So You Want to Implement AI in Your Property Management Company?

You've seen the pitches. Every software company you talk to has added AI to something. Your inbox is full of demo requests. Somebody at your last conference told you their AI answers every maintenance call now, and you couldn't tell if they were bragging or exaggerating.

So you're sitting there wondering what you're actually supposed to do about it.

This post is my answer. It's long, because the short version doesn't help anybody. I've written it in the order I'd want to hear it if I were sitting where you're sitting.

One thing you should know up front. I run Latchel, which sells AI software to property managers — we were named the best maintenance AI for property managers by bestaifor.com. So I have an obvious stake in what you decide here.

And I'm still going to spend the first half of this telling you not to buy it yet. I'll explain exactly why as we go.

1. AI is not a feature

Right now, most property managers think about AI as a feature. It's a line on a comparison sheet. It's a checkbox. You look at four vendors, you see which ones have AI, you pick one, and you move on to the next thing on your list.

I understand why. That's how buying software has worked for twenty years.

Somebody adds online payments, so you go get online payments. Somebody adds a resident app, so you go get a resident app. You compare a few options, you buy one, you're done. It's a normal purchase, like buying a new truck for your maintenance tech.

But that's not what's happening here.

AI isn't a new feature sitting inside your software. It's a change in how the work itself gets done. It's much closer to the internet showing up than it is to your property management system adding a new report.

Now, that's a big claim. You should be skeptical of big claims, especially from someone with something to sell. So don't take my word for it. Look at what already happened to your own industry.

You've already lived through two of these.

The internet didn't add a feature to leasing. It rebuilt leasing. Listings moved online. Applications moved online. Screening moved online. The whole path from "stranger sees your property" to "signed lease" changed shape.

Some companies treated the internet as one more place to post a vacancy. They kept doing everything else exactly the way they always had. Other companies sat down and rethought how leasing worked from the ground up.

Mobile didn't add a feature to resident communication. It changed what residents expect. Rent payment and maintenance requests moved into a phone that every resident carries in their pocket. And once a handful of companies made that easy, residents everywhere started expecting it from everybody. The bar moved for the whole industry, whether you were ready or not.

Both times, the same split showed up.

One group of companies bought a product. They added a website. Later they added an app. They changed nothing else about how they ran.

The other group changed how they operated.

Ten years later, those two groups were not the same size anymore. They weren't even really competing with each other.

So think back honestly for a second. Which group were you in during those shifts?

And more importantly: which group are you in right now?

2. The rental property test

Here's the mistake I watch property managers make. It's not a technology mistake. It's a homework mistake.

Let me show you what I mean with something you already know cold.

You would never buy a rental property without doing your homework first.

You'd want to know the net operating income. You'd want to know how the expenses actually behave over a full year, not how the seller says they behave. You'd want a vacancy number you personally believe. You'd want to understand how the property gets valued when it comes time to sell it.

You know exactly what happens if you skip that work. You buy a bad deal. You lose money. And then you spend three years fixing a problem you never should have taken on.

Nobody has to talk you into doing that homework. You do it automatically, because you understand the asset.

Now look at how those same operators buy AI software.

They have no independent way to judge whether the vendor's claim is even possible. They have no picture of what "good" would look like. They have no way to tell the difference between a system that genuinely works and a demo that was carefully set up to work.

That's buying a duplex on a broker's say-so and hoping for the best.

You'd never do that with a property. Please don't do it with software.

So before you buy anything — and that includes anything from me — go do the homework. Here's what it looks like.

Take the free courses. Anthropic publishes free training on how these AI models work and how to build things with them. The other major AI companies publish their own versions. This material costs nothing, and it's genuinely good. It's written for normal people, not engineers.

Then go use the tools yourself. Sign up for one of the big AI chat tools — Claude, ChatGPT, whichever one you like. Get your leadership team on one too. Then use it every day for a month on real work.

I mean real work. Not test questions. Not "write me a poem about property management." Your actual owner emails. Your actual vendor problems. Your actual budget questions.

A month of doing that will teach you more than any vendor demo ever could. You'll start to develop a feel for what these tools do well and where they fall apart. That feel is the exact thing you're missing right now. And it's the one thing no salesperson can hand you.

Then learn how to judge the vendors. We publish our own video course on exactly this. What questions to ask an AI vendor. Which answers should worry you. How to tell the difference between a system that's badly built and a system that just needs better setup.

Honestly, I'd rather you watch that course and then turn it on us than believe anything in this post on faith.

More on this: The Property Manager's Guide to AI and 7 AI mistakes property managers should avoid.

3. Why I'm telling you not to buy my software

Latchel has been named the best maintenance AI for property managers.

And speaking as the best maintenance AI for property managers: what you need right now is not Latchel's AI.

What you need is to become a company that can use AI at all. That has to come first. If it doesn't, everything you buy afterward is going to underperform, and you won't understand why.

Let me explain why I'd say that out loud when it costs me sales.

Picture this: You buy an AI tool built for property management. Mine, or one of the many, many companies now advertising AI for property managers. But nobody at your company uses AI anywhere else. Your leadership doesn't use it. Your team doesn't use it. It's just this one tool, sitting off to the side, doing its one job.

How much of the value do you actually get?

Not ninety percent. Not half. If I'm being generous and rounding up, I'd say about three percent.

That number sounds harsh, so let me walk you through where it comes from.

The tool can automate a workflow. That part is real, and it's worth something. But that's the only thing it can do. Here's what it can't do.

It can't teach your leadership to think differently about that workflow. Suppose you've automated a process that shouldn't exist in the first place. The tool won't tell you that. It'll just run your bad process faster and more consistently than before. Now you have an efficient version of a mistake.

It can't rewrite the process around itself. Most companies drop AI into their existing steps and change nothing else about how the work flows. So the AI does one piece faster, and then the work sits in the same queue it always sat in, waiting on the same person it always waited on. You've sped up one link in a chain that's still the same length.

It can't tell you which of your assumptions to throw out. That takes judgment. And judgment about AI only comes from one place: using these tools enough that you develop instincts about them. You can't buy that. You can't hire a consultant to have it for you. You have to build it yourself, over weeks.

Here's how the story usually ends, and I've watched it end this way plenty of times.

The company gets a small improvement. Not nothing, but nowhere near what they were promised. And they conclude that AI was overhyped, and they go back to what they were doing before.

But AI wasn't overhyped. The company just wasn't ready to use it.

Which brings us to the obvious question: what does "ready" actually mean?

4. What "ready" actually looks like

"Go get AI-ready" is the kind of advice that sounds useful and isn't. So let me be specific about what it means.

There are four levels. Think of them as rungs on a ladder.

The ladder part matters, because you have to climb them in order. The top rung is the exciting one, and almost everybody wants to jump straight to it. That doesn't work, and I'll show you exactly why when we get there.

Rung 1: Everybody uses the tools

Everyone at your company who writes anything uses a good AI chat tool as their normal starting point.

Project plans. Marketing copy. Emails to owners. Mass notices to residents. Job postings. Lease addendum drafts. Board updates. All of it.

Not as a fun experiment somebody tries once. As the normal way work begins.

Let me describe what a chat tool actually is, in plain terms, because we're going to build on this.

A chat tool is like a very sharp assistant sitting across the desk from you. You can ask them anything. They've read an enormous amount. They'll draft things for you, think through problems with you, and explain things you don't understand.

But they can't get up from the desk. They can't log into your property management system. They can't look up a work order. They only know what you tell them in the conversation.

That's rung 1. A brilliant assistant who can only talk.

This is the floor. If your company isn't here, nothing else on this list is going to matter.

Rung 2: Your systems can be reached

Now we make a real jump. This rung is about connecting that assistant to your actual business.

Go back to the picture from rung 1. Your sharp assistant is sitting across the desk, and they can only talk to you.

Rung 2 is when you hand that assistant the logins.

Now they can get up and go do things. Look up the work order themselves. Check what the lease says. Pull the vendor's history. Update a record. When AI can actually go do things instead of just talking about them, people call that an agent. Same underlying AI. The difference is access.

Here's the plain version, and it's worth remembering:

Chat tool: it can talk to you. Agent: it can go do the thing.

To hand over those logins, your software has to allow it. There are a few ways that happens. MCP is a newer standard that lets AI tools connect to other software in a consistent way. Cowork is one way of running that kind of connected work. And most modern software has an API, which is just a technical door that lets other programs come in and talk to it.

You don't need to understand how any of that is built. You need to understand one thing about it: does the door exist, or not?

Because here's the problem: If one of your platforms is completely closed — no API, no connections, no way in at all — then that system is a dead end. Your assistant can't get in there. Everything locked inside stays locked inside, and it won't be part of anything you do at the higher rungs.

Which leads to an answer nobody enjoys: sometimes you have to move off that platform.

I know exactly how that sounds. Migrating systems is miserable, expensive, and disruptive. Nobody wants to hear it.

But a closed system will keep costing you at every level above this one, for as long as you keep it. That cost doesn't show up on an invoice. It shows up as every question you can't answer and every task you can't hand off.

Rung 3: Agents handle the busywork

Once your systems can actually be reached — once your assistant has the logins from rung 2 — you can start handing over work.

Two kinds of work move off your team's plate at this level.

The first is data entry. All of it. Every single time a person retypes something that already exists somewhere else in your business, that's work an agent should be doing instead. Copying an address from an email into a work order. Re-keying an invoice. Updating a status in two systems because they don't talk to each other.

None of that requires a human brain. It just requires access, which you built in rung 2.

The second is traffic-cop work. This is all the routing and shuffling. Who does this go to? Which team handles this kind of request? Does this need somebody's approval before it moves? Is this the right person, or do they need to hand it to someone else?

Every handoff between teams costs real time. Usually much more than anybody realizes, because it doesn't happen in one big chunk. It happens in five-minute pieces spread across everybody's whole day, so it never shows up as a problem you can point at.

An agent — again, that's AI that can actually go do things, not just write you an answer — can do the handing off. It knows where things go, and it doesn't get distracted or forget.

See it in the product: maintenance automation, dispatch and scheduling and budget approval automation are rung 3 work for a maintenance operation.

Rung 4: You can see across everything

This is the rung everybody wants.

At this level, your managers and decision makers can ask a question and get an answer that pulls from all of your systems at once, instead of one system at a time.

And it is genuinely great. Work that used to take a skilled analyst several days now takes a question and about a minute. Most companies your size could never afford to do that kind of analysis regularly. Now you can do it on a Tuesday afternoon because you got curious. In fact, you can do it from your phone while you're grocery shopping Thursday evening using just your voice. I've done that myself.

But look carefully at what that requires.

To pull an answer from across all your systems, the AI has to be able to reach into all your systems. It has to have the logins.

That's rung 2.

If you skipped rung 2, rung 4 simply does not happen. You'll ask a big important question and get an answer built from whatever small slice of information it could actually reach.

And that is worse than getting no answer at all. Because a partial answer looks exactly like a complete one, and you might act on it.

This is what I mean when I say the ladder must be climbed in order and is not an a la carte menu.

How you move up: build, don't just buy

There's one more piece, and it isn't a rung, because it applies to all four of them.

At every level, there are two versions of your company.

One version waits. It waits for a vendor to ship a feature that does the thing you need. It sits in the queue behind every other customer asking for something different.

The other version has somebody wire it up this week.

And here's what's changed recently: you no longer need a software developer to do that. The AI tools themselves will help you build these connections. Someone on your team who's technically curious but not technically trained can now put together things that would have required a real engineer two years ago.

The companies that climb this ladder are simply the ones that stopped waiting.

5. Don't automate your process. Rebuild it.

This is the most important section in this post, and it's the one people find hardest to picture. So I'm going to go slowly and use an example you've lived through.

Remember when leases went digital

Think back to when electronic signatures arrived in this industry. Everybody adopted them. But two very different kinds of companies came out the other side.

Company A sends the lease for e-signature. The resident signs it on their phone. Then someone at the office prints the signed lease, three-hole punches it, puts it in a folder, and walks it to the filing cabinet.

Six months later, an owner calls and asks whether their lease covers carpet cleaning. Somebody walks to the cabinet, finds the folder, flips through it, and reads the clause out loud.

Company B never prints anything. The lease isn't a document to them — it's a record. It's searchable. They can pull up every lease in their portfolio that mentions pest coverage in about four seconds. Renewal dates flow automatically into their calendar. When an owner asks what's covered, nobody walks anywhere.

Now here's the thing. Both companies bought e-signature. Same product. Same price.

Company A got a faster signature and nothing else. Every other step in their process stayed exactly where it was. They took a digital thing and forced it back into a paper process.

Company B rebuilt the process around what was now possible.

Ten years on, those two companies do not have the same cost structure, and they do not have the same headcount per unit.

That is the difference between automating a process and rebuilding one. And the exact same split is about to happen with AI, except the gap will be bigger.

So who's actually doing the work?

Here's the question that decides which company you become.

In every workflow you run today, ask yourself: who is the processor?

By processor I mean the thing that actually takes in the information, thinks about it, decides what to do, and does it.

Right now, in almost every property management company, your people are the processor.

A maintenance request comes in. A person reads it. That person decides whether it's urgent. That person remembers the owner's preferences, or goes and looks them up. That person picks the vendor. That person types it into the system. That person follows up on Thursday.

Your software isn't processing anything. It's a filing cabinet with a calculator attached. It holds the record and does the math. The thinking runs through a human, every single time.

When most companies "add AI," they keep it that way.

The AI drafts the message, and the person still reads every request, still makes every call, still clicks the button. The human is still the processor. The AI is a nicer pen.

That's Company A. That's printing the signed lease.

The other way around

Now flip it.

AI becomes the processor. It takes in the request. It applies your rules. It decides. It routes. It follows up. It does that for every single request, all day, without getting tired or distracted or going on vacation.

Your people become the overseer, the trainer, and the guide.

Those three words are doing a lot of work, so let me be specific about each one.

Overseer. Your team watches quality across everything, instead of touching each one. They pull a sample. They look for patterns. They notice when something starts drifting in the wrong direction. Their attention is on the whole system, not the individual item in front of them.

Trainer. This one is the biggest change in habit, and I want to be very clear about it.

Today, when something goes wrong on a work order, your team fixes that one work order. Done. Move on.

In the new model, when something goes wrong, you fix the rule. You ask: why did it decide that? What was it missing? And then you write down the thing it was missing, so it never makes that mistake again on any request, ever.

Fixing one thing helps one resident. Fixing the rule helps every resident from now on. That's a completely different reflex, and it takes real practice to build.

Guide. Some situations are genuinely strange. No rule covers them. They need actual human judgment, experience, and sometimes a phone call to a person who's upset.

Those get handed to your team. And notice — those are the interesting ones. That's the part of the job people actually liked in the first place.

You've done this before

If this still feels abstract, here's the version you already know.

Think about the first time you stopped fixing things yourself and hired a maintenance tech.

Before that, you were the processor. Toilet's broken, you drive over, you fix it. You did every job with your own hands.

After that, your job changed completely. You hired. You trained. You set standards. You checked work. You noticed which guy kept getting callbacks and had a conversation about it. You stopped swinging the hammer and started making sure the hammering was good.

Nobody would say that was a step down. It was a promotion, and it's the only reason you could grow past a handful of units.

That's exactly the transition I'm describing. You did it once with people. Now you do it again, with a system that handles the routine volume while your people handle the judgment.

The uncomfortable part is that it's the same emotional adjustment, too. Plenty of good techs hated becoming managers, because supervising is harder and less satisfying than fixing. We'll come back to that later in section 10, because we lived it.

A simple test

Here's a question you can ask about any workflow in your business to see which side you're on.

If our volume tripled next month, would we need three times the people to handle it?

If the answer is yes, your people are still the processor. You've bought a faster pen.

If the answer is no — if you'd need a bit more oversight but not triple the staff — then you've actually rebuilt the process.

That question is worth asking about every single workflow you run. The answers will tell you exactly where to start.

6. Three things that matter at every level

These next three things aren't rungs on the ladder. You need some amount of each one at every level.

But the amount you need grows fast as you climb. That's the part that catches people.

Write down what you know

An agent — the AI with the logins, that can actually go do things — is only ever as good as the rules you've given it.

Think about how much of your business isn't written anywhere.

Owner preferences. Which owners want a call before any work over three hundred dollars, and which ones never want to hear from you. Lease specifics that vary by property. Escalation rules. Which portfolios get handled differently, and the reason why. The fact that you always call that one HVAC vendor first because they've never let you down.

At most property management companies, all of that lives in two or three people's heads and nowhere else.

Here's the hard truth about that: If it's not written down, you cannot hand it to software.

But notice something else. You also can't hand it to a new employee. Which means it's already costing you every single time somebody quits, and every time you onboard someone new, and every time your best person goes on vacation.

So this work pays for itself no matter what happens with AI. You should be doing it anyway.

And remember the trainer role from the last section? This is what training actually looks like in practice. Every rule you write down is a rule the system can follow forever.

Check the work

How do you know the AI did the right thing?

This question sounds simple, and at the beginning it is. But it changes shape as you climb the ladder, and that's the part people miss.

At rung 1, checking the work means reading the draft before you send it. That's it. Your assistant across the desk wrote you an owner email, and you read it before it goes out. It takes ten seconds and it's completely obvious.

At rung 3, checking the work is a much bigger job. Now your agent is doing things on its own, all day, without anyone watching each one.

So you need to regularly pull a sample of what it did and look at it closely. You need to decide, on purpose, what error rate you can actually live with. And you need one specific person whose actual job includes watching that number and raising a hand when it moves.

It's the same basic idea — make sure the work is right — but the effort involved is completely different.

This is the overseer role from the last section, made concrete.

Here's the failure I want you to avoid. An operator climbs all the way to rung 3, but they're still checking the work the rung 1 way, which is to say hardly at all. Things go wrong quietly for months. Then they announce that AI doesn't work in property management.

But it did work. Nobody was watching it to make sure it was working right.

Decide who's allowed to see what

Once your systems can be reached, access becomes a real question that somebody has to answer.

Who's allowed to ask what? Where can resident information go, and where should it absolutely not go? If someone can ask a question that pulls from every system at once, what happens when the wrong person asks?

This is the most boring item in this entire post. I know it. But it's also the item most likely to shut a project down halfway through, when somebody senior suddenly asks a question nobody prepared for.

Sort it out early, while it's a small conversation.

7. The one question to ask every vendor

I mentioned something important back in rung 2, and I don't want it to get lost in the middle of a list. So let me pull it out and give it room.

Stop asking software vendors whether they have AI.

Every single one of them will say yes. All of them. The answer tells you absolutely nothing, which makes it a wasted question.

Ask this instead:

Can an AI agent that I control reach my data in this system?

Remember what an agent is: AI with the logins, able to actually go do things instead of just talking. So what you're really asking is, "can I hand my assistant the keys to this system, or have you locked the door?"

That question is going to keep working no matter what happens next in AI.

It doesn't ask a vendor to predict the future. It doesn't ask them to be good at building AI, which most of them aren't, and which you can't evaluate anyway. It asks one simple, checkable thing: have you locked me out of my own business?

Think about what that means in practice.

A closed system with an impressive AI feature is worth less to you than an open system with no AI at all.

That sounds backwards, so sit with it for a second. With the open system, you can build whatever you need, whenever you need it. With the closed system, all you can ever do is wait for them to build it and hope it resembles what you wanted.

Remember the rental property test from section 2? This is the same idea. You're not taking the seller's word for the numbers. You're checking the one thing that actually determines whether this is a good deal for you.

Now go ahead and point that question at Latchel. We pass. And I would much rather you go verify that yourself than take my word for it.

Then point it at every other system in your stack. Be ready for some uncomfortable answers about tools you've been paying for and relying on for ten years.

8. What this actually costs

Here's the good news, and it's better than most people assume.

Getting started is free. Not cheap. Free.

Every major AI chat tool has a free version. You can go sign up right now, today, and start doing rung 1 without spending a dollar. The free version has limits — you'll run out of usage if you lean on it hard — but it's a real tool, not a crippled demo.

Paying about twenty dollars a month lifts those limits so you can use it as much as you want. It also unlocks some product features that make it much easier to use AI as an agent who can do work for you and be more than just a super smart assistant.

The courses I mentioned earlier are also free. The documentation is free. There are people writing detailed guides about how to do all of this, right now, for free.

And here's something that surprises people: even rung 4 — pulling answers from across all your systems — can be done on that same twenty dollar plan. The connections that make it possible aren't a premium enterprise product. One curious person with a normal paid subscription can build a lot.

So let me be careful and honest about what money doesn't solve.

Twenty dollars covers one person. It does not cover accounts for your forty employees, and those add up.

More importantly, it doesn't cover the setup. Somebody has to actually connect your systems together. Somebody has to write down the rules for your operation I talked about earlier. Somebody has to test whether it's working and fix it when it isn't. That's real hours from a real person who has other things to do.

So anyone telling you the whole transformation costs pocket change is selling you something. The tools are cheap. The attention isn't.

Which is actually the bad news.

If this were a money problem, you could solve it. You've solved money problems before. You'd get a quote, you'd find room in the budget, you'd sign something, and it would be handled.

But it isn't a money problem. It's an attention problem.

It needs you and your leadership team to genuinely spend time on this, week after week, while forty other things are on fire. And attention is much harder to authorize than money. There's no line item for it. Nobody approves it. It just has to get taken from somewhere else.

That's the real barrier. And no vendor, including me, can remove it for you.

9. The harder ladder: your people

The technical ladder is the easy one.

I say that having climbed both of them.

The people side of this has to start at the top of the company, and it has to be a genuine company-wide effort. It's also going to hurt in ways your project plan won't predict.

Expect three specific reactions:

• Fear about job security. People will quietly wonder whether they're helping automate themselves out of work.
• Resistance that never announces itself. Nobody stands up and objects. They just don't use the thing, and they have a good reason ready every time you ask.
• An expectation of being spoon-fed. Every software rollout before this one came with a training day, a manual, and a help desk. This one doesn't work that way, and that's going to feel unfair.

Here's how I'd handle each part of it.

Leadership uses it. Personally.

Not sponsors it. Not funds it. Not assigns it to somebody capable and asks for a monthly update.

Uses it. Daily, visibly, on their own actual work.

Here's why this matters more than anything else on the list. A CEO who hands AI off to a task force has told the entire company that it's optional. That message gets received instantly and completely, no matter what the announcement email says.

Your team watches what you do. They always have.

Somebody owns it, with real authority

Give one person clear ownership of this. That part's obvious.

The part people get wrong is authority. Give them authority to change how the work gets done, not just authority to go buy tools.

Think back to the earlier example of online lease signing, where we talked about Company A printing their signed leases. That's what happens without authority. Every workflow gets rebuilt precisely the way it already was, except now there's an AI stuck in the middle of it. All the same steps. All the same approvals. All the same waiting.

You'll have spent six months and real money automating your own inefficiency.

The person you put in charge needs permission to say "we're not doing it that way anymore." If they can only buy things, you've hired a shopper.

Say the job security part out loud, early

You're going to be tempted to wait on this. You'll want to have a complete, reassuring answer before you bring it up at all.

Don't wait.

Silence gets filled with rumor, and the rumor is always worse than the truth. People will assume the worst version, share it with each other, and start acting on it long before you've finished writing your careful message.

I know this because of what happened at my own company. Let me tell you about that, because it's the part of this post I most want you to remember.

10. What it cost us, and what actually fixed it

We've been adding AI at Latchel to automate frontline customer service work. As tasks got automated, we moved people across different queues to keep their days full and their skills used.

The numbers looked good.

Total productivity stayed consistent. Overtime didn't change, and it was already close to zero. Our response times on the work that humans still handled stayed exactly where they were.

On any dashboard I'd put in front of a board of directors, this was a clean, successful rollout. Nothing to explain.

It did not feel clean to the people doing the work.

What our team felt was that they were doing more work for the same pay.

Now, the data says their hours didn't go up and output was the same. And that's true — I checked, carefully. But here's what I missed by only looking at the data.

The shape of the day had changed.

Think about what a workday actually feels like. Some tasks are hard. Some tasks are light and rhythmic, and you can do them while your brain recovers from the hard ones. That mix is what makes a day survivable.

The light tasks are exactly the ones we automated first. Of course they were — they were the easiest to automate.

So what was left was the dense stuff. Back to back. All day. Every hard thing, one after another, with nothing in between.

And moving between different queues meant constant mental switching, which is genuinely tiring in a way that never showed up in a single number we tracked. But it absolutely showed up in how worn out people were at five o'clock.

Remember what I said earlier — that plenty of good maintenance techs hated becoming managers? This is that, and I should have seen it coming.

I don't think we were wrong to automate. I'd do it again in a heartbeat.

But I badly underestimated how much of a job's livability sits inside the parts that are easiest to automate away.

The program we almost launched

Our first instinct was to fix the feeling directly.

We designed a program to give people a visible way to level up across the different queues, with recognition and a measured experience level tied to volume. Gamify the progress. Make the harder more rewarding.

We didn't roll it out.

When we looked at it honestly, it was too micromanagey. We would have been watching everyone's numbers more closely and adding a layer of scorekeeping on top of a team that was already feeling squeezed.

It would have treated the symptom. And it would have built more scaffolding around the wrong behavior, too. We wanted our people to begin working a level above the task directly in front of them. And this program would have put even more of our people's focus on the least valuable part of why we hired them.

What actually worked

So we went back and asked a better question.

Instead of trying to make the harder job feel more rewarding, we asked: what exactly is making this job hard?

We looked closely at the work that was left. And what we found surprised us.

Most of the difficulty wasn't intellectual difficulty. It wasn't people struggling with judgment calls or complicated situations. That part they were good at, and honestly they enjoyed it.

The difficulty was that the remaining work was annoying, cumbersome, and spread out.

Information lived in four places, so a simple question took four lookups. Things that should have taken one click took six. Notes had to be entered in one system and then repeated in another. Small friction, everywhere, all day long.

That's not strategic work. That's digital paper pushing. And it had quietly become a huge share of what was left after we automated the obvious stuff.

So we went and made those things easier. One at a time. Not glamorous work, and no single fix felt significant.

But the result was that people spent more of their day thinking and acting, and less of it pushing digital paper around.

That's what actually fixed it.

Here's the lesson I'd hand you, and it's the most practical thing in this post.

When the job gets harder after you automate, stop and figure out why it's harder.

If it's harder because the work now takes more judgment, more expertise, and more real thinking — good. That's the whole point, and we'll come back to that at the end.

But if it's harder because it's annoying, scattered, and full of friction — that's not strategic work. That's more stuff to automate. Go get it.

Most companies never make that distinction. They automate once, hear complaints, and either dismiss the complaints or throw money at them, or worse, throw away the solution thinking it didn't work. The actual answer is usually a third round of automation aimed at the friction nobody thought to measure.

Change the training expectation — but make time for it

One more piece of this.

Nobody is going to spoon-feed this the way your team was once spoon-fed a new property management system. The tools change too fast for a training day and a binder. Getting better at this is now part of everyone's job, not a class you attend once and finish.

But that message only lands if you give people real hours to learn, on the clock, as part of their actual workload.

If you tell people to teach themselves and then give them zero time to do it, you will create resentment. And then — this is the trap — you'll mistake that resentment for resistance to AI. It isn't. It's resistance to being set up to fail, which is a completely reasonable thing to resist.

Change what you measure and what you reward

This is the one I'd push hardest, and the one almost everybody skips.

Go back to what our team told us. They felt they were doing more for the same pay.

That's not only a morale problem. It's a measurement problem.

Think about what we asked them to become: overseers, trainers, and guides. Then ask what happens if you still measure them as processors.

If the work has gotten harder and more strategic, but you're still counting tickets closed and calls handled, then your scoreboard is actively punishing the exact change you asked for. The person doing the deep, valuable work looks worse on paper than the person doing volume.

People are not confused about incentives. They read the scoreboard, not the email reminders and messages you send saying how important something is.

11. Where AI helps, and where it's already cheap

Alright. Let's get to the part you probably came here for.

Where does this actually help your business?

I'm going to be blunt about which of these areas deserve your attention and which don't. And I'll warn you now: three of the four are things you can mostly get for free.

When I say something is a commodity, I mean it's widely available, roughly the same everywhere, and nobody should be charging you a premium for it. Like gasoline. You might have a preferred station, but you're not paying double for it.

Data and insights: already commodity

This is what everybody thinks they want AI for. And it's the most commoditized thing on the entire list.

Pulling insights out of your data is genuinely easy now. Ask a question, get an answer, see the trend. What used to require an analyst and a week now takes a few minutes.

It's so easy that we give it away. Our Property Manager Visor product does exactly this, and it's free for everyone — not just Latchel customers. You don't have to buy anything from us to use it.

I'm telling you that so you know what the going rate is. If somebody is charging you a premium for "AI insights," now you know what you're actually paying for.

Marketing and sales: mostly commodity

Same story, mostly.

Content, campaigns, lead capture, follow-up sequences — there are a dozen good tools for all of this, and honestly there isn't much daylight between them. This is table stakes now. It's not an advantage over your competitor, because your competitor has it too.

But notice where the value actually starts.

The moment a lead turns into a signed unit, you're in operations. That's the handoff that matters, and it's where the real work has always been.

Strategy: commodity until you connect your data

Here's where it gets more interesting.

A chat tool — remember, that's the sharp assistant across the desk who can only talk to you — is genuinely good enough to help you build a plan and poke holes in your thinking. You can describe a decision you're facing and have a real back-and-forth about it. That's valuable.

But it's the same value your competitor gets for the same twenty dollars (or free, even!). It's commodity value.

The version that isn't commodity is what happens after rung 2, when you've handed that assistant the logins to your actual systems.

Now they're not giving you general advice about property management. They're looking at your portfolio, pulling from all of your systems at once, instead of giving you an answer stuck inside whichever single tool happens to hold that one piece of the puzzle.

That's the difference between a smart intern and someone who actually knows your business.

And notice — that difference isn't about having better AI. It's about access. Everybody has the same models. Not everybody has connected them to anything.

Operations: this is where the value is

Operations is where the real money is. It's also where the smoke and mirrors are thickest.

So let's walk through it slowly and carefully.

12. Operations, step by step

I want to walk you through one workflow from start to finish — maintenance — because the gap between how vendors describe this and what it actually takes is enormous.

As you read, keep the question from earlier about human vs AI workers in the back of your mind: at each step, who's the processor? Is a person thinking this through, or is the system doing it while a person watches?

Intake, from everywhere

It starts when somebody reports a problem. And they're not all going to report it the same way.

Not everyone is going to fill out your web form. Some people will call. Some will text. Some will use the resident portal. Some will email you a blurry photo with no explanation at all and expect you to figure it out.

That's not them being difficult. That's just how people are.

So every one of those channels needs the same tools and the same rules sitting behind it. A request that comes in by phone has to get handled the same way as one that comes in through the portal.

If they don't, you haven't built one operation. You've built four separate operations that happen to share a company name, and they'll drift further apart every year.

See it in the product: AI maintenance intake takes requests from every channel, and 24/7 emergency support covers what comes in after hours.

The diagnostic tension

Now here's a real trade-off, and there's no clean answer to it.

How many questions do you ask when somebody reports a problem?

Ask more questions, and you get better information for your vendor. They show up knowing what they're walking into, with the right parts on the truck. But submitting the request takes longer, it's more annoying, and your residents get less happy with you every time.

Ask fewer questions, and submitting is easy and pleasant. But your vendor rolls a truck without knowing what they're dealing with. Now they need a second trip, and you're paying for both.

There's no setting on that dial that wins both. Every property manager reading this has felt that tension.

The way out isn't a better form. It's building up information about the property itself over time, so that each new request needs fewer questions than the one before it.

Think about what you'd know after two years of doing that. You already know how old this water heater is. You know the second-floor bathroom has a history. You know which appliances came with the property and which ones your team installed. You know that this particular unit's breaker panel is mislabeled.

Ask once, know forever.

That information compounds quietly. After a couple of years it's genuinely one of the most valuable things you own — and it's completely invisible in any vendor demo, which is exactly why nobody talks about it.

Triage

Before anything gets assigned to anybody, you need answers to a handful of questions.

• Is this an emergency, or can it wait?
• Does this owner have specific preferences we need to follow?
• Is there a warranty on this property that covers the problem?
• And is the description specific enough to act on, or is it too vague to send anyone out?

Every single request needs all four of those questions and more answered. Multiply that by your monthly volume and you can see why this eats your team alive.

This is exactly the kind of work that should move to the system, with your team overseeing rather than deciding one at a time.

The mistake nobody demos for you

Here's a failure you will not see in any sales presentation. I want to spend real time on it, because it's the thing most likely to burn you.

AI figures things out by reading intent. And when it isn't sure, it tends to go broad. It generalizes.

That instinct is useful most of the time. It's what lets you type something sloppy and still get a good answer. But in property management, going broad can be expensive.

Let me give you a real example.

A lease covers certain pests. Mice and other rodents. Wasps and other flying stinging insects. It does not cover roaches, ants, or spiders — those are the resident's responsibility.

That's a normal lease. You've probably written one like it.

We have watched AI platforms read a rule about some pests and conclude that the owner covers all pests.

That is not a small mistake. That's you paying for treatments you never agreed to cover. Or worse, telling an owner they're not going to foot the bill and then having to walk it back or pay it from your own pocket after the work is already done.

Our approach at Latchel is deliberate about this, and I'd encourage you to hold every vendor to the same standard:

Generic instructions get interpreted generically. Specific instructions get interpreted specifically.

In plain terms: if you told us something precise, we hold the line at precise. We don't fill in gaps that you didn't ask us to fill.

That sounds completely obvious when you say it out loud. It is not how these systems behave by default. You have to build it that way on purpose.

So test any vendor on exactly this, using your own lease language. Give them a rule with a specific list in it, then ask about something similar that isn't on the list. Watch what happens.

Who do you send? Nobody gets this right 100% of the time

Now you know what's wrong and who's responsible. Next question: who goes out — a plumber or a handyman?

This looks like a small decision. It isn't.

Send a handyman when you needed a plumber, and you've bought yourself a trip charge and a delay. The resident waits longer, they're unhappy, and you're paying twice for one job.

Send a plumber when a handyman would have handled it, and you just overpaid on labor for no reason. Do that repeatedly and your owners start asking why maintenance costs so much.

There are platforms out there that will promise you they get this call right every time.

They are lying.

And you already know they're lying, because you know your own operation. You and your team don't get this right one hundred percent of the time. And you have context that no software has — you know the resident, you know the property, you've been doing this for fifteen years.

There is not enough information in the world to get that call right every single time. Any system, mine included, is making an educated guess based on what it has seen before. Sometimes it will guess wrong. That's not a defect. That's the nature of the problem.

Now, that is not a reason to avoid automating it.

It's a reason to set expectations that match what's actually possible. Decide what error rate you can genuinely live with. Then design the review step that catches the rest.

Remember the overseer role from earlier and how it's important to sample what the AI is doing to be sure it's handling things the way you want? This is exactly what they look like in practice. You're not hoping it's perfect. You're deciding what "good enough" means, measuring it, and having somebody watch the number.

And if a vendor won't have that conversation with you — if they won't tell you their failure rate — that tells you something important. Either they've never measured it, or they don't want you to see it. Neither one is good.

And then all the rest of it

Getting the right person assigned is maybe a third of the job. Here's the rest of the lifecycle.

All of this can be automated. None of it is glamorous. And every item on this list is probably eating somebody's afternoon right now.

• Matching the resident's availability to the vendor's availability
• Confirming the vendor is actually planning to show up
• Tracking whether they showed up when they said they would
• Surfacing the vendors who no-show or reschedule over and over
• Checking whether the resident was actually satisfied at the end
• Reviewing invoices for errors
• Catching the vendors who routinely overbill you
• Catching the vendors who ask for more budget approvals than they should need
• Following up on every one of those steps so that nothing goes quiet and forgotten

See it in the product: work order tracking, invoicing and payments and reporting and analytics cover the back half of that list.

Look at that list again and count how many of those are things you know you should be doing but currently don't, because there aren't enough hours.

Now here's the hard truth.

You will not get one hundred percent coverage on any of it. No system will ever give you that.

And that's part of the point of leveraging AI in the first place. Let me explain what I mean, because it's the most important idea in this whole post.

13. The job gets harder. That's the point.

Here's what the vendors selling you easy street won't tell you.

I say this having lived through several rounds of automation, including a few that happened long before today's AI existed. This pattern is older than the technology.

As you automate the routine, repetitive work, the job gets harder. Not easier.

That's not a flaw in the transition. That's the entire return on it.

Let me show you what I mean, using the two examples we've already walked through in this post.

On the sales side, your team stops spending its day chasing and calling leads. That's good — that work was repetitive and most of it went nowhere.

But now they spend that time figuring out why your units aren't leasing faster.

That's harder work. It takes judgment. It takes knowing your market. It takes being willing to form a theory and then be wrong about it in front of other people. There's no script for it, and nobody can tell you if you did it right today.

On the maintenance side, your team stops spending its day chasing vendors and residents for status updates. Also good — you saw that list in the last section.

But now they spend that time actually managing those vendor relationships.

Why has this vendor's performance slipped over the last four months? Is it fixable, or is it time to replace them? Who's going to have that conversation, and what are they going to say?

That's harder too. And we have quite a bit of time before AI can take over that job.

And notice what your team has become in both examples. Overseer. Trainer. Guide. They're not executing on the tactical and the routine. Their new role is harder, more strategic, and more valuable.

Here's the part I really want you to sit with.

That new job always existed. You just weren't doing it. You couldn't.

You couldn't spot the vendor who was quietly creating more work than they resolved, because you were completely occupied resolving the work. You couldn't fix your leasing problem, because you were on the phone with leads all day. The analysis never happened because the person who would have done it was busy doing the thing the analysis was about.

The automation doesn't create that job. It reveals it. That work was sitting there the whole time, costing you money every month, and you never had a free hour to go look at it.

Now, remember what we learned the hard way when we rolled out new AI internally at Latchel. When the job gets harder, you have to check which kind of harder it is.

If it's harder because it now takes more judgment, more expertise, and more real thinking — that's the good kind. That's the return you were after. Support your people through it, change what you measure, and let them get good at it.

If it's harder because it's annoying, scattered, and full of friction — that's the other kind. That's digital paper pushing that survived your first round of automation. Go automate that too.

Being able to tell those two apart is most of the skill in doing this well.

So here's my honest pitch, and it's not the one you'll hear from most vendors.

Not fewer hours. Better hours — spent on the problems that actually move your business, which are by definition the ones that don't come with a script.

If that sounds like less fun than easy street, I understand completely.

But easy street was never on the table from anybody. And the people promising it to you right now are going to be very hard to reach in eighteen months, when you call to ask why it didn't work.

14. Where to start

If you've read this far, you don't need any more convincing. What you need is a sequence.

We've put the full step-by-step into an ebook: the rung-by-rung checklist, the exact questions to ask a vendor before you sign anything, what order to do all of this in, and the mistakes we made ourselves so you don't have to repeat them.

But if you only do one thing this week, make it this one.

Go sign up for a good AI chat tool. The free version is fine to start. Use it every day for thirty days on real work: your actual owner emails, your actual vendor problems, your actual budget questions.

That's rung one. It costs nothing but attention.

And everything else in this post is built on top of it.