As vibe coding and off-the-shelf agent tooling has become more accessible, we're hearing a version of this question more often: “why pay for a specialized voice AI product when I could just build my own on ElevenLabs."
I've come to relish the question.
At Super, we have to think about the role of voice AI and where specialization comes into play... constantly. Every day, our job is to understand and rapidly evolve our own understanding of what best-in-class looks like for voice AI while also balancing the distinct requirements of a vertical-specific voice agent. It's a uniquely tricky and specialized segment of AI, and one of the reasons why it's not as easy as just bolting on ElevenLabs or Retell. The space is shape-shifting rapidly—from “it can speak?" to “why don't I roll my own"—in just the span of a year. Pretty wild.
So here's how I would think through a voice AI strategy.
The first question to ask is whether this is a competency that you should build, buy, or partner on. If voice AI is core to your business model — not just a tool you use, but core to your service offering — then building your own makes sense. Own the stack, own the differentiation. That's where Super sits.
Similarly, if your use case is genuinely hyper-specific in ways that no off-the-shelf product will ever accommodate, and it's a big enough problem for you to invest in staff and time to fine-tune and maintain, building your own will give you that fine-grain control for extremely specialized use cases that vendors may not cover.
But for most people we talk to — including property management systems and proptech companies — the honest truth is that voice AI is a component to their business, not the end-all-be-all.
I'll be candid that building a functioning voice agent that can handle more than a 2-step flow is extremely difficult. We have entire teams laboring over how to make a system, and it is still constantly requiring fine tuning. The state of the voice AI industry is not mature enough to pick one single partner and set it and forget it.
So if you're willing to make the investment, the honest bar should be whether this your bread and butter? Is this a core product you want to build out, iterate, maintain, train (and re-train)? If yes, build. If not, read on.
If your needs are simple — taking a message, very basic call routing, no integration with your property management system — a generic agent builder or off-the-shelf voice tool is probably fine. The configuration work is manageable, the cost is low, and you don't need vertical-specific intelligence to handle “what's your name?"
Small shops with minimal call volume and no integration requirements are a reasonable fit here. If cost is the primary driver and the use case is genuinely shallow, a generic solution will likely do what you need. I'll be the first to recommend exploring this pathway to small operators with lean use cases.
The limitation shows up when the calls get more complex, as quickly happens in property management. The unique thing about real estate as a vertical, and particularly property management, is not only are you dealing with regulated subject matter, you are also dealing with a wide variety of inquiries. This isn't a local spa or restaurant where phone inquiries are pretty much always the same.
This is where it gets hairy. Property management calls aren't generic. If you need to pull live data from other systems, such as your currently available listings and prices, need to know how to evaluate urgency (is this regular maintenance or an emergency), handle Fair Housing guardrails, match addresses across your portfolio, push call logs and actions into other systems... you're going to be left building a lot of this on your own. A generic platform will not support any of this out of the box, you're having to build all of the vertical expertise and integrations.
The vertical-specific layer — properly identifying caller scenarios, the PMS integrations, the Fair Housing guardrails, multi-step and conditional call flow logic — is where I approximate 70% of the effort lies. A specialized platform has already done that work, across a lot of customers, which means the edge cases are already handled.
But, the other 30% is the infrastructure layer. It's easy to underestimate this with a bundled pipeline offering, such as what ElevenLabs or OpenAI Realtime offers. But with these fully managed offerings, there are trade-offs based on the complexity and customization you're building in your voice AI agents.
For example, a “speech-to-speech" model such as OpenAI's realtime is currently limited in its ability to do things like “tool calling" and observability. Both of these limitations are fairly large trade offs. Tool calling is the ability to pull up information externally in the moment (such as your available listings). In a property management voice application, this is critical because you are always balancing speed and context: how do I not overwhelm the agent with context and also slow down its responses with how do I ensure it can gather the necessary information at the right moment. The lack of ability to see exactly what was decided and why throughout parts of the conversation also make it harder to pinpoint issues and continually improve upon it.
In a fully managed pipeline, like ElevenLabs, you're trading off flexibility. Right now, there's a very real race with providers for each component of the voice AI pipeline, and being able to select and swap in your speech-to-text (transcription and turn detection), LLM, and text-to-speech (the audio generation of the voice itself) providers is key to rapid improvements.
That brings us to why we have such a strong thesis at Super for what we're building. A vertical-specific, voice AI specialized platform that can bridge all of these components.
We think deeply about what makes a great voice AI experience and are not only actively iterating, enhancing, and improving our own system to this end, we work closely with other infrastructure providers to ensure that our ecosystem is improving for our customers. One recent example of that: we work with several speech-to-text (STT) providers and our product and engineering teams have direct lines of communication with these platforms in order to give targeted feedback and ideas for ways to improve our customer's experience. This has led to platforms releasing updates and early beta testing of features that we uniquely have access to because of our specialization.
We're also embedded and growing our integration ecosystem, including the capability for our customers to bring their own integrations and build on top. In this world, they get the best of everything: a specialized system that meets their needs and that they can expand upon if they wish to.
The cost question usually resolves itself once you add up the true cost of building anything independently: API costs, integration maintenance, ongoing tuning, quality assurance testing, security audits, and leadership focus and engineering time that could be spent elsewhere. For most property management companies, the math favors partnering with specialists— not because building is impossible, but because maintaining and improving it is the real commitment.
The question isn't whether you can build it. You probably can, and I'm always encouraging people to try something first. But once you move beyond that, the question is what you're optimizing for: cost, specialization, unique IP, or an operational capability improvement? Get crisp on that, and then evaluate. We're always here to help with that evaluation.
Ready to get the voice AI platform built for real estate operations? Book a demo with our team to deep dive into how Super is solving the responsiveness gap for property managers.
