Decagon has launched Alto, a purpose-built conversational judgment model that powers the duplex layer of its Voice 3 architecture, delivering a 5+ percentage-point lift in resolution rates on live customer calls [1]. The gain compounds on top of the up-to-7-point lift already delivered by Chord, Decagon's speech model [1]. With customer support and experience ranking as the leading generative AI use case at 56.5% of surveyed organizations [2], Alto's production results arrive at a moment of peak enterprise demand for reliable voice agents.
What is Covered in this Article
- The speed-versus-accuracy tradeoff stalling enterprise duplex voice AI adoption [2]
- How Alto's judgment model resolves the speak/wait decision on every conversational turn [1]
- Compounding resolution gains from Decagon's purpose-built model stack: Chord plus Alto [1]
- Enterprise demand for agentic AI in customer experience [2][3]
- Alto's training methodology: reinforcement learning across 90+ million simulated calls [1][1]
The News: Decagon announced Alto on November 11, 2026, a new model from Decagon Labs that powers the conversational layer of its duplex Voice 3 architecture [1][1]. Alto determines on every turn whether to answer immediately or wait for the reasoning model to confirm an action before speaking [1]. On live customer calls, Alto increased resolution rates by 5+ percentage points with no other significant changes to the agent's logic [1]. The agent responds in about 700 milliseconds on average without relying on generic filler statements [1]. Alto was tested across more than 90 million simulated calls before any live caller interaction, then ran in shadow mode alongside live production traffic [1]. Deutsche Telekom's Christian Niedworok noted that callers "start talking to it naturally without even thinking about it" [1].
Decagon's Alto Closes the Voice AI Reliability Gap
Analyst Take: Alto is not a general-purpose model applied to voice, it is a narrow, production-hardened judgment engine built for one specific problem [1]. That specificity is precisely what makes the 5+ percentage-point resolution lift credible [1]. Decagon is demonstrating that compounding gains from a purpose-built model stack can close the reliability-versus-naturalness gap that has kept enterprise voice AI deployments cautious.
The Duplex Tradeoff and Why It Has Stalled Enterprise Adoption
Duplex voice architectures split work across two parallel models: a fast conversational layer that listens and speaks, and a slower reasoning layer that handles complex decisions. The fundamental tension is that smaller, faster models make more mistakes, while waiting for the reasoning layer disrupts conversational flow. The wrong call in either direction is costly, answering too soon risks telling a caller an action is complete before it is confirmed; waiting too long makes the agent feel broken. AI agent reliability and hallucination management in production is the top AI adoption challenge, cited by 55.4% of surveyed organizations [2]. That statistic maps directly to the problem Alto is designed to solve, and it explains why the enterprise market has been slow to commit to voice agents at scale.
Alto's Judgment Model: A Targeted Solution to a Specific Failure Mode
Rather than patching a general model or adding filler logic, Decagon built Alto to do one thing well: decide when to speak, wait, or keep the conversation moving while the reasoning layer works [1]. Alto continuously tracks caller context and grounds its responses in a shared record of the call, waiting for confirmation before reporting that an action is complete [1]. This confirmation-gated design is a direct response to the governance concerns intensifying around AI agents, formal monitoring of unauthorized access by AI agents rose from 42.8% to 52.2% in six months, the largest increase of any risk area tracked [2]. For risk-conscious enterprise buyers, that design choice is a differentiator, not just a performance feature. Deutsche Telekom's experience reinforces the point: callers who were trained for years to speak in short fragments at IVRs are now engaging in natural dialogue [1].
Compounding Gains Validate the Purpose-Built Stack Thesis
The headline number is 5+ percentage points, but the more important signal is that this gain is additive [1]. Chord, Decagon's speech model, independently lifted resolution by up to 7 percentage points [1]. Alto's gain builds on top of that baseline, not in place of it. This compounding dynamic validates Decagon's architectural bet: that a stack of specialized, production-trained models outperforms a single monolithic pipeline. The training methodology reinforces this credibility. Alto was fine-tuned on simulated calls, then trained via reinforcement learning with responses scored on correct speak/wait decisions and policy adherence [1]. Before any live caller interaction, it was tested across more than 90 million simulated calls and validated in shadow mode against live production traffic [1]. That is a production-readiness bar most enterprise voice AI has not cleared.
Market Timing: Enterprise Demand Is at Peak Alignment
Decagon is launching Alto into the strongest possible demand environment. Customer support and experience is the leading generative AI use case, cited by 56.5% of surveyed organizations [2]. Customer service automation has consistently ranked at the top across survey cycles, reaching 63.1% in the prior period [3]. Looking at agentic AI specifically, customer experience ranks among the top deployment areas for the next 18 months at 48.6% [2], and customer service and support was the single most-cited area for agentic deployment at 52.8% [3]. Meanwhile, 61% of respondents in a September 2026 AI Assistants pulse expect their usage mix to move further toward agentic workflows over the next three months [2]. The AI platforms market supporting this shift carries a base-case CAGR of 52.5% through 2030 [4]. Alto's voice-native, production-validated approach sits at the intersection of every one of these vectors.
What to Watch
- Resolution rate trajectory: whether the 5+ point Alto gain holds or compounds further as Decagon expands to new customer verticals beyond telecommunications [1]
- Competitive response: how cascade-pipeline voice AI vendors reposition or retrain their models in response to Decagon's published production benchmarks over Q1 2027
- Agentic deployment acceleration: whether the 61% of enterprises expecting to shift toward agentic workflows translate that intent into signed voice agent contracts in Q1-Q2 2027 [2]
- Governance and reliability standards: whether enterprise procurement teams begin requiring shadow-mode validation and confirmation-gated response design as baseline criteria for voice AI RFPs [2]
- Decagon Labs model cadence: whether Alto signals a repeatable research-to-production pipeline that yields a third compounding model layer in the Voice 3 stack
Sources
1. Introducing Alto: the voice model driving a 5+ point lift in …, Decagon, November 2026
2. 2H 2026 AI Platforms Global Enterprise Decision Maker Survey Report, Futurum Research, October 2026
3. 1H 2026 AI Platforms Decision Maker Survey Report, Futurum Research, March 2026
4. 2H 2026 AI Platforms Market Sizing & Five-Year Forecast, Futurum Research, November 2026
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Read the full Futurum Group Disclosure.
Other Insights from Futurum:
Siemens Energy's Record Q3 Makes It AI's Power Backbone
Fin Operator Routines: Proactive AI or Just Scheduled Tasks?
eClerx Acquires RegEdge to Deepen AI-Powered Compliance Capabilities
Author Information
This content is written by a commercial general-purpose language model (LLM) along with the Futurum Intelligence Platform, and has not been curated or reviewed by editors. Due to the inherent limitations in using AI tools, please consider the probability of error. The accuracy, completeness, or timeliness of this content cannot be guaranteed. It is generated on the date indicated at the top of the page, based on the content available, and it may be automatically updated as new content becomes available. The content does not consider any other information or perform any independent analysis.
