How Aquafin is rethinking asset data management through AI-supported extraction
Every pump, motor, valve or electrical component installed in an industrial facility arrives with information that will influence its operation for years. Flow rates, electrical characteristics, tolerances and maintenance intervals are documented in manuals, drawings and dossiers. Before this information can support maintenance, it must be found, interpreted, validated and entered into the organisation’s systems.
That apparently administrative step is becoming a hidden bottleneck, consuming technical resources and delaying the transfer from projects into operations.
At Aquafin, Asset Data Teams process AsBuilt dossiers, transfer structured information into SAP and archive plans and electrical diagrams in systems such as Autodesk Vault. Much of that work still requires manual searches.
At Asset Performance, Joris Dircx, Manager Asset Data at Aquafin, and Jakub Nowak, Manager Asset Data Installations, explain how Aquafin is moving from manual data input towards AI-driven extraction. Their case is not simply about using technology to read PDF files. It addresses a more fundamental question: how can organisations transform unstructured project documentation into reliable asset information without removing human expertise from the process?
When more data does not mean better information
“Efficient data management is no longer purely an IT matter,” Dircx explains. “We are facing a combination of increasingly complex infrastructure, ageing assets and an ageing technical workforce. That makes the availability of reliable information a strategic concern.”
As infrastructure ages, maintenance and asset management teams must decide which equipment requires intervention and which assets can remain in service. Without trustworthy information about characteristics, history and condition, those decisions become more dependent on assumptions.
Organisations are also generating more information than ever through sensors, BIM models, specifications and digital maintenance records. Yet more data does not automatically lead to better decisions.
“Data must become a reliable source on which an asset manager can base decisions,” says Dircx. “Otherwise, the organisation is not working with usable insight, but with a large volume of fragmented and unstructured information.”
That fragmentation will be familiar to many technical managers. Asset information is often distributed across EAM systems, engineering databases, network folders, document-management platforms, spreadsheets and personal archives. Individual systems may each contain part of the truth, while experienced employees provide the context needed to interpret it.
This creates data silos, spreadsheet proliferation and uncertainty over which version is authoritative. It also makes organisations dependent on knowledge held by individual employees. As experienced technicians and asset managers retire, decades of practical knowledge risk disappearing with them.
Efficient data management is therefore not only about storing more information. It is about structuring technical knowledge so that it remains accessible, dependable and usable throughout the asset lifecycle.
The fragile bridge between projects and operations
This weakness often becomes most visible during project handover. A completed infrastructure project is not necessarily an operationally ready asset.
Before Operations can maintain newly installed equipment, information must be extracted from project documentation and translated into the organisation’s data structures. A manual may contain hundreds of pages, while the asset management system requires only a limited set of critical parameters, such as a flow rate, electrical specification, maintenance interval or operating tolerance.
“Our objective is to make the information flow from projects towards Operations and Asset Management faster, smoother and more efficient,” Jakub Nowak says. “Today, we still have to search through technical specifications, identify the information relating to our assets and translate it into structured data. It is a process that requires time and is susceptible to error.”
A single installation may contain many pumps, drives, valves, instruments and electrical components. Manual processing can delay complete asset information and create a risk that values are omitted, entered incorrectly or linked to the wrong equipment. An error during handover can later affect maintenance preparation, spare-part selection or operational risk assessment.
Aquafin is therefore investigating how AI-supported extraction can identify relevant parameters and present them for review before they enter the appropriate systems. The objective is a faster and more controlled connection between technical documentation and operational asset information.
Supporting rather than replacing the expert
Technical documentation is rarely standardised. The same parameter may appear in different units or sections, while tables may contain several values requiring interpretation. Fully autonomous extraction is therefore neither the immediate objective nor necessarily the most responsible approach.
“When we apply AI to technical asset data, the role of the human expert shifts from processing data to validating data,” Nowak says. “Technical knowledge remains extremely important. What we are trying to remove is the repetitive part of the work.”
Instead of manually reading every page and transcribing each value, data specialists can concentrate on assessing proposed information, resolving exceptions and applying technical judgement where it adds the most value.
Human validation remains the final quality gate. AI can locate, recognise and structure information, but a competent employee must still determine whether the result is plausible, complete and applicable to the installed asset.
The resulting model is one of augmented expertise. The technology carries out repetitive search and extraction work, while employees remain responsible for validation, exceptions and contextual interpretation.
Building stronger foundations for advanced asset management
Predictive maintenance, advanced analytics and digital twins all depend on the quality of the underlying asset information. An algorithm cannot compensate for incomplete characteristics, inconsistent naming or incorrect relationships between equipment and documents.
“Correct and structured asset data forms the bridge between the construction phase and several decades of safe and efficient operation,” Dircx emphasises. “Without that data, an EAM system has very limited value.”
When AsBuilt information is processed correctly, planners retrieve specifications faster, technicians access the correct manuals and reliability engineers connect failure histories with the right technical context.
“Better asset data first creates operational calm,” Dircx says. “Technical employees know where to find information during work preparation or in the event of a failure, and they know that the information is reliable.”
Over time, structured data supports better maintenance intervals, more reliable predictive models and more thorough root cause analyses. It also helps identify recurring problem assets, extend asset life and direct resources towards the greatest risks.
From experiment to operational discipline
Many organisations have tested AI through predictive models, document recognition or automated classification. The harder challenge is turning those experiments into repeatable operational processes.
“The greatest opportunity for AI in asset-intensive sectors is not necessarily another, even more complex predictive model,” Dircx says. “The real transformation is the move from technology experiments towards an automated operational discipline.”
A straightforward application integrated into daily work and scaled across projects may deliver more value than an advanced pilot disconnected from operations.
For organisations still dependent on manual input, Aquafin’s advice is pragmatic: think big, start small and automate as close to the source as possible.
“You do not need to wait for the perfect AI agent,” Nowak observes. “Apply the 80/20 rule. Identify the asset categories that generate most of the AsBuilt documentation and start with those.”
Introducing AI also forces organisations to clarify which information is needed, where it should be stored, who owns it and how exceptions will be handled. Technology can therefore become a catalyst for better data governance.
Better data, calmer operations
The immediate benefit of better asset data is operational clarity. During maintenance preparation or an unexpected failure, technicians need to know where information can be found and whether it can be trusted. Time spent searching through project folders or comparing conflicting documents directly affects maintenance efficiency.
“Better asset data first creates operational calm,” Dircx says. “Technical employees know where to find information during work preparation or in the event of a failure, and they know that the information is reliable.”
Over time, structured data supports better preventive maintenance intervals, more reliable predictive models and stronger root cause analyses. It also helps organisations identify recurring problem assets, reduce unplanned interruptions and make more informed lifecycle decisions.
Reliable information can also support longer asset life and help direct resources towards the equipment presenting the greatest risk. As Dircx concludes: “Where data was previously viewed mainly as something to be stored, it is now becoming the engine behind a growing number of processes.”
A practical case with broad relevance
The Aquafin presentation at Asset Performance starts from a problem found in virtually every asset-intensive organisation. New and modified installations are delivered, final project dossiers arrive, and technical teams must convert extensive documentation into information that can support Operations and Asset Management.
The case examines how AI can assist that process without removing the human expert from the loop. Participants will gain insight into the bottlenecks Aquafin encountered, the potential of automated extraction and the controls needed to preserve data quality.
The broader message reaches far beyond document processing. Digitalisation in maintenance does not begin with the most advanced algorithm. It begins with ensuring that the organisation can trust the information on which its people, systems and models depend.
For technical managers, the case offers a practical view of how one of the least visible parts of asset management may become one of the most consequential applications of AI: converting technical documentation into dependable operational knowledge, while allowing human expertise to focus on the work that genuinely requires it.