How to Budget for an AI Platform
On this page
Ask most operators what AI does for them today and you will get a version of the same three answers: a bot that handles questions after hours and takes maintenance requests, follow-ups that chase leads, reminders on collections. All three are useful, and all three can be built on Travtus in a sentence, without waiting for a vendor to add them.
But notice where that work sits. Every one of those jobs happens at the perimeter of the business, with a resident or a prospect on the other side. The bot takes the maintenance request, and then everything after it runs exactly as it did before. Somebody still prioritizes the ticket. Somebody still finds a technician who is not already on another job.
The perimeter gets faster. Nothing behind it changes.
Two things would change it, and neither appears on that list. The first is your own people getting answers without waiting on the data team, which alters who is able to make a decision, not merely how quickly it gets made. The second is work that begins inside the business rather than with a resident: between a regional and maintenance, between asset management and operations.
No vendor sells you that, because your operations are not their product. That is the case for a platform, and it is also why a platform is difficult to put in a budget. What follows is how to build the line, what to take out to fund it, and who will challenge it.
How do you build the budget for an AI platform?
Start with the easy part. Everybody charges per door, so the rate is not where this is won or lost. What matters is that it is one cost for the platform rather than a cost per function. When somebody asks for another workflow or another report next year, that is not a new product, a new contract, a new integration or a new security review. It is the same line.
So begin by counting how many times you are already paying for the same doors: one vendor for leasing, another for collections, another for reporting, another for procedures. That figure is the one your CFO has never seen laid out in a single place, and it is the real shape of AI vendor sprawl.
The hard part is what sits next to it. A brand-new line with nothing above it looks like brand-new money, and brand-new money attracts exactly one question: what comes out? Have that answer ready before anyone asks. For most operators the list is roughly the same:
| Offset | Where it sits today | Why it gets missed |
|---|---|---|
| Data warehouse upkeep | Data or IT budget | Approved as a project, so the ongoing cost of keeping it fed and mapped never gets counted |
| Reporting and analysis | Bought on the side, per report or per seat | Spread across enough small invoices that it is never consolidated |
| Single-job tools | Operations and marketing budgets | Pushing procedures to site teams, alerting somebody when a number moves, reading messages for sentiment. Each one too small to trigger a review |
| The workarounds | Payroll | The spreadsheet holding the real version of the number, the pack rebuilt by hand every month, the requests stacked on your analyst's desk |
That last row is the biggest item on the list and it is not a vendor at all. You are already paying for it, out of payroll. It simply never had a line of its own.
Then phase it. Nobody funds the whole thing in year one.
| Phase | What you fund | What it proves |
|---|---|---|
| Year one | Access to information: your own people getting answers without waiting on the data team | That the context layer is real, and that people use it unprompted |
| Year two | Automation on the same line: workflows, scores, scheduled reporting | That the same line absorbs the next thing your teams build |
That halves the first ask, and it gives you a plan instead of a number.
What does an AI platform replace, and what does it not?
This is where most business cases fall over, so be blunt about it. Start with what genuinely comes off the budget.
| Line item today | What it really buys | Priced | Status |
|---|---|---|---|
| Mystery shopping | A few sampled calls, scored quarterly. Every conversation your team has is already scored, every day. | Per shop | Replaced |
| Resident survey program | A slice of residents asked once a quarter how they feel, when most have been telling you all year in ordinary messages. | Per door | Replaced |
| Review responses | Somebody in marketing writing replies on Friday afternoon. | Per property | Replaced |
| Review generation | Requests blasted to every resident regardless of sentiment. | Per door | Replaced |
| BI seats | Licenses for people who only ever opened a dashboard to find one number. | Per seat | Reduced |
| The monthly reporting pack | An analyst rebuilding the same pack by hand. | Payroll | Replaced |
| Procedure and SOP tools | A library somebody refreshes once a year. | Per user | Replaced |
| Resident communication | Inbound and outbound messaging. | Per door | Replaced |
Here is the part worth putting in front of your CFO: every one of those is a separate vendor today, and most of them charge you per door. The same doors, over and over.
Now the other half of the list, which matters just as much.
| System | Why it is untouched | Status |
|---|---|---|
| Property management system | The platform sits on top of your system of record. Nobody moves your ledger, your leases or your payments. | No change |
| Leasing CRM | Pipeline and lead management stay in the system your leasing teams already work in. | No change |
| Rent pricing | Pricing is not in scope, and claiming otherwise is how a business case loses credibility. | No change |
| Compliance training and certification | Certification stays with your existing provider of record. | No change |
Say this second half early, and say it to everybody. It costs you nothing and it does two jobs at once. It removes the suspicion that this is a system migration wearing a different hat, which is the first thing your IT lead will assume. And it stops you promising savings you cannot deliver, which is what puts a sponsor in difficulty next summer when finance comes looking.
Only promise what you will actually cancel. A short honest list beats a long hopeful one.
How do you make the case inside your own company?
The number is the easy part. What you are really being asked is whether anybody will build anything with it. Six moves answer that before you walk in.
- Start with a decision somebody actually makes. Not a feature. A decision. Sit down with whoever owns the outcome and ask them to name one call they make regularly where better information would change what they do. That single sentence turns "a platform" into something with a dollar figure attached, and it makes the value theirs rather than yours.
- Look at your own data before you pick the first project. This feels backwards, and it is the strongest move available to you. A product demo becomes a feature comparison. Sitting down with what you actually hold, and what shape it is in, becomes a conversation about what you could do with it, and the first project then gets chosen against something real.
- Get other people to own it with you before you ask for money. You want three people in this: you, somebody who owns the business outcome, and your sponsor. This is not politics. It is the defense against the one objection that ends champions, which is coming up next. If operations and asset management picked the first project, it is a business decision. If you picked it alone, it is your bet.
- Pick the worst workaround, not the biggest opportunity. The biggest opportunity involves the most people and the longest road to proof. The worst workaround comes with a team that will use the thing on day one, and a before-and-after anybody can see. Maintenance coordination is usually sitting right there.
- Brief your sponsor in their own words. Find the phrase they already use about this part of the business and hand it back to them. You are not translating your idea into their language; you are showing them that something they already believe now has a way to happen.
- Ask for a commitment, not just a check. Most people skip this, and it is the one that holds the case together. Do not ask for approval to spend. Ask for approval to spend, plus a commitment from named teams to build a certain number of things in the first year.
That last point matters more than it looks. A platform is poor value if you build two things and stop. Saying so yourself, before anyone else does, is what makes the rest of your case believable.
What will you be asked?
Seven questions, and knowing who asks each one is half the work.
| The question | Who asks it | The answer that holds |
|---|---|---|
| "What comes out?" | Your CFO, and it will be first | You have the list. Keep it short and keep it real. Only what you will genuinely cancel. |
| "We already tried this." | IT or data | Rarely an argument about the product. You win it by not being the only person in the room who wants this. |
| "Why can't we just use Claude or ChatGPT?" | Your CIO, increasingly your CEO | Do not argue the model; Travtus runs on those models. A chat window has no context, no initiative and no governance. |
| "Why not just buy a leasing AI?" | Nearly everybody | A packaged set of workflows is somebody else's assumptions. And the second thing you automate is another purchase. |
| "Our PMS already has AI agents. Isn't it included?" | IT, or whoever owns the PMS relationship | Four ownership questions settle it: the logic, the data path, the model, the roadmap. |
| "My teams won't use it." | Your COO, and they are right to ask | Worst workaround first. One team, one region. Small enough that it cannot fail quietly. |
| "Another vendor, another security review." | Procurement, legal or IT, usually late | Fewer vendors is less to govern. Bring counsel in early and briefly. |
Three of those deserve more than a row.
"Why can't we just use Claude or ChatGPT?" This is often asked by people who have been using those tools themselves all year and find them genuinely good. Do not argue with them on that point, because they are right, and Travtus runs on those models. This was never a choice between them. The question is what sits around the model, and there are three things a chat window does not have.
- It does not know your business, and keeping it knowing your business is the actual work. A chat assistant answers from whatever you paste into it. Knowing what is happening at one community this week, across the work orders, the conversations, the lease and the ledger, kept current, is a data problem rather than a prompting problem.
- Nothing starts on its own. A chat window waits for somebody to open it. It will not notice that a number moved on Tuesday and go tell the regional who owns that region.
- It is per person, not per company. Everybody builds their own version in their own window. Nothing is versioned, tested, approved, published to a named audience, or scoped to a portfolio, and nobody can see what anybody else built. When they leave, it leaves with them.
If your own team would rather wire a model up to your data themselves, that is a real option and worth pricing. Just go in clear-eyed that you are taking on the pipelines, the permissions and the housing know-how, and that you will own all three permanently.
"Why not just buy a leasing AI?" It is a fair question, and there are three honest answers. First, a packaged set of workflows is a fixed set of somebody else's assumptions about your business: you can automate what they built and nothing else, and the work that is particular to how you run is out of scope for good. Second, none of it is internal, because your own operations are not that vendor's product. Third, and this is the one to sit on: the second thing you want to automate is another purchase. On a platform it is already paid for. That is the whole platform versus point solution question in a single line.
It is worth noting quietly that small AI companies in this market have a habit of being bought by bigger ones, and the product you signed for is not always the product you end up with.
"Our PMS already has AI agents. Isn't it included?" Bundled looks free. Four questions sort it out.
- Who owns the logic you build?
- Who owns the path your resident information travels, and how many companies does it pass through?
- Who picks the model?
- Who can change how it behaves: you, today, or somebody else's development list?
Bundled agents also only reach as far as that vendor's own products, and they do not come with you if you ever leave. Those four belong on the shortlist in any enterprise AI platform evaluation, and the first is really a question about whose data model you are building on.
How Travtus approaches this
Travtus is the everyday AI platform for multifamily. Somebody on your team describes the workflow, report or score they want, and they build it themselves, in a sentence. It stays yours. It sits on top of the systems you already run instead of replacing them, and it is built for the governance an enterprise actually needs.
A tool that does one job automates that job. A platform changes how the company runs. The budget question is really a question about which of those you are buying, and that is most of what it means to be an AI-native operator.
Before your 2027 budget locks
The instinct in a budget cycle is to treat AI as a line item, one more tool sitting beside the others, justified on its own. That framing is what makes it so hard to fund, and it is also the wrong shape for what this is.
A platform is closer to a shared service. You stand it up once, the way you stand up a data warehouse or a network, and every team then draws on the same foundation. The first team to build something carries the setup. The second draws on what is already there. By the tenth, the next use case has no marginal cost at all.
Which changes what you are actually approving. Not a tool for a department, but a capability the whole organization draws on, where the constraint stops being budget and becomes how quickly your teams can think of things worth building.
