Why Most Strata AI Pilots Fail (And What the 14% Get Right)

Most AI pilots in strata don't fail because the AI is bad. They fail because nobody thought hard enough about the strata.

That's the uncomfortable conclusion when you stack the numbers up. A March 2026 survey of 650 enterprise technology leaders found 78% have AI agent pilots running, but only 14% have reached production scale. MIT's NANDA Initiative put it more bluntly: 95% of enterprise generative AI pilots deliver no measurable financial impact. S&P Global tracked the wreckage from the other side: 42% of companies abandoned most of their AI initiatives in 2025, up from 17% the year before.

Strata is not immune to this. If anything, strata is the worst possible terrain for a generic AI pilot.

Why strata breaks generic AI

The Digital Applied research clusters scaling failure into five root causes that account for 89% of stalled deployments: legacy integration complexity, inconsistent output quality at volume, no monitoring tooling, unclear ownership, and insufficient domain training data.

Now picture an Australian strata business. The platform is StrataMax or Strata Master or Mastract, none of which were designed with modern APIs in mind. The data is split across email inboxes, PDFs, accounting ledgers, and the manager's head. Compliance varies by state. Ownership of any given decision is split between the manager, the committee, the chair, and sometimes the developer. And the domain knowledge — what a Section 106 certificate is, when a special levy needs an EGM, how to read a sinking fund forecast — is mostly tacit.

A horizontal AI tool dropped on top of that environment doesn't have a chance. It hits all five failure modes on day one.

This is why MIT found that buying AI from specialised vendors succeeds about 67% of the time, versus 33% for internal builds. It's not that internal teams are worse engineers. It's that they're solving the AI problem when the actual problem is integration, workflow, and domain.

The scale of what's being missed

Australia has over 368,000 strata schemes covering 3.19 million lots, served by roughly 3,900 strata managers. That's about 820 lots per manager, on average. In WA alone, strata property is valued at around $112 billion, with 10% of the state's population living in it.

The MRI Software Voice of the Strata Manager report from 2023 found that 60% of Australian strata managers work beyond a 38-hour week, 19% are working 51+ hours, and owner communication eats 60–80% of the working day. A third change employers in any given year.

The point isn't that AI should rescue an overworked workforce. The point is that the workload makes a failed AI pilot expensive in a specific way: every hour spent rolling out a tool that doesn't ship is an hour that didn't go to actual owners, actual buildings, actual levies. Strata businesses can't afford a 12-month pilot that produces a slide deck.

What the 14% actually do differently

Stanford's Digital Economy Lab analysed 51 enterprise AI deployments that delivered measurable value. The pattern was consistent: none used waterfall planning, all used iterative "start small, learn, expand" methodology, and the failures across the broader sample stemmed from workflow integration and organisational incentives, not model quality.

Translated into strata, that means three things.

One: pick a workflow narrow enough that quality can be measured weekly. Not "AI for the whole business." A specific lane — levy arrears chasing, maintenance triage, owner email responses — where you can count what the agent did and what a human would have done.

Two: integrate before you generate. The agent has to read from and write to the strata platform on day one. If the workflow ends with "and then the manager copies the answer into StrataMax," you've built a chatbot, not an agent.

Three: assign ownership. One person inside the business owns the rollout, sets the quality bar, and decides when to expand the agent's remit. Not a committee. Not "the technology partner." A named operator.

The Perth case

We deployed our agents inside one of Perth's largest strata groups starting with a single workflow: inbound owner emails. Read the email, understand the intent, draft the response, file it against the right lot and scheme, escalate the ones that needed a human. We didn't try to do levies, maintenance, and meetings all at once.

The reason that worked wasn't the model. It was that we'd already built integrations into their platform, the agent was trained on strata-specific language and compliance rules, and there was one operations lead inside the business who owned the rollout. When something looked wrong, we knew within a day. When something worked, we expanded it.

That's the unglamorous version of what the 14% do. Pick one thing. Wire it into the system properly. Give it to one person. Measure it. Then add the next thing.

The claim

Generic AI will keep failing in strata for the same reason it's failing everywhere else: it's being deployed as technology when the work is operations. The businesses that get value out of AI in this sector over the next two years won't be the ones with the biggest pilot budgets. They'll be the ones who treated AI as a vertical operations problem from the start.

If you're working out where AI fits in your strata business and want a straight conversation with people who've actually shipped it into production, book a 30-minute call. No deck, no pitch — just a working session on what would make sense for your operation.