A familiar pitch, a new industry
Commercial real estate has been a slow adopter of AI, and for a reason that will sound familiar to anyone who has run enterprise IT: the operational data is fragmented across facilities systems, vendor portals, finance, compliance records and building-management platforms that were never designed to talk to each other. A start-up in the space has just raised a Series B on the promise of fixing that — work orders, compliance, maintenance, tenant communication, equipment tracking and finance running through one interface with AI at the core, in place of the patchwork.
Swap the nouns and you have the pitch being made to every industry this year. One assistant over the ERP, the CRM, the ticketing system and the document store. One agent that understands the whole business because it can see all of it. The promise is real, and so is the demand. What varies enormously is what “one interface” turns out to mean.
Two very different architectures behind the same slide
The first version is a front door. The interface sits over the systems the organisation already runs, reads from them through governed connections, and — this is the important part — writes back to them. A work order raised through the AI lands in the maintenance system. An insurance certificate the agent chases is filed in the compliance record. Finance still lives in finance. The systems of record stay the systems of record; the interface makes them usable together.
The second version is a migration. The interface unifies the data by moving it inside the new platform, which becomes the place the work happens. The old systems are integrated in the sense that data is pulled from them, until they are eventually switched off. It is a perfectly legitimate strategy for a start-up — it is how platforms win — but it is a different proposition for the buyer, and the slide rarely says which one is being sold.
For a mid-size property manager the second version may be exactly right. For an enterprise with a general ledger, a regulatory reporting obligation and a warehouse that every other system depends on, it is a replatforming decision disguised as an AI purchase.
What makes the front door actually work
We have built the first version enough times to know where the effort goes, and it is almost never the model.
- Integration to the systems of record, in both directions. Reading is easy. Writing back — with the right validation, the right approval chain, the right audit entry — is where the weeks go, and it is the whole point. An AI that can only read produces a dashboard.
- A shared data model. The interface has to know that the “asset” in the maintenance system, the “equipment” in finance and the “unit” in the tenant platform are the same thing. That reconciliation is a data-warehouse problem, and organisations that have already done it for reporting are far ahead.
- Agents with permissions, not just access. The certificate-chasing agent in the property pitch is a good example: a narrow, well-defined task, acting through a specific system, with clear rules for when it stops. That is a job description, and it is the model for every agent that follows.
- Ownership of the data underneath. Someone has to own the reconciled model, the connectors and the access policy — otherwise the front door decays as the systems behind it change.
The question to ask any vendor
When the pitch arrives — and it will, from a start-up or from an incumbent adding AI to its suite — ask where the data lives after a year. If the answer is “in our platform,” you are buying a migration, and should evaluate it as one: exit terms, data portability, what happens to the systems being displaced. If the answer is “in your systems, with our interface on top,” ask to see the write-back integrations working against your actual platforms, not a demo environment.
Then ask what it does not cover. The best front doors are honest about the systems they cannot reach yet, because those gaps are where the on-site team will keep doing things the old way.
The version that lasts
The single-interface promise is not hype. Unifying fragmented operational data behind one AI-capable layer is, in most organisations, the highest-value thing an AI programme can do. But the durable version is built on the systems you already run, writes back to them, and leaves the truth where the auditors expect to find it. The other version can work too — just make sure you know which one you have bought.
