In the annals of technological progress, few trajectories have been as frustrating as that of voice AI. For years, we've endured the ineptitude of Siri and the infuriation of automated phone systems, wondering if the promise of conversational AI would ever be realised.
But the tides are turning.
If you've seen the GPT-4o demo or spent 30 seconds cloning your voice on Play.ai, you know: voice AI has levelled up. The question is no longer whether it works, but how it will reshape our world.
Two distinct camps have emerged in this new landscape:
The Horizontalists: The platform players
The Verticalists: The industry specialists
In one corner, we have the Horizontalists: startups racing to build the definitive voice AI platform. But this space is already hyper-competitive, with well-funded startups jockeying for position. More importantly, does anyone want to bet against OpenAI or AWS swooping in and dominating this market? The economies of scale and existing customer base give tech giants a formidable advantage.
In the other, we find a more intriguing cohort: the Verticalists. These are the companies crafting AI workers with deep domain expertise. They're not just adding a voice layer to existing software; they're engineering digital workers that truly understand context and nuance.
Imagine:
Car dealerships where AI agents handle 80% of initial inquiries, schedule test drives, and even begin price negotiations.
Logistics companies employing voice agents to manage carrier communications, provide real-time shipment tracking, and dynamically optimise routes, reducing manual call time by 60%.
This isn't speculative. Companies like Toma and Happy Robot are already bringing these visions to life.
It isn’t really about Voice at all, it’s about new software that’s multi-modal and uses AI to complete workflows only a person could have completed before.
Why does this approach matter? Three reasons:
Bridging the knowledge gap. Most businesses lack the technical expertise to implement and customise generic AI solutions. Vertical products offer immediate value without extensive onboarding or the need for in-house AI talent.
Employee empowerment. Just as accounting software allowed bookkeepers to evolve into strategic financial advisors, AI workers will free up employees to focus on high-value, relationship-driven work. The result? Increased job satisfaction and better outcomes for customers.
Budgetary sleight-of-hand. Instead of requiring a new "AI budget," these solutions slot into existing line items. "We can do the work of an employee, but cheaper and 24/7" is a much easier sell than "invest in the future of AI."
The Road Ahead
Building these systems isn’t easy. It requires deep industry knowledge, robust integrations with existing systems, and a willingness to start with an 80% solution, relying on early adopters to refine the offering.
Finding a compelling insertion point matters and we expect industries where it’s hard to find talent will be quickest to adopt technology that could enhance their existing workforce to improve productivity.
It will also be easiest to get traction in industries where the software tools they’re currently using either don’t capture the information required to build differentiated models or don’t have the talent to ship new features quickly — Gong is much more likely to ship AI workers than the Dealer Management Software at the Car Dealership.
But for those who can thread this needle, the opportunity is immense. We're looking at potential category-defining companies emerging in multiple industries.
If you’re building industry-specific software incorporating voice AI, I want to hear from you. If you disagree with any part of this analysis, let’s debate! This is the first sketch of an idea — it’s designed to be improved with feedback.
I think you're spot on with the opportunity and view on vertical vs horizontal - however I think like most AI SaaS tools, the distribution (and hence GTM) would be the moat among the plays possible here. It can't be the usual SaaS plays but more like a staffing agency approach.
Really good article. The jury is out on this one. The discussion goes back to whether a swarm of specialized agentic AIs working as a cohesive network would be better at solving problems at a lower computational cost than a superintelligent AI. Companies like Aitomatic are pioneering a new type of Agentic AI called Small Specialised Agents. They trained these for highly specialized applications in semiconductor manufacturing and others. See their AI Virtual Advisor here. https://www.aitomatic.com/products/aiva
I believe specialized Agentic AI is the future. A long way to go to get to seamless handover and coordination between various SSAs. We are building SSAs to maximize the lifecycle value of industrial equipment with circular economy. Its a sweet spot for Agentic AI with sparse and disjointed data, as well as simple but domain knowledge driven manual decision making.