XBP Global has announced an executed contract with British Airways to deploy its Plexus AI platform across aircraft maintenance documentation and asset-lifecycle records.
According to XBP, the system will capture, vectorise, and store maintenance events from British Airways’ fleet in a private vector database while automating parts of the surrounding document and maintenance workflow. The company also says Plexus will be used to reproduce servicing histories when aircraft are sold or reach the end of a lease.
The contract was announced on 26 August, but XBP has not disclosed its financial value, deployment timetable, initial fleet scope, or targeted cost savings. British Airways has not issued a parallel public announcement located during verification, so claims around future operational efficiency remain attributable to XBP rather than independently demonstrated results from the airline.
The application is nevertheless materially different from using generative AI for passenger enquiries or office administration. Aircraft accumulate large volumes of maintenance, component, inspection, modification, and servicing records over operating lives measured in decades, and those records need to remain searchable and associated with the correct aircraft or component as ownership and maintenance responsibility change.
XBP says the platform will run inside British Airways’ private cloud and use open-weight AI models, keeping operational data within the airline’s controlled environment. Agentic functions are intended to manage access to the datastore as well as retrieval and workflow, while the company describes the architecture as a sovereign deployment designed around data privacy.
The vendor also says Plexus will replace some legacy third-party software, although it has not identified those systems or quantified how much of the existing software footprint will be removed. That omission matters because consolidation can mean anything from eliminating several genuine production systems to replacing a narrow document-handling layer, with very different implications for cost and integration.
The engineering problem sits in the quality of the record chain rather than the speed of the search box. A maintenance system needs to distinguish aircraft, assemblies, serialised components, revisions, dates, and servicing events without losing provenance as records move between technical teams and organisations. A convincing summary generated from incomplete or wrongly associated source material would be considerably less useful than a slower search that preserves the evidence trail.
That makes controlled automation particularly important. XBP’s system will have to interact with existing engineering and enterprise processes rather than operate as a disconnected AI layer, while access permissions, version control, approval routes, and historical records remain intact around it.
Similar issues are appearing elsewhere as agentic software moves into engineering environments. UKAEA’s recent engineering-automation spinout, for example, is targeting agents that coordinate engineering tools while preserving permissions, design baselines, and review gates. Aviation maintenance is a different application, but the governance requirement is familiar: automation becomes useful only if the responsible engineer can establish what the system did and which records informed it.
XBP describes the British Airways programme as its second major sovereign AI deployment following work with France’s national health insurance organisation, CNAM. It intends to offer the same core architecture to other airlines and industrial companies managing complex, high-value assets where service histories influence uptime, safety, and lifecycle cost.
That wider industrial proposition is plausible because the underlying documentation problem is not unique to aviation. Rail fleets, turbines, defence platforms, process equipment, and heavy machinery all accumulate service histories, inspection results, component replacements, and configuration changes that have to remain usable long after the original engineer has moved on.
The complication is that industrial records are not interchangeable. Each sector brings different terminology, data structures, regulatory requirements, retention rules, safety cases, and approval responsibilities, meaning a common AI architecture still needs substantial domain integration before it can be trusted with operational work.
XBP’s announcement also makes several regulatory and performance claims, including describing the deployment as compliant with the EU AI Act and capable of delivering significant efficiencies. Those statements have not been accompanied by a published independent assessment or quantified operating baseline, and the system has only just entered deployment.
The useful measurement will therefore come after implementation. British Airways should be able to establish whether maintenance engineers and records teams spend less time finding, checking, assembling, and transferring documentation, while tracking whether retrieval errors, missing records, or manual interventions increase or decrease.
Aircraft maintenance offers little tolerance for software that is impressive until somebody asks it to prove where an answer came from. If Plexus can shorten the administrative workload while maintaining traceable servicing histories, the contract could become a useful industrial AI reference. Until those production results exist, the engineering significance lies in the deployment itself rather than the efficiency claims wrapped around it.
