Introduction

For most of the past decade, if a large company wanted to deploy voice AI for customer service, the realistic choices were limited to what Microsoft, Amazon, and Google had built into their cloud ecosystems. Azure Cognitive Services, Amazon Connect, and Google CCAI — Contact Center AI — were the enterprise defaults, bundled into the cloud infrastructure agreements that most large organisations had already signed. A bank wanting to automate phone banking, a hospital system wanting to handle appointment scheduling at scale, or an airline wanting to manage customer inquiries without expanding its call centre headcount would typically evaluate those hyperscaler tools first, because the procurement pathway was already open and the enterprise relationships were already established.

That distribution advantage is now being directly challenged, and the challenge is coming from a company most people know primarily from automotive dashboards rather than corporate contact centres. SoundHound AI, through a sequence of moves that accelerated dramatically in 2026, has assembled a position in enterprise voice AI that gives it access to a customer base that rivals the scale of what Microsoft, Amazon, and Google have been serving — and does so with a model-agnostic, voice-specialist platform that the hyperscalers, constrained by their own ecosystem interests, cannot easily replicate.

What SoundHound Actually Built

Before examining what the new position means competitively, it is worth being precise about what SoundHound has actually built, because the company's technology history is more specific and more relevant to the enterprise advantage it claims than most coverage suggests.

SoundHound's core technology is voice recognition and natural language understanding built from scratch rather than licensed from a foundation model vendor. The company's proprietary speech-to-meaning architecture — which processes voice input directly into semantic understanding rather than first converting speech to text and then parsing the text — produces latency advantages in conversational applications that matter enormously in the specific use cases that enterprise voice AI is deployed for. A customer calling a hospital's patient access line does not want to wait three seconds between speaking and receiving a response. A caller managing a financial services query over the phone needs an interaction that feels as natural as speaking to a human agent. The sub-second response times that SoundHound's architecture can achieve in production environments are not a benchmark difference — they are the difference between a voice AI interaction that customers accept and one they immediately abandon by pressing zero to reach a human.

The Amelia 7 platform, launched as the company's enterprise AI agent product in 2025, adds agentic capability on top of that voice foundation: the ability to retrieve information from backend systems, complete transactions, process requests, and escalate to human agents when the interaction requires judgment that the AI cannot yet handle autonomously. The Autonomics platform extends that capability into IT service desks and internal operations. Together, the two platforms address the specific operational problems that have made contact centres one of the most expensive and most resistant-to-change functions in large enterprises for three decades.

The LivePerson Acquisition and What It Changed

The most consequential single event in SoundHound's 2026 positioning was not a technology announcement. It was a corporate acquisition. SoundHound announced in April 2026 that it was acquiring LivePerson, a digital customer engagement company that has been building enterprise messaging and customer service relationships for more than a decade. The terms and the strategic logic are both captured in SoundHound's SEC filing: the combination pairs SoundHound's AI platform with LivePerson's digital engagement capabilities, which power one billion customer messages per month.

What LivePerson brings in concrete terms is a customer roster that no enterprise AI startup could build from scratch in under a decade. Combined with SoundHound's existing customer base, the merged company works with enterprise customers across more than 30 countries, including 12 of the top 15 global banks, 4 of the top 5 global airlines, 4 of the top 5 global automakers, and more than 10 of the leading global telecommunications providers. These are not leads or prospects. They are companies that have been paying LivePerson for customer engagement services for years, many for over a decade, which means they already have the data integrations, the compliance frameworks, the procurement approvals, and the organisational familiarity with vendor relationships that enterprise sales cycles require.

SoundHound CEO Keyvan Mohajer described the combination as bringing together "two complementary conversational AI pioneers," and the logic of the description holds on examination. SoundHound provides the voice AI capability and the agentic intelligence. LivePerson provides the enterprise distribution and the digital engagement layer that handles the non-voice channels — messaging, chat, and web interactions — that a complete customer experience platform must cover. The result is a company that can credibly address the full spectrum of customer interaction channels for organisations that are currently running those interactions through a patchwork of legacy telephony platforms, CRM integrations, and point solutions from multiple vendors.

The Experis Partnership and Why Healthcare Is the Opening Move

Alongside the LivePerson acquisition, SoundHound formalised a strategic partnership with Experis, a global technology services company within the ManpowerGroup family, naming SoundHound as its exclusive conversational AI technology partner for the EXCELERATE AI services portfolio launched in March 2026. The collaboration is initially focused on healthcare, financial services, and retail, with healthcare positioned as the first deployment priority.

The healthcare choice is not arbitrary. Healthcare contact centres represent one of the most painful and most expensive voice AI deployment targets in the enterprise landscape, for reasons that go beyond volume. A hospital system's patient access function — the team that handles appointment scheduling, referral management, insurance verification, and general patient inquiries — typically operates at perpetual capacity, with long hold times, high agent turnover, and significant patient dissatisfaction. The interactions are also more complex than most contact centre scenarios: patients are often anxious, their inquiries frequently involve sensitive personal health information, and the backend systems that need to be accessed to answer questions and book appointments are deeply fragmented across Electronic Health Records systems, scheduling platforms, and insurance verification tools.

SoundHound's Amelia 7 platform is specifically designed for exactly that complexity. It can handle the conversational nuance of a patient who cannot clearly articulate their symptoms, navigate the system integrations required to check appointment availability across multiple clinic locations, and escalate appropriately when the interaction requires clinical judgment. Experis brings the implementation expertise and the healthcare IT consulting relationships that allow those deployments to happen inside organisations that have strict change management processes and significant regulatory constraints. The combination is what Experis described as enabling organisations to "modernize service delivery across patient access, contact centers, and IT service desks."

Why SoundHound's Advantage Over the Hyperscalers Is Real

The instinctive response from anyone familiar with the enterprise software market is to ask why a company with SoundHound's size — revenue of roughly $85 million in 2025 — can maintain a durable competitive position against Microsoft, Amazon, and Google, each of which has access to vastly greater computing resources, far more training data, and enterprise relationships that dwarf anything SoundHound currently serves.

The answer begins with the specific nature of voice AI in regulated industries, and it is encapsulated by the distinction that SoundHound CEO Keyvan Mohajer drew in his May 2026 interview when asked directly about the hyperscaler competition. "We're not a plug-and-play vendor," he said. "We act as a strategic partner, working closely with enterprises to integrate AI into their real-world operations, continuously optimize performance, and deliver measurable outcomes over time." That statement describes a go-to-market model that is structurally different from the hyperscalers' approach, and the structural difference matters more than the resource difference in this specific market.

Microsoft's voice AI offerings are optimised for enterprises that are already deep in the Microsoft stack. Azure Cognitive Services for voice delivers acceptable accuracy across a broad range of applications, but it is not specifically tuned for the acoustic environments, the interaction patterns, or the backend integrations of a healthcare contact centre or a financial services call floor. Deploying it effectively requires significant customisation work that Microsoft's standard implementation partners are not always equipped to provide at the level of specificity that regulated industries require. Amazon Connect has similar characteristics: strong infrastructure, adequate accuracy, complex customisation requirements for anything beyond basic use cases.

SoundHound's OASYS platform, announced in May 2026, adds a further dimension to this advantage. OASYS is a model-agnostic framework that allows enterprises to switch between AI models — from any provider, including models from Anthropic, OpenAI, Google, or Meta — based on performance and use case, without rebuilding the voice AI infrastructure around which model is currently performing best. SoundHound CEO Mohajer made the strategic argument for this directly: "No company can predict which model will win over the next two years, and enterprises shouldn't be forced to bet on it today." A financial services company deploying voice AI for customer service cannot afford to rebuild its customer-facing systems every time a better foundation model becomes available. OASYS means they do not have to. The voice platform and the enterprise integrations remain constant; the underlying intelligence layer can be updated without disrupting the deployment. Microsoft, Amazon, and Google cannot offer that model-agnosticism because each is fundamentally motivated to keep customers inside their own AI ecosystems.

What Enterprise Voice AI Does in Practice

For any business currently evaluating voice AI tools, it is worth being specific about what these systems do and do not do in production environments, because the gap between vendor descriptions and operational reality in this category has historically been wide.

Enterprise voice AI in a contact centre does not replace the entire agent workforce on day one. It handles what Mohajer described as the broad-based demand for automating interactions, improving efficiency, and delivering better customer experiences — starting with the interactions that are high-volume, structured enough to follow predictable patterns, and low-stakes enough that a mis-handled AI response can be escalated to a human agent without significant consequence. Appointment scheduling, order status inquiries, account balance checks, standard insurance queries, basic troubleshooting for telecom or utilities: these categories typically represent 40% to 60% of a contact centre's total call volume, and they are the categories where voice AI automation produces the most immediate and measurable cost savings.

The operational outcome that enterprise voice AI deployments are measured against is not raw automation rate. It is cost per resolution — the total expense of handling one customer interaction from first contact to satisfactory completion. A voice AI system that handles 45% of calls fully autonomously while maintaining customer satisfaction scores comparable to human agents produces a different cost structure than a system that handles 70% of calls but generates significant escalation traffic when customers reject its responses. SoundHound's Amelia 7 platform is specifically optimised for the latter metric, with escalation logic designed to route appropriately before a customer becomes frustrated rather than after, which is the failure mode that makes most early-generation voice AI deployments generate negative customer satisfaction data.

What This Means for Any Business Evaluating Voice AI

For a business currently deciding whether to deploy voice AI for customer service or internal operations, the SoundHound-LivePerson-Experis combination represents a materially different option from the hyperscaler default than existed twelve months ago. The practical implication is that the evaluation criteria need to expand beyond which platform has the highest speech recognition accuracy to include: which platform can integrate with the specific backend systems that your operations depend on; which platform's partner ecosystem can provide the implementation depth that a regulated industry deployment requires; and which platform's architecture gives you the flexibility to change underlying AI models without rebuilding the customer-facing deployment when the model landscape evolves.

The OASYS model-agnosticism is particularly relevant for any business that has been reluctant to commit to a voice AI deployment because of uncertainty about which foundation model will produce the best results in 24 months. That uncertainty is reasonable and well-founded given the pace of model development that the AI industry demonstrated in 2025 and 2026. A deployment architecture that separates the voice interface and enterprise integration layer from the underlying model makes it possible to start deploying now and update the intelligence layer later, rather than waiting for model certainty that the market is not going to provide.

Conclusion

SoundHound's position in enterprise voice AI at the end of the first half of 2026 is qualitatively different from what it was twelve months earlier, and the difference is the result of deliberate strategic choices rather than organic growth in a market that was already moving in its direction. The LivePerson acquisition brought an enterprise customer base that includes a majority of the world's largest banks, airlines, and automakers. The Experis partnership opened a channel into healthcare and financial services contact centre deployments through an implementation partner with the domain expertise those industries require. The OASYS platform created an architectural response to the competitive dynamic that the hyperscalers cannot match without undermining their own ecosystem strategies.

What this combination means for Microsoft, Amazon, and Google is a credible specialist competitor in the specific enterprise voice AI deployment scenarios where their general-purpose cloud tools have historically underperformed expectations. What it means for any business evaluating voice AI is that the menu of credible options now extends beyond the hyperscaler defaults, and that the specialist option has enterprise relationships, implementation depth, and model-agnostic architecture that deserve serious evaluation alongside the familiar hyperscaler alternatives. The enterprise voice AI market is growing fast. The competitive structure of that market just became considerably more interesting.