The next five years of AECO information management

Predictions about this industry usually arrive as tooling. Five years out, we are told, the model will federate itself overnight, the site will be scanned every evening, and a chat window will answer any question about the building. Perhaps. None of that settles the argument that decides money on an ordinary job, which is older and much duller: when three firms disagree about which drawing the work was built from, who can produce the answer, and how long does it take?

The most expensive document on a project is still the one nobody can put a name to. Not the PDF that failed a search — the stretch of time in which nobody can show, from a file made at the time, which issued set the site built from. Everything below follows from that, in the two climates this pillar has to serve: United Kingdom golden-thread duties, and Gulf permit and handover regimes that do not share a filing system with anyone.

So the next five years will not be decided by a new dashboard category. They will be decided by four questions about who holds what. Who can name the issued set. Who accepts the asset information model. Who is allowed to treat a machine reading as though it were an issue. And who can produce a UK golden thread or a Gulf handover pack without assembling it in the week the letter arrives. Each of those already has law, appointment practice and published cost research behind it, which is why none of them needs an invented survey of 2031 to be argued about.

The cost has not moved

In both markets the reassuring version runs like this. The system of record has been bought. Folders have names, models have federations, someone is "on BIM", and the plotter still sings at half seven. In that version information management is a purchase, and the next five years are a catalogue refresh.

The pattern underneath the purchases runs the other way. Rework is an information problem before it is a site problem. Forty systems can all be rational purchases and still leave no single source of truth, because none of them was asked to own the issue. The unofficial set in the trailer exists because the official cupboard cannot be trusted at the point of work. Email becomes the contract record because the issued set was never closed as an event. The claim that dies dies for want of a contemporaneous file, not for want of a remembered conversation. Owners wait a year for data they already paid for because handover was a pack. Facilities rebuild the register because the cut from project information model to asset information model had no owner. The twin freezes at practical completion because nobody was appointed — or paid — to keep it current. Construction AI pilots mostly die of three absences: no pin to the issued set, no owner once the champion leaves, and no production workflow.

Each of those is the same failure in different overalls.

The published cost research has not been repealed either. The Get It Right Initiative's measured direct cost of recorded error still sits around five per cent of construction cost — already above typical contractor margin — with a Delphi exercise in the same programme nearer 21 per cent once indirect cost, unrecorded process waste and latent defects are counted, and a wider literature band of 10–25 per cent when unmeasured costs are included. Those figures are not a single measured ledger, and they are not all "lost drawings". They remain the industry's best published attempt to price error. PlanGrid and FMI's 2018 Construction Disconnected survey of nearly 600 construction leaders attributed 48 per cent of United States rework to poor project data and poor communication, and reported that respondents spent 35 per cent of their time — more than fourteen hours a week — looking for project information, resolving conflict, and dealing with mistakes. FMI priced the United States labour cost of those activities at $177.5 billion for 2018. A gang standing about while somebody "just finds the latest" is still paid labour waiting on a set that should never have been ambiguous, and the wait is still cheaper than the wrong install that follows when nobody waits.

Nothing about that needs a new coefficient. The same cost simply keeps turning up in harder rooms: coordination, contracts, the jobsite, handover, occupation, and the tools the industry keeps buying.

Who owns the issued set

An issued set is not a folder of PDFs. It is a dated, attributed act: these sheets, this model state, this revision of the specification, released to these companies, for this purpose, in this information state. ISO 19650's language of work in progress, shared and published information exists because those states are different legal facts. A common data environment that stores every state and cannot tell them apart is a warehouse.

The next five years will not invent a fifth state. They will raise the price of pretending that four folders are four states.

On a Tuesday in 2031 a mid-size frame will still be going up in weather that has not improved. Somebody will still print A1s, because gloves and low winter sun do not care about a tablet. The test will still be whether three people who do not share an employer can name the current issued set for the work in front of them, and whether the record agrees with them. What an information-first project looks like in 2030 set out that test already; the intervening years have not replaced it with a gadget.

Ownership does not follow the purchase order. It follows whoever can sit down on a quiet afternoon, without pulling three people off their work, and reconstruct which containers went out at which suitability, to which firms, on a named day. If the hut set and the published container disagree and neither carries a closed issue history, the commercially right argument still loses, however obvious it looked at the time. A system that cannot answer that question is doing storage, whatever the demonstration promised.

The unofficial trailer set will not vanish because a vendor added a search box. It goes when the official set is trusted at the point of work: current, plotted from the published container, withdrawn when superseded, and reconstructable six months later without ringing the person who has since left. Until then the site will keep a second record, and the second record is the one the claim cannot use.

Who owns the AIM

Delivery builds a project information model. Operations need an asset information model. The conversion is a separate job, and it has no natural owner.

ISO 19650 already names the two models. The project information model answers delivery questions: can this duct pass the beam, which revision went to the trade, did the hold point pass. The asset information model answers operational questions: which unit is this, what was actually installed, where is the certificate that accepted it, when does the warranty end, what residual hazard must a later maintainer manage. They share objects. They do not share purpose, query or reader. A federation renamed "as-built" is a project information model with a new label, and publishing the common data environment archive to the client is a delivery cupboard with the client's name on it.

The next five years will keep producing that cupboard, because delivery is paid to complete and operations are paid to operate what they receive. The space between those two invoices is unowned labour. When a thing has no invoice it has no owner, and when it has no owner it gets performed as a rename.

What operations inherit is a vacuum that looks full. Folders, models, PDFs, a drawing register, a commissioning archive filed by system. Watch a facilities manager three weeks after taking over: on a phone camera in a plant room, photographing nameplates, because none of what she was handed answers the only question she has — show me this asset, the certificate that accepted it, the warranty that still runs, the residual I have to manage. So she builds a third model. A spreadsheet typed from nameplates, a maintenance system rebuilt from PDFs, a golden-thread file assembled after occupation. That third model is the twelve-month wait. The data was not lost from storage; it was never transferred into anything operations could query.

A digital twin does not close that vacuum by being named in the bid. The United Kingdom's Gemini Principles were plain back in 2018: a twin is a realistic digital representation that is dynamically updated. The industry kept the noun and dropped the verb. A twin that is only true on handover day is a souvenir of delivery, and operations do not need souvenirs. They need to know what is installed this week.

So the ownership question for 2026–2031 has less to do with whether owners buy twins than with whether anyone is appointed — a role, not a platform — to accept the asset information model before the delivery team demobilises, and whether anyone is still paid to keep it after taking-over. Where those two names are missing, expect what we already get: a day-one inoperable building, a register rebuilt from scratch, and a conference sentence about lifecycle data that never names the cut.

The machine can read. It cannot issue.

Language and vision models have made it cheap to point a window at last Thursday's export and get a confident paragraph back. When AI reads every drawing, the first job that still belongs to the issued set is register reconciliation: title block against register row against supersede state. That is a reading job, and it can be done badly by a person or badly by a model. Neither reading becomes an issue event. Work in progress is not shared, shared is not published, and a chat window publishes nothing.

Published labour-exposure research typically puts construction near the bottom of current AI coverage and office work near the top, which describes the work rather than the ambition. Construction labour is presence — weather, access, materials, other people's bodies, a drawing that has to be true on a slab poured yesterday. Current models do not pour. The hours that are exposed sit in the trailer and the regional office: finding the issued revision, reconciling a register, writing a query, checking a clause. Those are the hours we already fail at without a model. Point a copilot at a picnic of files and you spend the exposed slice while missing the issued set entirely.

The next five years will make retrieval cheaper. They will not turn retrieval into issue. Pilots will keep dying of the same three absences: the model was never pinned to a published container, the champion left, and the live transmittal never moved into the trial. A firm that treats a generated summary as the record has bought itself a faster unofficial set. The claim, the golden thread and the Gulf permit desk will still ask for the dated act.

What would change this part of the argument is a production workflow in which a machine reading is bound to a named container, a named suitability and a named person who is allowed to publish — not a more fluent paragraph. Until that binding is ordinary, "AI retrieval" is mostly a very quick way to read the wrong file. The ownership question stays human: who may change the live set, and who can prove they did.

Golden thread and Gulf handover are one test

The UK golden thread and Gulf digital handover do not share a filing system. They share a test. Can a named holder produce the current record, in a form a stranger to the job can use, without assembling it from mailboxes in the week it is demanded?

In England, the Building Safety Act 2022 and the later keeping-and-provision rules for occupied higher-risk buildings turned a phrase from the Independent Review of Building Regulations and Fire Safety into a records job with named holders. The golden thread is a duty to keep a digital, current, producible record of the building — who is responsible for it, what the building is, and how the spread of fire and structural failure are being managed — and to hand that record on. A ZIP named GOLDEN_THREAD_FINAL is not that duty. The golden thread in practice is retrieval plus keeping: the accountable person, or the principal accountable person, must be able to produce prescribed information and support a safety case report. The safety case is the argument. The thread is the file that argument has to sit on. If the live fire-strategy sheet is a revision behind the site, the duty has already failed, whatever the folder is called.

In the Gulf, "digital handover" is not one mandate. Dubai Municipality, Abu Dhabi and MOMRA are three information and submission regimes. They do not share a portal, a licensed person, a filename rule, or a meaning of the word model. One coordinated issued set does not become three authority packs by being uploaded three times; mapping is the job, and somebody has to be paid to do it. The construction published set is allowed to be richer than any permit model. It is not allowed to be a different building. Completion packs — as-built sheets, a completion model, the operation-and-maintenance objects a facilities team will actually query — are a later extract from the same identity, not a crate assembled after the crane comes down.

What each region gets right is not a slogan contest. The UK makes a named keeper and a producible current record unavoidable on higher-risk buildings. The Gulf makes a stamped, jurisdiction-specific pack unavoidable at permit and again at completion. Neither makes the issued set optional, and both punish the project that treated the archive as the deliverable.

Those regimes will not merge. Thresholds, forms and platform names will move, as they always do; the pattern will not. A job that works in both climates can name one published coordinated set and then extract the thread or the pack the named authority asked for. A job that keeps a UK cupboard and a Gulf cupboard, and hopes the two describe the same building, will fail in at least one climate — and will not be able to say which issued identity the works were built from.

Read the method

You will be shown percentages this year about how many firms will be "information-first by 2031". Ask for the denominator, because most AECO numbers arrive without one. Documentation volume without incident rates. Lifecycle data "lost at handover" without the moment of loss. Automation percentages without a definition of non-physical work. Read the method, and if there is no method you are reading a brochure.

What does not need inventing is already in force. The issued set already decides rework, standing time, wrong-revision installs and claims. The asset information model is already what operations need, and already the cut nobody is paid to make. Machine retrieval is already cheap enough to trial and already insufficient as issue. Golden-thread duties are already law on the buildings they apply to. Gulf permit and completion packs are already gates. ISO 19650 already separated the project information model from the asset information model, and work in progress from shared from published. The Gemini Principles already said that a twin which does not update is not a twin.

The next five years will make some of those duties more expensive to ignore: more occupied higher-risk buildings inside the UK regime, more municipal portals that will not accept a construction federation, more owners who have waited a year once and will write acceptance of the asset information model into the next appointment. The same years will make it cheaper than ever to generate a paragraph about a folder. There is no contradiction in that; it is one market, in which a photogenic tool gets pointed at an unowned record.

There is a plain test for all of it, and it needs no new survey. On an ordinary Tuesday, produce four things: the issued set the site built from; an asset information model a stranger to the job can query; a machine reading bound to a published container rather than to a heap of files; and a golden thread or completion pack extracted from that set rather than rebuilt. Most mid-size work cannot yet produce all four. When it can, the problem described here will be a history lesson.

Who owns the record in 2031

Ownership in 2031 is still a person.

A platform can store containers. A model can summarise them. A portal can reject a file that is too large or stamped by the wrong licensee. A holder is something else: a role on an appointment, with the authority to publish and to withdraw, the job of accepting the asset information model, the duty of keeping a golden thread current, the signature on the extract a municipal desk will stamp, and the ability to find the contemporaneous file after everyone else has moved on.

Where that role is vacant, the work still gets done badly by somebody. It is the BIM manager who treats the asset information model as a facilities consultant's problem, the document controller who treats handover as a compilation exercise in the last month, the contractor who treats as-built as a redline, the owner who believed they bought a record when they bought "BIM", and the vendor who will happily renew a seat. Each of them can show you a cupboard. None can tell you which containers went out, to whom, at what suitability, on the day the work was done.

So write the holder into the next appointment instead of hoping one emerges. Name, in the schedule of services, the person who may publish and withdraw. Name who accepts the asset information model, with an acceptance test that can fail, and require that acceptance before the delivery team demobilises rather than after. Keep paying somebody to hold the record after taking-over, because an unkept record rots within months of the last invoice. And where an authority pack or a golden thread is in scope, require it to be extracted from the published set, with the mapping written down at the start of the job and not in the week the letter arrives.

That costs money and it is worth saying so: a retained fee after practical completion, an information manager's hours that somebody has to carry, and a redrafted schedule of services that a lawyer will want to read. Against a year of operating a building nobody can query, that is a modest bill. Only clients and appointing parties can impose it, and principal contractors and lead designers are the ones who then have to resource it. If you want to know where your own job stands, ring round tomorrow and ask who holds the record. When that takes more than one phone call, you have found the line to write into the next appointment.