Property Meld can already predict (for customers on the Property Meld Ops plan) what most repairs should cost before a technician or vendor is dispatched. Now, we’re focused on making those predictions even more actionable.
And as we’ve dug deeper into that work, we’ve learned something important: greater precision doesn’t just improve the cost estimate. It creates new opportunities to make better decisions throughout the entire maintenance process.
Here’s what we’ve learned recently
Accurately predicting the cost of a repair requires much more than historical invoice data. You need to understand what was reported at intake, how the issue was diagnosed, what happened throughout the repair, and how it was ultimately resolved.
That level of detail matters.
At intake, Property Meld can classify repairs across more than 6,700 subcategories. That gives our models the context to begin determining whether something is more likely to be repaired or replaced, the probability of each outcome, and what each outcome should cost.
And the more we learn, the more interesting the possibilities become.
If we can understand what a repair is likely to cost before dispatch, could we anticipate when investor approval will be necessary before it causes a delay? Could we determine whether an internal technician or vendor is likely to be the better choice? Could we eventually recommend the provider most likely to deliver the best outcome based on cost, repair history, speed, and resident satisfaction?
Those are some of the questions our team is exploring now.
So, why hasn’t this existed before?
We’re learning that the hardest part isn’t necessarily building the AI. It’s building the data foundation the AI can trust.
To make reliable predictions, you need clean, structured information across the entire repair journey, from intake and diagnosis through work performed and final resolution. Then you need enough volume within those classifications for a prediction to actually mean something.
The combination of data cleanliness, depth, and volume creates an incredibly high bar.
Even at Property Meld, where we’ve been focused exclusively on maintenance data for years, our data science team has continued improving how information is captured and classified to make this possible.
And we’re beginning to see what that foundation can unlock.
Many customers using Property Meld Ops can already see predicted costs for more than 80% of repairs before dispatch. We’re continuing to tighten those predictions, but what excites us most is what becomes possible as they get better.
Because knowing what a repair should cost is useful. Knowing it early enough to influence what happens next is much more powerful.
It creates the opportunity to move maintenance intelligence upstream, helping operators make better decisions about approvals, assignments, vendors, technicians, and costs before those decisions become invoices.
We’re still early, and there’s a lot left to learn. But each improvement gives us a clearer picture of where maintenance operations can go next.
We’ll keep sharing what we learn along the way.