance still depends on lighting, camera position, image quality, product variation, and the way rejected items are investigated. False rejects reduce output and waste saleable product, while missed defects undermine confidence in the system.
Forecasting and scheduling draw on a wider collection of data, including retailer orders, promotions, weather, shelf life, ingredient availability, labour, changeovers, storage, and transport. Algorithms can process those variables rapidly, but weak source data produces an equally rapid route to an inaccurate plan.
Governance must define who can approve automated recommendations, how changes are documented, and what happens when a model moves outside its validated range. Food safety, legality, and product-release decisions require clear human responsibility even where software performs much of the analysis.
Cybersecurity becomes more significant as operational information is connected with cloud services or supplier platforms. Remote access can support diagnostics and model development, while also increasing the number of routes into production systems.
Network separation, access control, backups, supplier management, and incident response need to develop alongside data connectivity. A system intended to improve production cannot be allowed to weaken the availability or integrity of the line it monitors.
Defined projects provide a more reliable route than broad factory-wide programmes. A manufacturer can target one recurring stoppage, one inspection problem, or one forecasting process, then establish baseline performance and measure the result before extending the method.
Models also require continued maintenance. Equipment wears, products change, sensors are replaced, and market behaviour shifts, causing performance to drift unless the system is monitored and retrained under controlled conditions.
AI can strengthen food manufacturing where reliable data and clear process ownership already exist. Plants with inconsistent records, disconnected systems, or poorly understood operating variation will need to address those foundations before automated analysis can deliver dependable production decisions.
