This is the boring bit that decides if your AI strategy is actually going to work

Occupancy, utilisation, churn, MRR. If your systems disagree on the basics, AI will simply scale the confusion.

Flex operates with one of the most fragmented technology stacks in commercial real estate.

A typical operator runs at least a CRM, a property management system, an access control layer, a billing tool, a community platform, sensors and comms, and each of those tools has its own opinion about what a “member” or a “renewal” or an “occupied seat” actually is.

Most of the time you can work around it.

You sit in a meeting, look at two different occupancy numbers, agree which one to trust for the next twenty minutes, and move on. It’s annoying, it costs time, but it works, because a human in the room is doing the reconciling silently. Judgement is patching the foundation in real time.

The reason that’s about to stop being good enough has nothing to do with AI being smart or dumb.

It’s about what those numbers are for.

They feed the NOI you report to the asset owner.
They feed the landlord report that justifies whatever you’re charging for the space.
They feed the pricing decisions that protect your margin
and, over a long enough horizon, the valuation of the building itself.

And right now, every one of those is being held together by a person in a meeting who knows which version of “occupancy” actually fits the question being asked.

When an AI agent reads the same stack, it doesn’t do that.

It isn’t that an agent can’t reason.
Agents apply probabilistic reasoning constantly, that’s the whole point.
What they can’t do is interrogate the structure they’ve been handed.
They take the definitions and the data as given, and they act on them confidently.

So the silent reconciliation that used to happen in the meeting just…doesn’t. A bad assumption stops being a one off slip and becomes a pattern, the same confidently wrong answer repeated across every report, every forecast, every customer touchpoint, faster than anyone can audit it.

That’s what changes the stakes.
Not that AI introduces a new problem, but that it compounds an old one at scale, with full confidence, and no obvious failure mode.

The meeting where you notice it

Every flex team I’ve worked with has had the same meeting.
Three people in a room, four definitions of occupancy, and a spreadsheet that was fine until someone asked the question.

Is it contracted desks over total?
Physical presence over capacity?
Revenue weighted?
Does a hot desk count?

Everyone in the room is technically right. The reports aren’t wrong. It isn’t a messy team failing to agree, it’s four very sensible definitions colliding.

Some of that disagreement is worth keeping.
If you run mostly private offices you’ll almost certainly measure occupancy differently to someone running mostly hot desks, because you’re running very different businesses. The mistake isn’t the definition varying between companies. It’s letting it vary inside one.

And inside one report that’s going to the asset owner.

Two shapes of the problem

All of this tends to land in two recognisable shapes.

One is when a single word is actually doing two jobs.
One operator I’ve been working with had been using “occupancy” as a single metric. The contracted side, what percentage of their offices had a paying customer’s name against them, turned out to be a different number from the actual use side, what percentage of the desks in the building were under contract. They’d been leaning on one of those and treating it as the whole picture. The number wasn’t wrong. It was a siloed view of what their occupancy actually looked like.

Once they ran both signals side by side, named them as different things, and understood what each really meant, the planning conversation finally had both halves of the picture.

Their approach to driving top of funnel totally changed, they then realised they had more budget to play with than they thought.

And the landlord report they’d been producing started telling a story that actually held together under questioning.

The other shape lives between systems rather than inside heads.
Another operator we work with had a member whose contract was terminated and reissued under a new arrangement. The original contract was being held as inactive in one system and as still live in another. The renewal was being counted in one place and missed completely in another.

The fix here wasn’t an SQL query.
It was the documented agreement about what an “active contract” actually meant in that business, and what counted as a rolling renewal versus a fresh signing. Once that was agreed and communicated, the systems could be aligned to it. Without that, no amount of integration work would have made the numbers make sense, and the MRR figure being reported up to the asset owner was off by enough to make a difference.

All of this is genuinely unglamorous.

But without agreed definitions, every decision built on top of them is at risk. The planning session running on a metric that means different things to different people in the room. The landlord report that doesn’t actually add up, and maybe isn’t questioned until it’s audited, or worse, until the asset owner runs their own analysis and notices. The public inventory feed you push to Valve showing listings that don’t match what’s actually true on the floor.

The internal version costs time.

The external one costs trust, credibility, the margin you’re meant to be protecting, and ultimately the asset value you’re meant to be creating. Which is probably tight enough already.

What it takes to fix, and when not to

We spent the best part of eighteen months on this at Koho.
Not handing operators a universal definition they need to adopt, but getting each operator’s stack to a single agreed set. Occupancy, utilisation, MRR that matches how they actually bill, and how the finance team needs to recognise it, churn that can tell a non renewal apart from a mid term exit.

That outcome’s only possible when you totally understand how an operation functions day to day. Walking through every metric that lands on every report, finding where it’s defined, finding where two systems disagree about it, deciding which is correct for that particular business, because the answer is hardly ever universal, then actually writing it down and treating it as gospel.

It’s worth being honest about when addressing all of this gets easier and when it gets harder.

Early is easier.
A two building operator with eighteen months of trading data has a much cheaper version of this conversation than a multisite operator with eight years of accumulated reporting and a dozen integrations to align. The definitions you set early embed in the team’s vocabulary, the training, the operational habits.

Wait, and now you’re untangling legacy noise alongside the actual decisions, paying for it twice. Once in the unpicking, once in every report that ran on the wrong number in the meantime.

For most multisite operators, the cost has already been quietly piling up.
The landlord report is already arguing with itself, the AI tools are being demoed and deployed, and the data underneath all of it is still disagreeing.

Most operators are implementing the features before they fix the foundations underneath.

The agent gets bought, the demo works, and the failure starts quietly afterwards. An AI reasoning over inconsistent definitions doesn’t fail loudly, it delivers a confident answer that happens to be wrong, and the wrongness compounds across every report and forecast it touches before anyone goes looking for it. And the first person who goes looking is usually sitting outside your operation, the prospect on a demo call, the auditor running their own numbers, the asset owner reading the report you handed up last quarter.

Tons of the conversations I’m in are spent on the tool at the top, the end result, the tangible piece the customer sees and acts on.
Because it’s fun, it’s exciting.

Hardly anyone gets excited about standardising CRM or PMS datasets.

The version of AI flex operators are being sold, the one that takes a workflow end to end, not just drafts the email before it’s sent, needs reconciled definitions underneath it.

Or it will make mistakes.

And those mistakes will land in the report you hand up the chain.

The unglamorous question of what your metrics actually mean was always going to decide what your building is worth.

AI didn’t change that.

It just guaranteed the asset owner gets to the answer before you do.