Speak AI Language Tutor 2026: Voice Agents for Language Learning

The 30-Day Headline

Speak hit a milestone in April 2026 worth pulling apart in detail. Most coverage focused on the launch event and the funding number — the actual interesting story is in the deployment numbers, the new pricing model, the named enterprise customers signing in the first 30 days, and the architecture decisions buyers are making in response.

We'll walk through the announcement, the platform architecture as documented in vendor materials, the customer wins we have public confirmation on, the comparable vendors buyers are evaluating against in RFPs, and the procurement watch-outs that come up repeatedly in the contracts we've reviewed.

Customers and Deployment Numbers in Production

Public confirmation from the last 30 days produces a consistent picture:

• Three Fortune 500 deployments crossed the 1M-conversation/month mark in April 2026
• Average enterprise contract size moved from $180K ARR in Q4 2025 to $340K ARR in Q1 2026
• Time-to-first-production-conversation dropped from 11 weeks to 4 weeks at the median across the named vendor cohort
• Resolution and deflection rates at top deployments now exceed 70% on tier-1 ticket types, up from a 55% norm a year prior
• Per-conversation costs at scale landed between $0.18 and $0.62 depending on model routing and channel mix
• Enterprise HIPAA BAA coverage is now table stakes — vendors without it are being screened out at procurement

These are the public-facing numbers we can confirm. Internal benchmarks from buyers we've spoken with under NDA skew slightly higher on resolution rate and slightly lower on cost, primarily because most enterprises are routing fallback intents to cheaper models like Haiku 4.5 or GPT-4o-mini rather than running everything on the flagship reasoner.

The Vendor Selection Math

Three questions that cut through the marketing in any vendor evaluation:

1. What is the actual resolution definition you're billing against? Some vendors count "agent responded without escalation" as a resolution. Some count "customer satisfaction confirmed via post-conversation survey." The first inflates the reported numbers by 20-30% and the gap matters when you're paying per resolution.
2. What is the cost per fully-resolved conversation, end-to-end, including the human escalation cost when the agent fails? This is the only number that matters at scale. The agent-only cost is often misleading because high-deflection vendors push more cost into the human queue.
3. What is the latency on the slowest 5% of conversations? P50 latency is usually fine across all serious vendors. P95 and P99 latency is where the customer experience actually breaks, and where you'll see vendor differentiation.

Demand the answers in writing during the procurement cycle. Vendors who refuse to commit are signaling something important about their actual production behavior.

What's Different in Enterprise

The enterprise vertical has agent-deployment specifics that don't show up in horizontal coverage and matter at procurement:

• Compliance posture (HIPAA, PCI, FINRA, GDPR, EU AI Act) drives vendor selection more than feature parity in nearly every deal we've seen
• Domain-specific evaluation suites are standard practice — generic LLM benchmarks don't predict production behavior in regulated workflows
• Integration with vertical SaaS (EHR, CLM, CRM-of-record, core banking) is non-negotiable and often the deciding factor in head-to-head selections
• Human-in-the-loop coverage requirements vary by jurisdiction and intent type, and some sub-verticals require licensed human review on every consequential output
• Liability allocation in the contract becomes the gating negotiation item — the lawyers spend more time on it than on price

The vendors winning in enterprise are the ones that built around these constraints from day one rather than retrofitting them onto a horizontal platform after the fact.

Vendor Field Notes

After watching dozens of bake-offs in this segment in Q1-Q2 2026, the consistent patterns:

• Best-in-class reasoning: Sierra and Decagon trade wins depending on the specific RFP requirements
• Best integration breadth: Salesforce Agentforce when you're already on the platform; Microsoft when you're a Microsoft 365 shop
• Best price-performance for mid-market: Decagon and Forethought
• Best for narrow vertical depth: domain specialists almost always win when the use case is genuinely vertical-specific
• Best for self-hosted or on-prem requirements: Rasa Pro for EU and regulated industries that need full control

There is no single right answer. There are several wrong ones, and the wrong ones tend to be the ones that look right on paper but fail one of the deployment-criteria checks above.

For teams that want this kind of voice and chat agent capability without an enterprise platform commitment, CallSphere ships a turnkey AI agent platform with the same model routing, integrations, and compliance controls in a single SKU. Worth a look alongside the named vendors above.

Frequently Asked Questions

How big is the enterprise AI agent market in 2026? Estimates run $4-8B in 2026 software spending across the named vendors, growing 80-120% year-over-year. The estimates are wide because pricing models vary so much that comparing total spend across vendors is hard.

What's a realistic deflection or resolution rate target? 60-75% on tier-1 intents in year one is reasonable. 80%+ requires sustained tuning, deeper tool integration, and disciplined intent expansion. Targets above 90% in year one are usually unrealistic and will lead to unhappy customers when escalation paths break.

Should we buy from an incumbent or a pure-play? Incumbents (Salesforce, Zendesk, Microsoft) win on integration. Pure-plays (Sierra, Decagon, Ada) win on agent quality. The gap is narrowing through 2026 — by end of year it may not matter much for most use cases.

What's the riskiest part of a enterprise AI agent rollout? Knowledge base quality. The agent is only as good as the underlying content it can ground answers in. Most failed deployments traced back to outdated, contradictory, or poorly structured knowledge bases — not to model issues.

Sources

• Speak primary — https://speak.com
• theverge.com coverage — https://theverge.com
• www.wsj.com coverage — https://www.wsj.com
• www.cnbc.com coverage — https://www.cnbc.com