PTQ Q3 2026 Issue

completion data, deviation logging, and comprehensive audit trails support both internal performance monitoring and external regulatory review. Document-based systems are structurally incapable of generating equivalent evi- dence, which represents an increasing compliance expo- sure as regulatory expectations around process safety performance documentation continue to develop. AI-assisted procedure quality management Legacy document migration problem The transition from document-based to digital procedure management encounters a material barrier at the point of legacy document migration. Refinery procedure librar - ies typically contain thousands of documents, including SOPs, maintenance procedures, emergency plans, and reference materials, developed over decades using incon- sistent formatting conventions, terminology, and structural approaches. Before these documents can be governed, executed, and improved within a digital platform, they must be structured and normalised into the platform’s content model. Manual conversion is labour-intensive and creates a migration bottleneck that delays the realisation of digital governance benefits for the bulk of the existing document estate. Artificial intelligence is increasingly applied to address this constraint across three primary capability areas: large-scale document digitisation, content normalisation, and componentisation Artificial intelligence (AI) is increasingly applied to address this constraint across three primary capability areas: large- scale document digitisation, content normalisation, and componentisation. Human-in-the-loop validation ensures that standardised content is reviewed by subject matter experts (SMEs) before publication and field use. Large-scale document digitisation AI pipelines developed in partnership with hyperscale cloud platforms extract structured procedural content from heterogeneous source formats, including PDFs, Microsoft Word documents, and mixed-format legacy files. These systems replace manually intensive, rule-based conversion workflows with batch processing capable of significantly higher throughput. Reported accuracy benchmarks exceed 90% section and statement match rates against source documents. Conversion time that was previously measured in days is reduced to minutes per document. For organisations managing large legacy libraries, this capability substantially reduces the labour effort of the digitisation transition. The primary practical barrier to modernisation – the effort required to bring historical

documentation into a managed digital environment – is materially lower with AI-assisted ingestion than with man - ual conversion processes. Content normalisation and component management Digital procedure management platforms decompose SOPs and related documents into reusable content statements: discrete, governed text components that can be reused across procedures, reference documents, and emergency plans. Inconsistency in terminology and phrasing across a large procedure library constitutes a silent safety hazard. When the same operation is described using different language in different documents, operators may interpret the guidance differently, producing execution variance that is difficult to detect and may not surface until a failure event occurs. Investigating the 2018 explosion of an FCC unit at a refinery in Superior, Wisconsin, the US Chemical Safety Board found that several key instructions, such as “Always keep the regenerator pressure a couple of pounds higher than the reactor pressure”, were understood differently by employees, with opinions diverging as to whether this implied only a minimum or also a maximum limit. The result, the report noted, was procedures “leaving vague state- ments up to each operator’s interpretation”.7 That is why HSG48 and HSG253 highlight the impor - tance of clarity and conciseness, providing specific guid - ance, such as the use of positive action verbs or the use of precise checklists and other aids, to ensure adherence. AI should not, by itself, attempt to clarify procedures, but AI-assisted processing can be applied at the point of document ingestion to normalise terminology and struc- tural patterns or flag procedures that do not meet HSG criteria before documents enter the governed environment. Central management of shared content components ena- bles updates to propagate automatically to all documents in which a given statement appears, reducing both the effort and risk of inconsistency in large-scale libraries. Human-in-the-loop validation and auditability AI-generated structured output is not published directly to the live procedure environment. Following extraction and normalisation, output is presented alongside the orig- inal source document for review by SMEs, who assess the generated content for accuracy, completeness, and correct interpretation of the source material. Corrections made dur- ing this review cycle are incorporated into the underlying model logic, progressively improving output quality for sub- sequent processing runs. The validation workflow produces a traceable lineage from source document to published procedure statement – a documented record of the conversion process that sup- ports both internal quality assurance and external audit requirements. Manual conversion processes rarely pro- duce equivalent documentation of the transformation from source to published content, which represents a gap in the evidentiary chain that process industry auditors increas- ingly expect to examine.

PTQ Q3 2026