Executive summary
People remember how a product makes them feel, and with the recent advancement of voice technology and its integration with AI agents, the product experience can be redefined. The experience shifts from something customers have to learn, into something that understands them and that is what voice agents can make possible, by removing friction, speaking your language, navigating on your behalf, and delivering what you ask for. This post explores the evolution of the voice interface, how it can be integrated into a product, and the steps and challenges involved.
Before voice, we interacted with technology through keyboards, screens, and touch. Each shift made technology feel more natural and accessible and voice was the next step. Voice assistants have been around for more than a decade, beginning with Siri, which brought voice interaction into the smartphone for the first time, followed by Alexa and Google Assistant, which expanded beyond the phone into smart speakers and connected devices. But these systems were purely reactive. A user speaks, the system responds to a specific question, and the session ends. No continuity, no memory, no ability to reason across a conversation. They could tell you the weather, set a timer, or play a song, but they could not hold context, lead a dialogue, or understand the situation in which you were asking. This was never a limitation of voice as a modality in real-life, it was a limitation of technology and systems built to react, not to reason.
That changed with the rise of large language models and agentic frameworks. Voice systems stopped being glorified search boxes and started being something genuinely different. They could hold context across a conversation, infer what a user meant rather than simply parsing what was said, and reason through a sequence of steps, making decisions along the way. They have access to the knowledge they were trained on, connect to external tools, and act on your behalf inside real workflows, not just returning an answer. Recent product integrations already demonstrate this in practice, a voice agent can call a client, schedule a follow-up, block time in a calendar, answer frequently asked questions, process a customer complaint, recommend a product based on past behavior, or guide someone through a complex application. It is no longer a system that simply answers you, it is one that genuinely understands you. But how can we bring this into the digital experiences we build as well?
Look at almost any digital product today and you will find the same underlying assumption, users navigate by tapping, clicking, and typing. Designers built every menu, filter panel, search bar, and checkout flow around that model. As a result, users are left to translate what they actually want into the language of a UI, and that translation takes effort. That is what we have normalized, simply because we have never seen the alternative at scale. Chatbots attempt to resolve this, but typing is never as natural as speaking. But what if a voice agent handled that instead? Rather than expecting users to navigate a product on its own terms, a voice agent understands what they want, navigates on their behalf, and guides them through the experience. That is a fundamentally different starting point and it changes what a product can feel like to use.
Let's put it into action so you can feel the value behind it.
Imagine you are on your favorite shopping website. Instead of browsing category by category, sorting by price and color, and opening tab after tab, you simply say: "Based on the jeans I bought before, what would be a good black top to go with them?"
The agent knows your purchase history, understands your style, and recommends options that fit. It can even show you how a piece would look on someone with your profile, all without typing a single word or opening a chat window.
Without agents, this experience looks very different: a series of isolated questions and answers. "Show me black tops under €35." One query, one response, no continuity.
With agents, it becomes a real conversation that carries context, builds on what was said before, and moves toward what you actually need. What makes this possible is not voice alone. It is the agentic framework behind it, the ability to hold context, reason across information, and act on your behalf. Voice is what makes it feel truly human.
This shift is already reshaping retail discovery more broadly. As our team has explored, AI shopping agents are reshaping how products are discovered and selected across the entire commerce journey, with voice as one of the most natural interface layers. [1]
This is not limited to shopping. Take the same idea and apply it to finding a home.
You open a real estate website. Instead of adjusting price sliders, filtering by bedroom count, and scrolling through hundreds of listings that may or may not match what you had in mind, you simply talk. You describe what matters to you, the neighborhood, the size, the feel, and the agent searches on your behalf.
It surfaces the listings that fit, answers your questions about each one, and goes beyond what photographs can show. It draws on its knowledge to help you understand a property more fully, visualize how a space might feel to live in, and make a more informed decision.
You speak in the language that feels most natural to you, and the conversation moves at your pace. That is what becomes possible when voice is treated as the interface of a product.
In the following demo, you can see the voice experience supports:
Watch the voice agent real estate demo:
The shift to voice agents is not just a design decision, it is a business one. According to Market.us, the global voice AI agents market is projected to grow from $2.4 billion in 2024 to $47.5 billion by 2034, a compound annual growth rate of 34.8% [2].
When a customer can simply talk and get what they need, the path to conversion becomes effortless. When a product speaks your language and understands your tone, customers stay longer, return more often, and trust the brand more deeply.
The operational case is just as strong. Gartner forecasts that conversational AI will cut contact center labor costs by $80 billion in 2026 alone [3], reflecting how voice agents now handle frequently asked questions, complaints, and routine workflows without human intervention. An agent that guides decisions and resolves queries means the product is always available, always helpful.
The financial return follows. Forrester research shows that companies deploying voice AI report a three-year ROI between 331% and 391% [4], a return that reflects not just cost savings but the compounding value of customers who return, trust the brand, and recommend it to others.
But perhaps the most significant value is differentiation. In a landscape where most digital products offer a similar experience, a product that responds to your voice, remembers who you are, adapts to how you speak, and acts on your behalf is one that customers will remember and return to.
A voice agent is only as good as the framework beneath it, the data behind it, and the conversation design around it. Three foundations matter most:
The voice is the surface. What sits beneath determines what the agent can actually do. Choose a framework that supports multi-turn conversation, tool use, and memory. Without these, the voice experience is just a search bar with a microphone.
An agent that cannot reach your product's data, user preferences, catalogs, live inventory, or external systems can only return generic responses. The more context it has, the more useful, personal, and reliable it becomes.
Teams built most digital products around clicks and taps: predictable, structured inputs. The voice is neither. It requires designing for the way people actually speak, fluid, sometimes ambiguous, and always contextual. That means anticipating how users will phrase what they need, how the agent should respond when something is unclear, and how to transition smoothly to a human or another resource when the agent reaches the limit of what it can do.
None of this comes without real engineering and design considerations. A few deserve particular attention:
Users expect voice to feel as natural as conversation, which means response times that feel slow in a chat interface feel even slower when spoken aloud. A pause that lasts seconds breaks the natural dialogue.
Voice agents need to stay within the context of the product they are embedded in. An agent that oversteps, guesses where it should not, or takes on requests beyond its role will quickly lose the trust of the user. Knowing what the agent should not do is just as important as knowing what it can.
A voice agent is only as trustworthy as the information it acts on. If it mishears a request or misinterprets intent, the consequences in voice feel more jarring than in text. Getting accuracy right, both in what the agent hears and how it interprets meaning, is fundamental to building an experience users will rely on.
A product that feels like it understands what customers need, navigates with them rather than making them navigate alone, saves user time, is useful, is something they come back to, recommend to others, and remember. Voice agents can create exactly that kind of experience. The technology is ready. The architecture exists. What remains is the willingness to ask a harder question than "should we add a voice feature?" and instead ask, "what would this product become if voice were at the center of it?"
ML6. How Do Retailers Stay Visible with AI Shopping Agents Choosing Products? https://www.ml6.eu/en/blog/how-do-retailers-stay-visible-with-ai-shopping-agents-choosing-products
Market.us. Voice AI Agents Market to hit USD 47.5 Billion By 2034. https://market.us/press-release/voice-ai-agents-market/
Gartner. Gartner Predicts Conversational AI Will Reduce Contact Center Agent Labor Costs by $80 Billion in 2026. https://www.gartner.com/en/newsroom/press-releases/2022-08-31-gartner-predicts-conversational-ai-will-reduce-contac
Forrester Consulting. The Total Economic Impact™ of PolyAI. https://tei.forrester.com/go/polyAI/PolyAITEI