Teleconferência de Resultados do 3T 2026 da SoundHound AI (SOUN): Meta de Lucratividade para 2027

Principais destaques

• A SoundHound AI reportou receita recorde no terceiro trimestre de 2026 de US$ 62,49 milhões, alta de 48% na comparação anual, incluindo cerca de US$ 14 milhões provenientes de 26 dias sob o controle da LivePerson.
• A margem bruta não GAAP ficou em 50,3%, queda de 9 pontos percentuais na comparação anual, em razão do lançamento de novos recursos de IA agêntica e do maior consumo de nuvem.
• A empresa estreitou seu guidance de receita para 2026 de US$ 230 milhões a US$ 260 milhões para a faixa de US$ 250 milhões a US$ 260 milhões, elevando o ponto médio, apesar do encerramento de contratos com margens negativas e da migração de acordos de licenciamento para estruturas hospedadas.
• A administração reafirmou o guidance de receita para 2027 de US$ 350 milhões a US$ 400 milhões e prevê um EBITDA ajustado positivo e sustentável no segundo semestre de 2027, inclusive no piso dessa faixa de receita.
• A SoundHound identificou pelo menos US$ 120 milhões em reduções de custos anualizadas. Mais de US$ 80 milhões já foram implementados e devem entrar em vigor antes de 1º de janeiro, com outros US$ 40 milhões previstos para impactar os resultados ao longo de 2027.
• A receita pontual caiu para cerca de 10% da receita do 3T, ante mais de 33% no 2T, à medida que a SoundHound priorizou contratos hospedados e receita recorrente.

Principais dados financeiros

Desempenho operacional e de negócios

A SoundHound afirmou que o OASIS continuou a impulsionar a conquista de novos clientes, expansões e implementações lideradas por parceiros nos setores de saúde, restaurantes, automotivo, hotelaria, serviços públicos, seguros, varejo e comércio por voz. Quase 20 implementações corporativas do OASIS estavam em andamento, enquanto outubro gerou mais que o dobro das oportunidades registradas no lançamento, incluindo oportunidades dentro da base de clientes da LivePerson.

A empresa iniciou tratativas com as 90 principais contas corporativas da LivePerson, que representam mais de US$ 100 milhões em ARR. A integração completa da plataforma e da estrutura organizacional continua prevista para o início de 2027. A administração afirmou que a combinação adiciona mensagens digitais e integrações corporativas aos recursos de voz e de IA agêntica da SoundHound.

A SoundHound está migrando clientes para seus modelos proprietários de IA, com centenas já transferidos para essa arquitetura. A administração afirmou que os modelos podem reduzir os custos estruturais de modelagem em mais de uma ordem de grandeza, além de melhorar a precisão e a latência em casos de uso corporativos específicos. O OASIS continua agnóstico em relação a modelos, permitindo que os clientes usem modelos da SoundHound, modelos de fronteira de terceiros ou ambos.

A atividade comercial incluiu implementações no setor de saúde, uma expansão com o Resorts World Las Vegas, novas unidades de restaurantes e um contrato automotivo na China na casa dos sete dígitos com garantia mínima. A SoundHound também anunciou uma parceria com a LG para levar seu canal de comércio Reach a aproximadamente 20 milhões de Smart TVs da LG nos EUA, com lançamento previsto para os próximos meses.

Guidance da administração

A administração estimou que a simples adição da LivePerson à base de consenso pré-aquisição da SoundHound para 2026 teria gerado aproximadamente US$ 290 milhões em receita. O novo ponto médio do guidance, de US$ 255 milhões, reflete uma redução de US$ 35 milhões: cerca de US$ 8 milhões decorrentes do encerramento de contratos com margens negativas e da despriorização de linhas de negócios menos rentáveis, somados a cerca de US$ 27 milhões provenientes do reconhecimento da receita de licenciamento ao longo do tempo em contratos hospedados.

A redução de custos planejada de US$ 120 milhões engloba cerca de US$ 48 milhões em sinergias típicas de M&A, US$ 15 milhões em operações orientadas por IA e sistemas, US$ 42 milhões em consolidação e otimização de nuvem e US$ 15 milhões na realocação de recursos para produtos essenciais e iniciativas de go-to-market.

Riscos e pontos de atenção

• A transição de licenças antecipadas para contratos hospedados reduz a receita pontual no curto prazo, embora a administração espere que isso melhore a ARR, o recebimento de caixa, a previsibilidade e a estabilidade trimestral.
• A margem bruta não GAAP recuou à medida que novos recursos de IA agêntica foram implementados e o uso de nuvem aumentou. A expansão futura depende em parte da consolidação de infraestrutura, do encerramento de contratos deficitários e da maior utilização de modelos proprietários.
• Alcançar a rentabilidade do EBITDA ajustado no segundo semestre de 2027 depende da execução dos cortes de custos remanescentes e da manutenção do ritmo comercial dentro da faixa projetada de receita.
• A integração da plataforma, da estrutura organizacional, da infraestrutura de nuvem e da base de clientes da LivePerson continua em andamento.
• Os contratos de licença antecipada historicamente oscilaram em tamanho, tornando a receita reportada menos previsível. A SoundHound está reestruturando esses acordos para reduzir essa volatilidade.

Destaques da sessão de perguntas e respostas com analistas

A administração afirmou que os modelos proprietários da SoundHound são geralmente preferidos para aplicações de voz devido à sua velocidade, latência, personalização e perfil de custo. Para cargas de trabalho agênticas, a empresa está utilizando modelos menores e otimizados para atendimento ao cliente, mantendo integrações com provedores de modelos de fronteira. A expectativa é que, em última análise, a maioria das interações seja executada em modelos da SoundHound.

Ao ser questionado sobre quais indicadores antecedentes os investidores devem acompanhar durante a transição do modelo de receita, o CFO John Collins afirmou que o EBITDA ajustado se tornará uma métrica regular de guidance em 2027. A empresa está avaliando indicadores operacionais adicionais e espera fornecer mais informações na teleconferência de resultados do 4T.

Sobre o cronograma de custos, Collins afirmou que economias adicionais um pouco abaixo da faixa média de um dígito em milhões de dólares devem entrar em vigor no 4T, potencialmente com reduções adicionais. Espera-se que cerca de US$ 80 milhões em economias anualizadas estejam totalmente em vigor no início de 2027, seguidos por outros US$ 40 milhões ao longo do ano.

A administração também afirmou que os modelos proprietários da SoundHound permitem reduzir ou eliminar limites associados a tokens em alguns contratos, diferenciando sua precificação da de provedores que dependem fortemente de modelos de fronteira de terceiros.

Transcrição completa da teleconferência de resultados

Transcrição completa da teleconferência de resultados

Comentários da administração

Operator

[Operator Instructions]

Please be advised that today's conference is being recorded. I would now like to hand the conference over to your first speaker today, John Perrocchio, Vice President, Investor Relations.

Unknown Speaker

Good afternoon and thank you for joining our third quarter 2026 conference call. With me today is our CEO and co-founder, Keyvan Mohajer, and our CFO, John Collins. We'll begin with some short remarks before moving to Q&A.

I'd also like to remind everyone that we will be making forward-looking statements on this call. Actual results could differ materially from those suggested by our forward-looking statements. Please refer to our filings with the SEC for a detailed discussion of the risks and uncertainties that could affect our business and for discussion of the statements that qualify as forward-looking statements. We may discuss certain non-GAAP measures. Please refer to today's press release for more detailed financial results and further details on the definitions, limitations, and uses of those measures in reconciliations from GAAP to non-GAAP.

Also, note that the forward-looking statements on this call are based on information available to us as of today's date, and we take no obligation to update any forward-looking statements except as required by law. Finally, this call is being audio webcast in its entirety on our Investor Relations website. An audio replay will be available following today's call.

With that, I'd like to turn the call over to our CEO, Keyvan Mohajer. Please go ahead, Keyvan.

Unknown Speaker

Thank you, John, and thank you to everyone for joining the call today. This is a defining moment for SoundHound. With a strong balance sheet, nearly zero debt, and increasing revenue, we are now poised for the first time to achieve sustainable profitability within the second half of 2027.

We are on track to generate about $1 million in revenue per day and are increasingly well positioned to take a sizable share of the predicted $1 trillion addressable market in the new agentic era. We are doing this by providing one of the largest customer rosters in our space with the highest quality, most comprehensive agentic AI solution for enterprises. We process billions of interactions across millions of devices and thousands of businesses, all at a competitive price for our customers thanks to our in-house models.

We believe every business, product, and individual will have an AI agent that listens, learns, and acts, interacting naturally with people and each other. SoundHound is building the operating system for that agentic world. Our OASIS platform integrates decades of innovation from 6 pioneering companies we have merged under SoundHound AI. This technical differentiation sets us ahead of the pack, and you don't have to take our word for it. The industry experts at analyst firms, including Gartner, Everest, IDC, and ISG, ranked SoundHound as a category leader.

And in Q3, I'm delighted to report that we delivered another record revenue quarter of $62.4 million, up 48% year over year. We delivered this with another significant improvement to the quality of our revenue. A year ago in 2025, our Q3 revenue mix contained more than 23% point-in-time contributions from certain deals. That percentage grew to over 33% in Q2 of this year. While the growth and repeatability of upfront license deals is a strong sign of demand for our solutions, their fluctuating nature can make our financials less predictable.

This quarter, we reduced the contribution from such deals to about 10%, and we expect to continue this improving percentage going forward. As a result, the vast majority of our revenue will become recurring, and as we will discuss shortly, we believe we can do this while still signing long-term customer commitments because of our OASIS platform.

In Q3, we closed new deals, renewals, and expansions across key verticals including healthcare, hospitality, automotive, utilities, restaurants, voice commerce, and more. I will talk through these in more detail shortly before handing over to John Collins, our new CFO, who will walk you through the numbers.

First, a LivePerson update. I'm pleased to say we remain firmly on target for full platform and organizational integration by the beginning of next year. SoundHound has brought businesses together before, and each time our playbook runs faster. We have already engaged with LivePerson's top 90 enterprise accounts, representing over $100 million in ARR. In those conversations, the feedback has been clear and overwhelmingly positive.

Before the acquisition, LivePerson faced customer hesitation and churn driven primarily by 2 factors: concerns over their financial stability and the absence of an enterprise-grade voice and agentic AI solution. With this merger, we have eliminated both friction points overnight. At close, we retired virtually all of LivePerson's debt, leaving SoundHound effectively debt-free, backed by over $200 million in cash and cash equivalents.

We also expect to reach sustained profitability next year, providing our enterprise clients absolute confidence in our long-term financial stability and operational durability. We can now also deliver SoundHound's market-leading voice AI and the OASIS agentic platform directly into LivePerson's existing customer base. And we are already showing enterprises what's coming.

At Customer Contact Week in Amsterdam just last month, our teams demonstrated how our platforms will work together. This speed of execution shows how well our teams have been able to innovate together. Attendees at the show were excited by what our combined technologies can do. In parallel with our product innovation, we are executing on cost synergies by removing G&A overlaps, consolidating software vendors and tools, and unifying our go-to-market teams. These savings are significant, and John will share more details shortly.

We believe our combined organization now has the most comprehensive AI solution for enterprises: 1 unified platform that brings together world-class voice, agentic, and digital messaging capabilities, natively omnichannel across voice, web, mobile, SMS, and social. Underneath OASIS is a deep technology stack backed by over 800 patents and built by SoundHound over decades. It spans speech, language intelligence, agentic orchestration, and increasingly our own specialized AI models. This is important both to our own business and for our customers.

Years of real-world data and model development allow us to optimize our in-house models for the specific enterprise use cases we serve. And with our recent investments in machine learning, we can now run complete customer interactions on our own AI stack without requiring third-party frontier models. We have already migrated hundreds of customers to this architecture. These deployments are delivering better accuracy and latency for our customers and our end users, while importantly reducing structural SoundHound model costs by more than an order of magnitude.

We plan to continue to migrate more of our customers over the coming months, and over time, we expect the majority of our traffic to be powered by SoundHound models. But to be clear, OASIS is model agnostic. Customers can run frontier models, SoundHound's own models, or a combination. This makes SoundHound a much safer choice for enterprises compared to going exclusive with a frontier model provider. We believe that for many high-volume enterprise implementations, our own models can offer an especially compelling combination of customer experience, control, and economics.

And I'm especially excited about the progress we are making in real-time full duplex and multimodal agentic AI, technology that allows AI to listen, respond, adapt, and take action continuously as the conversation unfolds. We believe this represents an important next step for enterprise AI and we look forward to showing much more of what we have built in the near future. And because OASIS is multimodal and omnichannel, the same platform can power agents across voice, chat, and messaging. That intelligence can also run wherever a business needs it to, allowing them to meet their customers where they are.

This quarter, we introduced OASIS Edge, bringing LLM-powered agentic voice AI directly onto vehicles and smart devices. Because processing can happen locally, interactions can remain private and continue even when network connectivity is unavailable. OASIS Edge allows manufacturers and service providers to deploy fully embedded agentic AI and extends a core OASIS principle: build an agent once, then deploy it in the cloud, on the edge, or across a hybrid environment.

Owning the technology stack gives SoundHound important advantages. It gives us control, it gives us independence, it lowers our costs, and it allows us to continuously differentiate the experience we deliver to our customers. It also opens opportunities with enterprises where security and privacy are fundamental requirements, including customers in highly regulated industries that want to protect sensitive data from flowing through third-party model APIs or that need AI to run inside infrastructure they control, using our models while still delivering the same OASIS agentic experience.

With this proprietary technology foundation in place, OASIS is driving momentum across every enterprise touchpoint. Last quarter, I told you we were winning in demos, in RFPs, in pilots, and in production, and that our biggest Q2 deals were driven by OASIS. That continues in Q3. Growth is coming from 3 places: new customers signing up, existing customers buying more, and partners opening new doors.

In Q3, we signed new customers across healthcare, restaurants, automotive, devices, voice commerce, insurance, retail, and corporate services. In healthcare, a leading academic health system chose us for both its hospital services and its health plan. And Randox, a health testing provider serving the U.K. and global population, chose us for customer care. Healthcare is where trust, compliance, and accuracy are non-negotiable.

And leading providers now consistently choose SoundHound to automate routine and complex patient access, easing operational strain with our agentic architecture. Via OASIS, our AI agents integrate directly into core EHR systems to resolve 24/7 appointment scheduling, prescription refills, and billing inquiries with deterministic guardrails to keep high-stakes workflows on track and keeping clinical staff in the loop when needed. This way, our platform delivers an effortless end-user experience anchored by enterprise guardrails and strict governance.

And while OASIS can handle most interactions end to end, it's also able to incorporate human oversight in real time on voice and chat without the need to transfer the conversation over. This is a unique advantage for OASIS: Human-Assisted Resolution, our patented technology that powers some of the world's largest companies. This technique is more precise and effective than escalation alone. It strategically inserts real-time human input into autonomous AI-driven conversations when necessary to deliver the best results and drive efficient resolution with the objective of avoiding a full human handoff. The end user continues to experience a seamless conversation and businesses get to save their experts for the moment they need human judgment.

Beyond patient and consumer contact centers, our technology is also performing for employees inside the enterprise where more businesses are looking to optimize their workforce with AI. A major U.S. energy company signed a new contract to bring AI to its back-office functions, and a multinational food and beverage company renewed its HR deployment. When AI serves employees as well as customers, resolving tickets, executing password resets, providing PTO status, or surfacing key internal documents, it stops being a single project and becomes a core part of how a company runs. This deepens the relationship, proves the performance and value of our platform, and gives us more place on the market.

OASIS is also reaching new places through our partner network. Last quarter, I told you about our multi-year partnership with a Latin American company whose customers span more than 20 countries. This quarter, our first joint build went live: Línea 195, Bogotá's 24/7 line for citizen services serving 8 million residents. An AI agent named Alex now answers residents' calls in Spanish across 76 city districts, handling up to 1,000 calls simultaneously, where every answer traces to an official source. Because of this deployment, the average wait time for residents is reduced significantly and call abandonment dropped. The number of inquiries handled increased by 30% in just the first phase.

Through our reseller partner, Alta Resources, we also won a major global consumer products company as a new end customer. OASIS layers over human staff service operations and lets AI do as much of the work as possible. Outsourcers, resellers, and cloud marketplaces, where we are now listed, give us a path to thousands of enterprise clients through partners who already hold the relationship.

Existing customers are also buying more, a sign our deployments are making an impact for their workforce, their customers, and their bottom line. In telecommunications and insurance, customers added volume. In financial services and healthcare, they expanded as they renewed, and in government, long-running programs were further extended. Our momentum extends directly from enterprise workflows into high-volume physical environments.

Hospitality shows where this is heading. A global resort operator, Resorts World Las Vegas, expanded in Q3, moving from an on-property digital concierge to an external-facing booking agent, the place where AI touches revenue directly. When a customer hands AI the job of booking the room, the technology is entirely out of its test phase. It's a vote of confidence and a preview of what agentic AI will be trusted to do across every industry.

And our research shows consumers are ready for it. A recent study commissioned of more than 1,000 Americans found that nearly 3 in 5 would let an AI agent complete a purchase on their behalf, a strong indicator of a significant cultural shift we believe is happening right now. The same study found that customer satisfaction more than doubled when people watched an AI agent resolve an issue end to end.

The returns on this kind of fast agentic AI are easy to measure in quick-service restaurants and they are striking. One large, well-known national sandwich chain is now using our technology to regularly generate multiples of what it spends on our service each month by allowing customized SoundHound AI agents to answer 100% of calls and take orders. The locations where AI is deployed have also seen an increase in basket size. A leading quick-service chain using SoundHound technology in its drive-thrus paid back its investment in just a couple of weeks. In addition, with AI handling the drive-thrus, their team could focus on guests inside the restaurant, and as a consequence, non-drive-thru sales actually improved compared with control stores. Perhaps unsurprisingly, retention of restaurant locations is an all-time high this quarter, and a household-name pizza chain added our AI to close to 100 additional locations.

We believe every business will have an AI agent and every product will have an AI agent. I have talked about how OASIS powers the AI agents of businesses across multiple industries. Let me talk about what happens when it leaves the building and enters the car, the device, or the TV. Cars are where our voice AI has been proven the longest, and the story is global. For several quarters in a row, we signed new seven-figure deals in automotive and IoT with new brands in Asia. This quarter in China, we signed a seven-figure minimum guarantee agreement with a top-tier automotive supplier for our voice AI solution. Chinese suppliers and automakers now consistently choose us because our technology performs better than the alternatives, including on-device support, even when the connection drops.

In India, a leading electric two-wheeler maker, Greaves Electric, signed a new deal for a voice assistant powered by SoundHound. And makers of industrial equipment and autonomous vehicles are now exploring voice and visual control of complex machinery. This is a validation of our vision. If it has a microphone and a purpose, it should have an AI agent, and OASIS can power it.

That brings me to voice commerce, which is evolving into multichannel agentic commerce. Consumers are ready. As I mentioned earlier, a recent study found that the majority of consumers would let an AI agent make a purchase for them. SoundHound is positioned to meet that growing interest with a new agentic channel where anyone can discover merchants, order, and pay directly from the car, the TV, and other connected devices. Businesses and brands reach customers at the precise moment that they decide.

Restaurants are our first test case, but this opens up incredible opportunities across hospitality, travel, entertainment, banking, retail, and more. This quarter, we built Reach. We just announced our partnership with LG to bring a new commerce channel to approximately 20 million LG Smart TVs in the U.S., expected to debut in the coming months. We also signed voice commerce deals with a large coffeehouse brand and a growing fast-casual chain, with more on the way.

Now a word on a strategic shift in how we run our business. Given our new scale, the cost synergies provide an opportunity for the first time to achieve sustainable profitability. We took this opportunity seriously and have identified $120 million of savings. We have already taken action on the majority of this and we expect to reach this number in 2027. We expect realizing this number will make us adjusted EBITDA profitable even at the lower end of our revenue guidance. This is a huge milestone for us.

As a publicly listed pure-play AI pioneer that is profitable with a strong, clean balance sheet and a technical moat, we expect to be a major force in the agentic era for years to come. Our pace of innovation has accelerated beyond our own imagination, and our data and experience benefit us in unique ways. We have also been strategically non-renewing and exiting certain negative gross margin contracts inherited from prior acquisitions and expect to continue doing so with a positive impact on both gross margin and EBITDA.

Finally, thanks to OASIS, we are able to prioritize predictable recurring revenue over licensed deals that have a point-in-time revenue impact. While these licensed revenues are not recurring by nature, they have been highly repeatable. We've had them every quarter since we went public and their magnitude has increased by nearly 50% year over year in the last 12 months. These are true organic deals we're closing when a customer is willing to make a large commitment to SoundHound. This is a testament to the growing demand for our solution. However, the size of these deals has historically been fluctuating, which makes them less predictable.

Now, with the self-learning feature of OASIS, we are able to structure these deals as hosted solutions with recurring revenue, while still signing long-term commitments from our customers. We have already started this transition in Q3 with the less desirable near-term point-in-time revenue decreasing and more valuable high-margin recurring revenue expanding. Again, this strategic shift is possible notably due to our OASIS platform, which delivers more value to our customers as a hosted service. The expected impact is more predictable and longer-term growable ARR. Together, these actions upgrade the overall quality of our revenue and transition SoundHound to a cleaner, more predictable, and higher-margin financial model.

Before I hand it over, I want to welcome John Collins, who joined us at the close of the LivePerson deal as our Chief Financial Officer. John brings more than 15 years of leadership at the intersection of enterprise software, data science, capital markets, corporate finance, and AI. He has been a founder and entrepreneur and he has served as the Chief Financial Officer, Chief Operating Officer, and Interim Chief Executive Officer at LivePerson. So he understands both the opportunity and the operating realities of the business we brought in.

He also brings the financial discipline this next phase demands. Prior to joining SoundHound, he led a transformation from more than $100 million of annual cash burn to positive free cash flow in a single year while supporting double-digit growth. He executed cost reduction programs in excess of $200 million and led multi-year debt restructurings that captured $227 million of debt discounts.

That rare mix of operator, financial steward, and entrepreneur makes John an ideal addition to our leadership team. John will oversee our unified financial strategy, instill financial discipline across the combined company, drive the operational synergies as LivePerson moves onto OASIS, and guide our path to sustainable, high-margin, profitable growth. John, welcome.

John Collins

Thank you, Keyvan. Considering this is my first call as CFO, I'd like to share my perspective on the opportunities for the combined company in this dynamic market. During my 7 years at LivePerson, I saw a great many enterprise implementations of AI-powered customer service offerings and gained a robust understanding of the CCaaS competition vying for our customers. While enterprise customers frequently swapped vendors, none of the major incumbents ever gained a commanding share of the market. I believe this was because the offerings, while differentiated on certain features, were ultimately too interchangeable. None were 10x better than the competition.

Since joining SoundHound, I've seen OASIS and its voice AI in action, and it was instantly clear that SoundHound's platform represents a step change in voice-based agentic capabilities that could deliver a 10x improvement relative to incumbent offerings. With the acquisition of LivePerson, SoundHound expanded its commercial footprint with dozens of Fortune 500 brands and hundreds of enterprise accounts across the globe. The acquisition also provided access to deep integrations across major digital channels and the enterprise software stack, which will further extend SoundHound's omnichannel offering and the surface area for agentic actions. We are excited to see the combined company leverage these assets to accelerate enterprise revenue growth.

As CFO, my objective is to pair SoundHound's product innovation and market momentum with rigorous financial discipline, improved quality of earnings, accelerated synergy execution, and strategic capital allocation for long-term value creation. With the integration actions we've already taken, which I'll discuss shortly, I'm confident that we'll both accelerate revenue growth and generate positive adjusted EBITDA within the second half of 2027.

My remarks today have 2 goals: to demonstrate a clear, credible path to sustainable, profitable growth in 2027, and to be direct regarding how revenue mix will evolve and improve earnings going forward.

Beginning with the most important message in my remarks, we are reaffirming revenue guidance of $350 million to $400 million for full year 2027. And, for the first time, guiding to sustainable, positive adjusted EBITDA within the second half of 2027, a milestone you can expect even at the low end of this revenue range. Further, we expect meaningful gross margin expansion as our cost actions continue to layer into the P&L throughout 2027. This is a first for SoundHound. Starting in 2027, adjusted EBITDA will be a regular component of our guidance, a numeric metric you can rely on going forward. Importantly, our profitability guidance is built on actions we've already executed, which I'll elaborate on now.

Since the close of the LivePerson transaction, we moved swiftly and decisively, executing actions that reduced the combined company's annualized cost structure by over $80 million. We expect these actions to be fully effective before January 1. We are also in the process of executing $40 million in additional cost reductions, which will layer into the P&L throughout 2027, bringing our aggregate cost reduction to at least $120 million.

Broadly, the reductions to our cost structure fall into 4 categories. First, standard M&A synergies. We have consolidated teams in each functional area and rationalized spans of control. Similarly, on the non-labor side, we're consolidating overlapping software, database, consulting and advisory, and other vendors across the two organizations. This category of reductions accounts for approximately $48 million.

Second, AI and systems-driven operations. We build agentic AI for a living, and we are applying it to our own operations so that we can do more with fewer resources across every dimension of the business. This category accounts for approximately $15 million.

Third, consolidation and optimization of cloud spend. Given the breadth of the combined company's cloud spend, these reductions warrant a separate category. SoundHound acquired LivePerson as it was nearing completion of a multi-year effort to migrate its production traffic to GCP. With that migration now complete, we have already begun decommissioning legacy infrastructure, and this will continue throughout 2027. In addition, with a comprehensive view of both companies' cloud workloads, we are eliminating unnecessary spend and shifting substantial volumes to pre-existing commitments to take advantage of discounts and superior fee structures. This category accounts for approximately $42 million.

Fourth, strategic reallocation of resources to products and go-to-market. We are concentrating investment where we win. That means realigning go-to-market product and engineering and pausing or exiting non-core dilutive business lines, including some with negative gross margins that we inherited in prior transactions. This category accounts for approximately $15 million.

Together, these actions total at least $120 million of annualized cost reductions, at least $80 million of which are actioned and scheduled to be effective before January 1.

Considering several actions directly lower our cost of revenue, particularly those in infrastructure, cloud, and product, we are confident in expanding gross margin over time. While the direction is clear, we are not guiding a numeric level at this time.

Next, I'll discuss key factors driving our reaffirmation of 2027 revenue guidance. First, accelerating momentum with OASIS across new logos, expansions, and channel partners, translating to nearly 20 enterprise implementations of OASIS already underway. Moreover, in October, we generated more than twice the opportunities that we did at launch, including many with LivePerson customers. Second, our guidance also reflects a change in how we structure licensing deals. In short, by hosting the solution for our customers, we expect multi-year contracts to yield revenue over time, instead of at a singular point in time.

Turning to 2026, taking into account our record third quarter, post-close contributions from LivePerson, and strategic decisions to improve earnings quality, we are tightening the range for 2026 revenue guidance to $250 million to $260 million, relative to our prior range of $230 million to $260 million. As we've already alluded, there are 2 primary changes driving improved earnings quality. First, we are exiting negative gross margin contracts inherited from earlier acquisitions and deprioritizing several low-growth lines of business. Together, we expect these strategic decisions to meaningfully increase profitability and center our focus on OASIS as the core driver of growth. Second, we are moving new multi-year agreements from upfront licenses to hosted structures.

On the second point, I want to underscore the growing and durable customer demand that underlies these upfront license deals. While such deals yield less recurring revenue, they are highly repeatable. For perspective, we've closed license deals for the last 18 consecutive quarters. Notably, we recognized approximately $20 million in revenue from license deals in Q2 alone, and we have a robust pipeline of deals for the remainder of the year. However, with the rollout of OASIS and its self-learning capabilities, we can now deliver more value to these customers if SoundHound hosts the solutions rather than the customer.

Converting licensed deals to hosted structures also improves our quality of earnings. While the hosted structure will reduce near-term point-in-time revenue, in return, we expect predictable ARR, faster cash collections, and improved revenue visibility and P&L stability quarter-to-quarter. These improvements are essential for sustainable profitability beginning within the second half of 2027.

To quantify the impact of these changes, recall that SoundHound's pre-acquisition consensus of $238 million excluded LivePerson. Here's the bridge from that figure to our new midpoint. First, if LivePerson revenue were purely additive, we would expect total 2026 revenue to be approximately $290 million. Second, we are guiding to a midpoint of $255 million, which implies that our measures to improve earnings quality result in a reduction of $35 million. Third, this $35 million can be divided into the following categories: approximately $8 million of lower revenue due to our decisions not to renew negative margin customer contracts and to prioritize OASIS over low-performing business lines, and approximately $27 million of lower licensing revenue in the second half of 2026. As implied by our 2027 guidance, these short-term revenue changes do not impact our prior 2027 outlook.

With that, I will turn to our third quarter results. In the third quarter, we delivered record total revenue of $62.49 million, up 48% year over year. That growth was driven by strong demand for our products and nearly 1 month of revenue contribution from LivePerson following the September 4 close. Non-GAAP gross margin was 50.3%, down 9 percentage points year over year. The decline was driven by the ramp of new agentic AI capabilities and higher cloud usage.

Looking forward, we expect gross margin expansion, driven primarily by 3 factors. First, reductions in legacy infrastructure and cloud consolidation and optimization post-acquisition and following LivePerson's migration of production volume to GCP. Second, non-renewal of negative gross margin customer contracts inherited in past transactions. And third, increased deployment of SoundHound proprietary AI models in place of more expensive frontier models.

As for cash, we ended the quarter with over $200 million in cash and cash equivalents, and the balance sheet for the combined company is debt-free with the minor exception of approximately $500,000 of outstanding 2026 stub notes, which will be retired at maturity in the fourth quarter. Finally, note that a reconciliation of non-GAAP to GAAP measures is included in today's press release.

To summarize, we are reaffirming 2027 revenue guidance of $350 million to $400 million, and for the first time guiding to sustainable, positive adjusted EBITDA profitability within the second half of 2027. That outlook rests on continued commercial momentum with OASIS and at least $120 million of annualized cost reductions, at least $80 million of which are already actioned and scheduled to be effective before January 1. As those cost reductions layer into the P&L, we expect gross margin to expand in 2027 and beyond.

We are raising the midpoint of our 2026 revenue guidance and narrowing the range to $250 million to $260 million, even as we take steps to improve earnings quality by exiting negative margin contracts and moving from upfront licensing to hosted structures. Those steps give us more predictable, more visible, and more stable revenue that is better matched to customer usage along with faster cash collections.

Q3 marked an important operational and financial milestone. We retired LivePerson's debt, prioritized revenue quality, and implemented a disciplined operating model built for sustainable profitability, all while maintaining our growth trajectory. I'm excited by the combined company's opportunities, and I look forward to keeping you updated as we execute. With that, operator, let's open the line for questions.

Operator

[Operator Instructions]

Our first question comes from the line of Brian Schwartz with Oppenheimer.

Perguntas e respostas

Brian Schwartz

Keyvan, I had a question on how you prioritize using your proprietary models versus kind of partner models because, you know, you're positioning the OASIS platform that now it's going to be integrating 6 businesses. So, what do you look at in terms of measurable customer revenue or unit cost advantage that requires you to use these proprietary models that you're building? And what investment guardrails allow you to pursue that differentiation without jeopardizing the second half '27 profitability commitment that you're giving to us today?

Unknown Speaker

Thank you for the great question. I'm glad you asked. That is a unique advantage to SoundHound. As you know, we've been a pioneer in the space of voice AI and we have our own foundation models. We've already made all the investments over the years to build these models and they outperform the frontier models and big tech in many ways, both in terms of speed and accuracy and significantly on costs.

So, to answer your question, for voice models, it's kind of a no-brainer to use our models because we check all the boxes in terms of being superior. It's faster, lower latency, and lower cost. You can customize them better for our customers.

The next level is when we upgrade our customers to agentic. We now have thousands of businesses using our platform. A lot of these businesses started using us before agentic was proven. And now as they upgrade to agentic, all of a sudden, if we use frontier labs' models, the cost goes up significantly, especially compared to what they are used to paying. In many cases, you're able to ask them for an upgrade, so it's an upsell to go to agentic, but still there are significant costs that most of our competitors and other vendors have to tolerate when they offer agentic for the first time to their customers.

We talked earlier this year about making an investment to have our own foundation models for end-to-end experiences. So, not just for voice, but also for reasoning and agentic orchestration. We achieved that milestone earlier. We have upgraded or moved hundreds of our customers to our own models, and it's actually performing better than the foundation models because they are focused on customer service.

When frontier model labs generate their models, it has trillions of parameters, it can answer questions about quantum physics and poetry, and it has a lot of information embedded in them. But in our case, we can have smaller models for customer service that outperform the frontier models. So we can run them at orders of magnitude lower cost, but higher accuracy and better performance. So we've already migrated a lot of customers and it's working really well. Our goal is to migrate the majority of our customers. Ultimately, most of the interactions will be powered by SoundHound.

We still integrate with frontier labs to have that option, but again, most of the interactions will go through our models. Now, the impact of that is lower COGS, significantly lower COGS, especially as we migrate all of the customers to our own models.

Brian Schwartz

And the follow-up question I have for John, thank you very much for all the disclosures and helping to bridge the changes with the guidance that you gave out today. Given the revenue model transition, what are now the best leading indicators that investors should monitor going forward? Should we place greater weight on ARR, your contracted backlog, consumption metrics, or is there another KPI because traditional revenue growth over the next several quarters is going to become less indicative of underlying demand?

Unknown Speaker

John, I think you might be on mute.

John Collins

Thanks, Brian, for your question. We're in a good position to reevaluate the metrics that we release on a recurring basis. And I've mentioned already 1, which is adjusted EBITDA, which will continue going into 2027. There are others we're considering, but we're not ready yet at this time to introduce new metrics beyond adjusted EBITDA. But you can expect that we'll have more, particularly in the way of leading indicators on the fourth quarter call.

Brian Schwartz

Great. Thank you for taking my questions.

Operator

[Operator Instructions]

Our next question comes from the line of Jon Hickman with Ladenburg Thalmann.

Jon Hickman

John, can you elaborate a little bit on—you did a good job on the revenue side. Can you give us a little color on the expense side for Q4? I know you're going to cut all these expenses out, but what was the base that you started with?

John Collins

Hi, Jon. I think what you can expect from us regarding additional cost-out inside of the fourth quarter will be a little less than, let's say, mid-single-digit millions that will be effective within the fourth quarter, and potentially considerably more. And then as we get—you know, that number will continue to grow in terms of full effectiveness as we enter January and have full-year effectiveness to the tune of approximately $80 million, as we discussed in the prepared remarks, for full year '27. As we get further into '27, there are additional cost actions that we've already identified that will be fully effective as we progress through the year, totaling an additional $40 million and $120 million in total for 2027.

Jon Hickman

Well, I guess, can you answer the question about what was the base that you started with? I mean, if you just added in LivePerson with your quarter, what would operating expenses have been? Give us a base that we can start subtracting the $80 million from.

John Collins

Right. We didn't put that in our PR today, but you can rest assured that it will be in our upcoming filing.

Jon Hickman

Okay. Thank you.

Operator

[Operator Instructions]

Our next question comes from Leo Carpio with Joseph Gunnar.

Leo Carpio

I actually have 2 high-level questions. Can you talk a little bit in terms of the competitive environment in the quarter? Have you been seeing—has it been stable? Are the hyperscalers moving into your space?

And then secondly, in terms of your pricing, since you seem to be moving toward a more strict pricing discipline here, was there any industry environment pressure that compelled you to make this shift that you're undergoing or is this more a review of the portfolio that you've conducted and decided to optimize?

Unknown Speaker

Yes, I'll take the second one first. On the pricing, actually we have an advantage because we have our own models. So, I'm going to mix the 2 questions together.

The 2 types of competitors we have: 1 is the big techs that have frontier models. They have high costs and they are also mostly API vendors, whereas we partner with our customers. Most of our customers don't want an API vendor; they want a partner. We sit with them, we listen to them, and we help them dream big and achieve those dreams. The relationship we establish with them is very strong.

And then there are new players in our space that are using frontier models to deliver agentic experiences. Their costs are high, and they have introduced this concept of token costs for their customers, which is discouraging for customers and also limiting. Now, because we have our own models, we can run everything end-to-end using our own models. Our costs are lower, and we don't have to worry about this concept of token costs. So in some of our contracts, we actually remove the token limits or costs associated with tokens, and that really gives us an advantage.

Maybe a little bit higher-level answer to your question about the competitive landscape: there's validation in our space that agentic AI is being heavily adopted by enterprises. Our vision is if there's a product that has a microphone, whether it's a car or an IoT device, it needs to have its own AI agent, and our OASIS platform can power the AI agent of physical products automatically. Or if it's a business, whether it's a very large healthcare provider or insurance company or a single-location barbershop, it can have and should have its own AI agents, and our OASIS platform can be the AI agent of businesses as well.

So we imagine a world where every physical product and business will have an AI agent. These agents can interact with the people that use them and with each other. We are building the operating system of this new agentic era. And while there's a lot of noise in the market, we are highly differentiated because we are a pioneer in this space. We've been at this for decades with the companies we've acquired. We have the most comprehensive solution, we have the largest customer roster, and again, we have our own foundational technology.

Leo Carpio

Okay, and then a quick housekeeping question. For the third quarter, you mentioned LivePerson contributed 1 month's worth of revenue. Could you break that out in terms of legacy SoundHound versus LivePerson in terms of the revenue contributions? Thanks.

John Collins

I'll take that, Keyvan. So we provided some math to bridge the guide. And if you work through the math, you'll see the implied contribution from LivePerson is roughly $52 million for the second half. And so in Q3, there's about 26 days of effectiveness for 1 quarter, which works out to be about $14 million of that figure.

Leo Carpio

All right. Thank you.

Unknown Speaker

Maybe I'll add a little bit on top of that. As we explained very clearly in our remarks, legacy SoundHound has had very large deals that had a point-in-time revenue contribution. If you look at the last 12 months, the contribution of those types of deals has gone up by 50% compared to the prior 12 months. Now, these are organic deals that SoundHound has secured and signed in that quarter, so they don't come through acquisition. The fact that we've had them every quarter for the last 18 quarters and the magnitude has gone up by 50% really shows the level of growth and demand for our products, and that demand has only gone up.

The shift that we are introducing starting Q3 is we are able to offer a hosted version of the same software to all of our customers because OASIS as a hosted service delivers more value. We are able to still sign those deals and get the big commitment from our customers, but recognize the revenue over time because we are the ones hosting it. That change started in Q3, shifting the composition of our point-in-time revenue versus over-time revenue to 10% versus 90%, which is really what we want, whereas a quarter before that, it was 33% point-in-time and the rest over time. So it's an improvement that we started in Q3 and we expect to continue that going forward.

John Collins

I'd add 1 more point on this. If you look at last quarter Q2, there was just over $20 million of point-in-time licensing revenue. And the pipeline, as I mentioned, continues to be extraordinarily robust. This is a sign of durable and growing demand from the customer base that we fully intend to sign, but under a different structure that will have more benefit to '27 and '28 relative to all point-in-time in 2026, just to ensure that point is clear.

Operator

Thank you. I'm currently showing no further questions at this time. This does conclude today's conference call. Thank you all for your participation. You may now disconnect.

This live transcript is auto-generated without human intervention or review.