Zendesk drops per-seat pricing as NiCE rebuilds Cognigy and AI agents start calling customer service
Zendesk used its AI Summit on 10 Nov to change how it charges for customer service software. Zendesk Unlimited removes per-seat fees for its new Outcome AI Platform plans and replaces them with an outcome budget that is drawn down each time an issue is resolved or a piece of business work is completed. In the same announcement Zendesk launched a public directory that helps personal AI assistants find a company's verified support channels. Both moves assume that more of the work in help desk software will be done by AI agents, and that some of the customers knocking on the door will be AI agents too.
Zendesk Unlimited swaps seats for an outcome budget
The Zendesk product post lays out how the plans work. Each Outcome AI Platform plan has a flat platform fee (the entry Essentials plan has a minimum spend instead), no charge per seat, AI features that vary by tier, and an outcome budget held in the customer's contract currency. Optional add-ons such as Sandbox and high volume API access are sold separately.
Zendesk counts two kinds of outcomes. A conversational outcome is a customer issue resolved through a Zendesk conversation, whether AI resolved it, a human agent resolved it with AI help, or both did it together, and one issue counts once even if it spans several messages or channels. Conversational outcomes start at $0.99. An operational outcome is a measurable AI task done outside a conversation, such as running a standalone custom agent workflow or producing an analysis in Agentic Analytics. Zendesk says simply opening a feature or chatting with AI does not trigger a charge, and reporting traces each charge to an individual outcome.
Will Poon, a Zendesk director of product management, told CX Network that "a successful outcome means the customer's issue is resolved, not simply that a ticket is closed or an AI interaction ends," and that customers can see why a charge qualified and raise a case for review. According to the Zendesk help center, the plans are in an Early Access Program for now, and customers sign up through Zendesk sales.
The rest of the platform has three parts. Zendesk Bridge lets a business connect the AI interface of its choice to Zendesk's AI, service context, and workflows, with identity, permissions, human approvals, and audit applied to every action. Specialized AI Agents bring industry and custom logic into service workflows, and Self-Improving Zendesk ties AI agents, human work, knowledge, and quality data together so teams can find root causes. Zendesk says customers on the platform can automate up to 84 percent of service requests.
Personal AI agents are already contacting brands
The second Zendesk launch is aimed at a new kind of caller. The Open Service Directory is an open specification and public directory where businesses register their official support channels and AI agents, so that a consumer's personal AI assistant can find the right place to ask for help. Businesses control what they share and when a human must review a request. Zendesk said the directory does not touch private customer data or take service actions, that it will support agent communication protocols such as PACT and PAP alongside it, and that it has no plans to charge for it. Betterment, ElevenLabs, Flex, TeamSystem, TBA Auctions, Telekom Payment S.A., and Wireless Logic are early partners.
ASAPP has early data on how much of this traffic exists. Its Consumer Agent Index, published on 4 Nov, found about 2,300 customer service chats over 60 days at two major airlines and two telecom providers where the "customer" was an AI agent acting for a person. Weekly volume rose from about 40 to about 400 chats in September and has stayed there. Fewer than three in ten of those agents said they were AI, and most called themselves an "assistant." At one telecom provider, 56 percent of the agents were asking for a lower bill or a retention offer, 4.3 times the rate of comparable human customers. Their median opening message was 366 characters long, against 35 for a typical customer. ASAPP plans to publish a new edition every month.
NiCE is preparing for the same traffic from the other side. In a SiliconANGLE column on the new Cognigy, analyst Zeus Kerravala reported that NiCE is co-developing a Personal Agent Protocol with Meta and Sierra. The protocol is not live yet. Clement Boutier of NiCE described Cognigy as the "front door" of the contact center, where a consumer's personal agent would have to authenticate before it can go further.
NiCE ships the next generation of Cognigy
NiCE announced the next generation of Cognigy on 12 Nov, a little more than a week after the third quarter results covered in the 9 Nov issue, in which NiCE put its AI annual recurring revenue at $392 million. Cognigy gets a redesigned Agentic Experience Studio with default speech-to-text, text-to-speech, and language models already provisioned, an AI Assist helper that can build, debug, and simulate agents, and reusable skills that bundle instructions with tools. Agents can coordinate sub-agents and connect to outside agents and systems through MCP and A2A, and native speech-to-speech models let an agent keep talking while it calls a tool.
The pricing change mirrors Zendesk's in one respect. Default models are now included in the per-conversation price, so customers no longer need a separate model contract. Premium models from Deepgram, ElevenLabs, and Anthropic are available with usage-based billing, and customers can also bring their own model contracts.
The bigger theme is control after launch. New tracing shows every turn of a conversation, including which skill loaded, which tool was called with what arguments, token use, and time to first token. The simulator can compare up to 10 variants of an agent using synthetic customers. Boutier, the NiCE director of product management who leads the release, told SiliconANGLE that a hybrid of natural language understanding flows and agentic reasoning "is actually where most of our customers are," and that fully agentic deployments are at the "beginning of the adoption curve." NiCE says existing production deployments keep running unchanged and customers can adopt the new features when they are ready.
Retell builds its own model for live calls
Retell AI launched Tell-1 on 12 Nov, a language model trained on hundreds of millions of calls and built for the moments when voice agents tend to fail: a caller gives half an answer, changes the subject, or needs the agent to look something up while they wait. The model reasons while it speaks. Retell says Tell-1 responds in under 100 milliseconds on its live benchmark and is built to stay under 300 milliseconds at peak volume, and that it scored highest overall on that benchmark against the latest models from OpenAI, Google, and Anthropic.
Retell runs the benchmark, and to its credit its own guide to purpose-built models tells buyers to treat it as a reason to shortlist and then test on their own calls. The same guide lists Tell-1 at 3.5 cents per minute for the model alone, ranked fifth on tool calling, and available in the same model selector as OpenAI, Anthropic, and Google models. Retell also launched Retell Dialer, which runs outbound campaigns across AI voice and SMS agents, schedules callbacks at the contact's local time, and hands qualified prospects to a specialist agent. Buyers comparing speech AI vendors now have one more company offering its own model rather than renting a frontier one.
Talkdesk and Cresta stretch AI across the whole customer journey
Two launches on 12 Nov treated a single call as one step in a longer relationship. Talkdesk announced Talkdesk Voyager, a long-horizon orchestration product built on its Customer Experience Automation platform. Voyager ties each customer journey to a goal, such as preventive care, repeat purchasing, or loan conversion, and coordinates AI agents, employees, and enterprise systems toward it over days, months, or years. It keeps long-term memory, so it can recall a shopper's fit preferences months after an exchange or wait for an appointment opening with a preferred doctor. Pre-built blueprints cover healthcare, retail, and financial services, including post-discharge follow-up aimed at reducing avoidable 30-day readmissions.
Cresta, which Salesforce agreed to buy on 2 Nov, launched Journey Intelligence the same day. It links calls, chats, and other touchpoints into one customer history. Journey Insights adds a per-customer journey view, callback analysis that shows why customers keep calling about the same issue, and a predicted next contact for each customer, which Cresta says is a first in the market. AI agents get memory of the full journey, and human agents get a Customer Snapshot before they pick up. Cresta's example is a home security customer who has called several times about false alarms: the AI agent can skip the apology and go straight to adjusting sensor sensitivity, the fix those customers usually call back for.
A day earlier Cresta launched in Japan with a Tokyo office and a Japanese interface and voice. Jim Tamura, who previously led ElevenLabs' business in Japan and Korea, is the country manager. Cresta named three Japanese relationships: retailer Shop Japan is deploying it, the BPO BELLSYSTEM24 will offer its conversation intelligence to Japanese contact centers, and Kotoba Technologies will supply low-latency Japanese speech recognition and synthesis. Cresta tied the timing to labor shortages and to a new Japanese law that requires employers to protect staff from customer harassment.
Zingtree pairs AI agents with fixed rules
Zingtree launched a new platform on 10 Nov for contact centers in healthcare, insurance, financial services, and other regulated industries. Required steps such as identity verification, eligibility checks, and policy thresholds run as deterministic flows outside the AI model, so they happen every time and use no AI tokens. Guardrails check each action before an agent takes it, for example blocking a refund above a policy limit, and every answer and action is logged for audit. The logic runs across self-service, a copilot for human agents, and back-office orchestration, and it can be called through MCP and API. Todd Sale, Corpay's senior vice president of customer experience operations, said that in payments, "even one incorrect answer or unapproved action is a real problem." The release fits Zingtree's place in the scripting category, and it lands on the same theme as the Cognigy launch: most buyers want AI reasoning where it helps and fixed rules where the law or money is involved.
The Dialpad and Ada deal, a week later
CX Today returned on 11 Nov to the Dialpad and Ada combination that led the 9 Nov issue. Dialpad CEO Craig Walker argued that software categories "describe how vendors organize their products, not how customers experience a business." Kerravala, quoted again, said the fit makes sense on paper but called the deal a possible sign of an "AI reset" for standalone conversational AI vendors. The companies did not disclose terms. The Globe and Mail reported an all-share deal valuing Ada at about C$600 million, or about US$430 million, against the US$1.2 billion valuation of its 2021 funding round. BetaKit reported that Walker said the combined company will have more than US$470 million in annual recurring revenue. The companies had said the deal would close on or about 12 Nov.
Also this week
Five9 published its Care-Completion Gap Report on 10 Nov, based on a Dynata survey of 1,000 US consumers. Forty-four percent said they had delayed care, given up on scheduling or a billing issue, or switched providers in the past year because dealing with a provider was too hard. Thirty-eight percent are comfortable using AI to schedule an appointment, but only 12 percent for medical advice, and the top factor that would raise trust in AI was being able to reach a human easily. Only 26 percent of Gen Z said their issue was fully handled on first contact, compared with 74 percent of Baby Boomers.
SoundHound reported third quarter revenue of $62.4 million on 5 Nov, up 48.5 percent, its first quarter since closing the purchase of LivePerson on 4 Sep. GAAP net loss was $46.4 million. SoundHound said it has contacted LivePerson's top 90 enterprise accounts, which represent more than $100 million in ARR, identified $120 million in annual cost synergies, and expects to be adjusted EBITDA positive in the second half of 2027.
3CX released V20 Update 10 on 11 Nov. It adds MCP support so tools such as ChatGPT or Claude can query the call data a user is authorized to see, lets the AI receptionist and customer service agent run on xAI Grok as well as OpenAI, and sends calls to a backup destination if the AI provider is down. A new 3CX AI Server runs transcription, summaries, and sentiment on a customer's own Apple Silicon Mac or Linux server, and 3CX says a Mac mini handles about 268,000 call minutes a month.
Quiq put its Voice AI agent in the Zoom App Marketplace on 16 Nov, so Zoom Contact Center customers can add it without custom integration work and hand calls to human agents with full context in the Zoom desktop app. Quiq said some of its own customers want to move to Zoom Contact Center and can now keep the AI agents they already run.
New in the CXSpy catalog
CXSpy added five company pages this week, bringing the catalog to 347. Two French CCaaS vendors join: Diabolocom, a Paris contact center platform founded in 2005 that runs its own private cloud and is a licensed telecom operator, and Ringover, a cloud phone and call center system for sales, recruiting, and support teams with more than 100 integrations. Intradiem, from Marietta, Georgia, joins workforce management with real-time automation that moves work, schedules training, and offers time off as contact center volume shifts. Two voice AI companies join call center AI: Desible.ai, a Bengaluru company building voice and messaging agents for banks, lenders, and insurers, and Vocca, a New York company whose AI answers patient calls and books appointments for medical practices and hospitals.
Previous issue: Salesforce buys Cresta, Five9 buys Level AI, and Dialpad and Ada combine in a week of CX AI deals.
Sources
- Zendesk: Outcome AI Platform and unlimited seats
- Zendesk: Introducing Zendesk Unlimited
- Zendesk help: Announcing Zendesk Unlimited (EAP)
- CX Network: Zendesk launches public directory for A2A service
- ASAPP Consumer Agent Index, Volume 1
- Unite.AI: NiCE upgrades Cognigy for governed enterprise AI agents
- SiliconANGLE: NiCE's next-generation Cognigy
- Retell: Tell-1 launch
- Retell: Purpose-built vs general LLMs in contact centers
- Talkdesk: Talkdesk Voyager
- Cresta: Journey Intelligence
- Cresta: launch in Japan
- Zingtree: agentic governance and orchestration platform
- CX Today: Dialpad-Ada deal signals a conversational AI reset
- The Globe and Mail: Ada to merge with Dialpad
- BetaKit: Ada to merge with Dialpad
- Five9: Care-Completion Gap Report
- SoundHound Q3 2026 results
- 3CX: V20 Update 10
- Quiq Voice AI in the Zoom App Marketplace