The major construction platforms have all shipped this year, and the updates are good. Procore added submittal syncing between collaborating companies and AI resource recommendations on the planning Gantt. Autodesk added automated drawing extraction and tighter control over which third-party apps can reach a project. Trimble put a 3D model viewer in Field View so a snag can be pinned to a location rather than described. Simpro shipped an accounts receivable agent that decides when to chase an invoice.
Every one of those makes a platform better at the slice of the project it already owns. None of them makes it better at the part it has never been able to see, and that part is where most disputes start.
What each one owns, and what it never sees
Laid out side by side, the pattern is hard to miss.
| Platform | What it owns | What it never sees |
|---|---|---|
| Procore | The main contractor’s project: RFIs, submittals, drawings, programme | Anything agreed by people who do not hold a licence |
| Autodesk Construction Cloud | The model, the drawings and the document control around them | Why a detail was changed on site at seven in the morning |
| Trimble Viewpoint Field View | Forms, snags, inspections and permits, now pinned to the model | The instruction that caused the work the form records |
| Simpro | The subcontractor’s own jobs, scheduling and invoicing | The main contractor’s version of the same events |
Each is excellent inside its boundary. The boundary is the problem, and it is not a boundary any of them can move, because the thing outside it is a conversation rather than a system.
Per-seat pricing decides what your record contains
This is the mechanism, and it is worth stating plainly because it is usually discussed as a procurement detail rather than a records problem.
A platform is licensed per user, and in practice licensed to the main contractor’s own staff. Count the people who actually move a project forward in a given week and most of them are not on that list.
The three moments it costs you
- The delay position. Most of the hours in an extension of time claim go on finding the entries rather than making the argument. If the sequence change was agreed in a message and the platform only holds the formal correspondence, that time is spent reconstructing rather than retrieving. A dated site diary that captured it when it happened turns days into minutes.
- The variation nobody confirmed. The work was instructed, it was done, and the confirmation exists as a photograph of a marked-up drawing on somebody’s phone. Whether you get paid for it depends entirely on whether that survived, which is the subject of controlling WhatsApp groups on site.
- Handover. Evidence assembled from four systems and several phones, months after the people involved have moved on. Collation is not the same as archaeology, and the difference is whether anything was recorded contemporaneously.
Is an integration the same as a record?
No, and it is the most common confusion we see. An integration moves data between systems so nobody retypes it. Procore’s APIs power over 500 integrations and that is a genuine achievement. A record is something different: a dated, attributed account of what was decided, by whom, with the original attached, that outlives the software it was created in.
The distinction is getting sharper rather than softer as AI arrives in these tools. An agent that can search everything in a platform is useful and will be confidently wrong about anything that never reached it. The quality of the answer is set by the quality of the record, and no amount of model improvement changes that.
What Construction Metric does about it
We are not asking you to move off your platform. We built the layer that sits under it.
The traffic a project already generates, the messages, the photographs at full resolution, the voice notes, gets captured into a dated record with the sender and the time attached, without anybody doing a second job by hand. WhatsApp capture is what reaches the people who will never hold a licence, which is most of the people on the job. What comes out the other side is a record you own, exportable into the systems you already pay for through our connectors, and into the programme tools a claim actually gets argued in via programme export.
The reason to build it this way rather than as another platform is that it survives a change of platform. Whichever system you are on in three years, the record of what happened in 2026 still has to exist, and it still has to be yours. How it works sets out the whole route, and why every subcontractor should keep their own records covers the same argument from the other side of the contract.
Questions and answers
Does an integration between our systems solve this?
No, and the distinction is worth being precise about. An integration moves data between systems so nobody retypes it, which is genuinely useful. A record is a dated, attributed account of what was decided and by whom that survives the software, the contract and the people who were there. Procore’s APIs already power over 500 integrations, and none of that makes an instruction given in a group chat part of any system at all.
Why does per-seat pricing matter to our record?
Because it decides who is inside the system and who is outside it, and the people outside it are doing the work. A platform gets rolled out to the main contractor’s own staff. It rarely reaches the groundworks gang, the cladding subcontractor two tiers down, the supplier’s driver or the client’s facilities manager. Those people are central to the job and peripheral to the licence count, so the conversation follows them out of the system.
Can AI search our project records for us now?
Better than it could a year ago, and only across what it can reach. Procore announced a class of APIs built for AI and agent use in March 2026, including deep search across drawings, photographs and video. That is a real capability. It searches Procore. If the decision was taken on a phone and never reached Procore, a confident answer comes back with nothing behind it, which is worse than no answer because it stops somebody looking.
Are we supposed to replace our platform?
No. Replacing a platform is a two-year project and it does not fix this, because the gap is between systems rather than inside one. Keep what your clients and supply chain already use. The question is narrower: what happens to the instructions, approvals and changes that never reach whichever platform you are on, and who can produce them in eighteen months.
What does this actually cost us?
It shows up three times. Assembling a delay position from memory and camera rolls rather than a dated record. Arguing a variation that was agreed verbally and confirmed in a message nobody kept. And handover, where evidence has to be reconstructed from four systems and several phones. Each of those is expensive, and each is cheap to prevent while the job is running.
Where should we start on Monday?
Pick one live job and ask where the last three instructions actually came from. If the honest answer involves a phone, you have found the gap, and it is the same gap on every other job you are running. Capturing that traffic into a dated record is a smaller piece of work than any platform migration and it keeps paying after the next software decision is made.
Where to start
Take one live job and ask where the last three instructions actually came from. Not where they were formally recorded: where they came from. If the honest answer involves a phone, you have found the gap, and it is the same gap on every other job you are running.
It is a smaller piece of work than any platform migration and it keeps paying after the next software decision is made. We build this for UK contractors and consultants at AI Metric, and Construction Metric is the product that came out of it. If you want to talk it through against your own projects, get in touch and we will look at it properly.