What the deliverable is supposed to be
COBie, the Construction Operations Building information exchange, carries the non-geometric facility data a building owner needs on day one of operations. It travels as a workbook, as IFC, or as ifcXML, and its sheets describe the facility down to the individual installed component.
The structure is sound. Contact, Facility, Floor, Space and Zone establish where things are. Type and Component separate the model of equipment from the specific unit installed, so a building with forty identical VAV boxes carries one Type record and forty Component records. System groups them functionally, Spare lists replacement parts, Job holds the maintenance tasks, and Document links the manuals.
Feed that into a maintenance management system and a facilities team starts work on day one with an asset register instead of a filing cabinet.
What actually gets delivered
A workbook with the Component sheet populated and half the fields empty.
Serial numbers missing or filled with the literal string TBD. Warranty start dates absent, or set to the contract completion date for equipment that was energized nine months earlier. Manufacturer names spelled four ways across the same sheet. Space names that match neither the architectural drawings nor the signage the owner installed. Barcodes absent because no one tagged the equipment during installation.
The owner's facilities director opens it, finds it unusable, and their team spends the first year rebuilding the register by walking the building with a clipboard. That register is the one they trust, and the COBie file becomes a contract deliverable that was accepted once and opened no further.
The scheduling problem
Every field in that workbook already existed, in a document, months before handover.
The manufacturer and model number were in the approved submittal. The performance data was in the product data sheet attached to it. The warranty terms arrived with the same package or with the purchase order. The serial number was on the equipment when it was delivered, and on the packing slip. The installation date is in the daily report and the inspection record. The maintenance tasks are in the operation and maintenance manual the manufacturer supplied.
None of that information is hard to obtain at the moment it appears, and all of it is hard to reconstruct eighteen months later by a project coordinator who was not on site when any of it happened.
COBie fails because it is produced as a closeout task rather than accumulated as a construction byproduct. Somebody sits down in month twenty-two and tries to assemble from memory and email a dataset that the project generated continuously and discarded continuously.
Why the usual fixes do not work
Asking the contractor to fill the spreadsheet earlier moves the pain without removing it, because at month ten the serial numbers do not exist yet and the model numbers may still change through a substitution request.
Asking the design team to populate the model with type data helps on the Type sheet and does nothing for Component, which is where the operational value lives. An owner does not need to know that the design specified a certain air handler. They need to know which unit is in which mechanical room, when its warranty expires, and who to call.
Buying a handover software product late in construction gets you a better-looking version of the same reconstruction exercise, performed by the same person from the same incomplete email history.
What has to change
The record has to be the object rather than the document.
When a submittal is approved, the equipment type it describes should become a durable record carrying its manufacturer, model, performance data, and warranty terms, linked to the specification section that required it. When the purchase order issues, it attaches to that type. When the delivery arrives, the packing slip's serial numbers create the component records against it. When the installer signs off, the installation date lands on the component. When the manual arrives, it links to the type rather than to a folder.
By substantial completion the register exists because it was assembled by the work, and generating COBie becomes an export rather than a project.
Document extraction is what makes this practical rather than aspirational. Reading a manufacturer, a model number, and a warranty term out of a submittal PDF is a solved problem now, and it is the difference between asking a project engineer to key three hundred equipment records and asking them to confirm three hundred extracted ones.
The honest caveat
An accurate asset register still needs a person to accept it. Extraction produces candidates, and a component record with a confidently wrong serial number is more dangerous to a maintenance team than an empty field, because they will schedule against it.
The realistic target is not an unattended pipeline. It is moving the human effort from transcription, which is expensive and error-prone and happens at the worst possible moment, to confirmation, which is cheap and happens while the equipment is still in front of somebody.