Buy the Ledger. Build the Brain.

I sat through a platform demo a few months ago where the salesperson spent 11 minutes on the AI.

It drafted an RFI. Summarised a document. Flagged a cost variance. The room nodded along, because it looked like the future, and because everyone in it had a job that involved too much typing.

Then the contracts manager asked whether he could see the licence and ticket register. The one that decides whether the crew can be on site on Monday morning.

It couldn't. That register lives in a spreadsheet on someone's desktop, and the AI only reads what's inside the platform.

Nobody said much after that. But it's the whole industry in one exchange, and I've thought about it a lot since.

So we did the homework on the four platforms a civil, trade or services contractor would actually shortlist: Procore, Autodesk Forma (which is what Autodesk Construction Cloud became back in March), CMiC, and Trimble Viewpoint Vista. The residential and GC-only products are out, because they have no civil or infrastructure story and padding a comparison isn't the same as informing a decision.

We went looking for meaningful differences in the AI. What we found was more useful, and it wasn't about AI at all.

They've all built the same AI

Strip the branding off and line up what these things actually do.

Four vendors. Three jobs: pull structured data out of documents, draft text a human would otherwise type, and flag odd-looking records the system already holds.

That's the category. All of it.

These are the stock capabilities of any current language model. Genuinely hard in 2022. In 2026 they're a fortnight's work against an API. Every vendor on that list shipped much the same feature set within eighteen months of each other, which tells you there was nothing proprietary underneath to slow anyone down. Just a user interface wrapped around a model somebody else trained.

Trimble is the instructive one rather than the embarrassing one. It has a "Trimble AI" umbrella brand. It announced an agentic platform called Agent Studio in November 2025, described then as "coming soon to Labs for select customers". At the same event it demonstrated a Viewpoint Finance Assistant, "expected to be in Labs in early 2026", and we could find no announcement that it ever arrived generally. Meanwhile the Viewpoint and Vista product pages carry no AI messaging at all. The ERP that most needs an operational copilot has one in early access and a help-desk bot in production.

None of which makes these bad products. It means the AI isn't the product. It's a feature, built on infrastructure none of these vendors own, shipped on a schedule none of their customers control.

The bit the vendors write down and nobody reads aloud

All four handle document control, change orders and cost tracking. That's where demos live, and nobody picks a platform on it.

The decision gets made on three capabilities. Here the four of them split straight down the middle.

Every cell there comes from the vendors' own documentation, which matters most for the "No" columns. Those are usually guesswork in a comparison like this. Not here.

Autodesk's own material on ERP integration describes the ERP as the thing that supplies general ledger, job costing, payroll and financial statements. Procore's ERP integrations page exists precisely because Procore doesn't do those jobs. These aren't gaps we inferred from silence. They're gaps the vendors publish.

The other side holds up just as well. CMiC's payroll product promises to "demonstrate compliance by delivering certified payroll reports", and its multi-company material, updated in August, describes one database giving "entity-level and group views from the same source data", with intercompany entries offsetting automatically and joint ventures handled by full consolidation or the equity method depending on who controls what. Trimble's Vista documentation, updated on 2 September, covers payroll across "multiple states" with tax defaulting to either the job's state or the employee's, and reciprocal agreements "between crafts/unions to cover cases where an employee from one local union is working on a job within the jurisdiction of another union".

That last one is the tell. Nobody writes that documentation unless real contractors with real union exposure made them.

Design capability runs the other way, and here I'm giving you a reading rather than a measurement. Autodesk is the deepest of the four on BIM and coordination, which is hardly a surprise given what Autodesk is. CMiC does have a native module, Construct BIM, though its own documentation describes managing and visualising models with version control and mentions neither clash detection nor model authoring. Vista has nothing native. Model work happens in Tekla and Trimble Connect and reaches Vista by integration.

So the two products with real accounting and payroll depth are the two with the least design capability, and the two with the strongest design stories have no general ledger and no payroll whatsoever. Not one platform on that list covers both ends.

Which means that if you're a civil or services contractor with genuine payroll complexity and any design coordination at all, you aren't buying a platform. You're buying two.

And the vendors know. That's the part worth sitting with, because they've put it in writing.

Procore's ERP integrations page lists what it connects to, and the list names Viewpoint Spectrum, Vista and CMiC, alongside QuickBooks, Sage, NetSuite, Acumatica, Workday and Xero. Two of the four platforms in this comparison appear on the third one's integration menu.

Autodesk goes further. Its own published article on construction ERP integration, updated in March, walks through pairing Forma with an ERP and uses CMiC as the worked example, describing it as providing "accounts payable, accounts receivable, billing capabilities, general ledger, job costing, payroll, and producing financial statements". In fairness that's a blog post rather than formal product documentation, and the connection it describes runs through a third-party integration platform rather than anything Autodesk built. Still. That's Autodesk, on Autodesk's site, explaining which competitor you'll need to do the half of the job Forma doesn't.

The recommended architecture, according to the vendors themselves, is two platforms.

Follow that to where it lands. Two systems, each with its own AI, each AI sealed inside its own vendor's data model, neither able to see the other's records.

You didn't buy one intelligent platform. You bought two confident fragments.

What actually happens when you buy AI as a feature

You inherit someone else's roadmap. The AI ships when their release cycle says, does what their largest customers asked for, improves at their pace. You don't get a vote. Trimble's finance assistant has been coming to early access since last November.

The AI only sees their data. Your licence register is in a spreadsheet. Tender packs arrive as email attachments. Subbie insurance certificates sit in a shared drive that three people can find. Payroll is in another system entirely. So the AI is blind to most of your actual operation, and confident about the sliver it can see, which is worse than being quiet.

Then there's the awkward one. Your differentiator is exactly what off-the-shelf AI won't touch. You don't win work because your change orders are well documented. So are your competitor's, on the same software you're both paying for. You win on how you price risk, how fast you turn a tender around, how reliably your crews stay ticketed. None of that is a standard module, and no generic agent drafting a generic RFI moves any of it.

And the AI turns up before your data is ready for it. Every feature in that table assumes clean, complete, consistently coded records. The reason the demo sings and your rollout doesn't is that the demo runs on a tidy fictional project. Getting your own data into a state an agent can reason over is the actual work, it's yours either way, and buying the AI doesn't do it for you.

What building the AI layer actually means

I'm not suggesting you build Procore. That would be idiotic, and it's where this argument usually gets strawmanned.

The systems of record are worth buying. General ledger, certified payroll, multi-entity consolidation, BIM coordination. That's decades of accumulated regulatory and engineering detail, and you should pay someone else for it. CMiC's union reciprocity handling and Vista's multi-state payroll are not weekend projects, and anyone who tells you otherwise has never run a prevailing-wage job.

Buy the systems of record. Build the intelligence layer that sits across them.

If you're going to end up running two platforms anyway, on the vendors' own advice, then the gap between them isn't an inconvenience to be integrated away. It's the only place in your stack where something can be built that knows your whole business. That's where the leverage is. It's also the one part nobody is selling you.

In practice the layer is your documents, registers, inboxes and platform data pulled somewhere an agent can reason over all of it, with the specific jobs your business needs automated running on top. Three from our own work:

The structural difference is in what you're buying. A platform sells you licensed seats or contracted volume, renewed annually, used or not. A layer you own runs on consumption. It costs something when it does work and nothing when it sits there.

You also get portability. Our builds let the client connect their own AI provider and key, so when a better or cheaper model ships, and one always does, you switch. You're not waiting on a vendor to re-platform their assistant.

The honest framework

If you need certified payroll and multi-entity consolidation, you're shopping between CMiC and Vista, and you should walk in expecting an implementation measured in quarters rather than weeks. If design coordination is the centre of your work, Forma is the deepest of the four. If it's project and field management across a broad portfolio, Procore is the most widely reviewed by a distance.

Then stop.

Don't buy a third platform to plug the gap between the first two. That gap is where the AI layer goes, and it's the one part of your stack that can genuinely be yours.

What I'd do in the first 90 days

1. Pick the two workflows costing you the most hours this month

   Not the most impressive ones. The most expensive ones. Nine times in ten it's tender response and compliance chasing.

2. Map where that data actually lives

   It'll be four places, and one of them will be a spreadsheet somebody maintains by hand and quietly worries about.

3. Then build one

   A single workflow, end to end, in production, measured against the hours it used to eat.

4. Only then decide what to buy

   You'll have a much sharper view of what a platform genuinely needs to do once the intelligence layer is off the shopping list.

The vendors aren't wrong that AI belongs in construction. They're wrong about who should own it.

Clickpoint Consulting builds AI operations layers for trade, civil and service contractors. Compliance tracking, tender automation and document intelligence that sit across the systems you already run. If you're halfway through a platform evaluation, we'll happily tell you which parts of it to buy.