PolyAI combines enterprise dialog agents, Agent Studio and a developer toolkit

PolyAI builds dialog agents for enterprise customer engagement, with voice at the center of its product history and current positioning. Agent Studio gives teams a place to build and operate agents, while the ADK provides a local development route. The buying decision is about a complete conversation service: its knowledge, integrations, spoken behavior and ongoing maintenance. PolyAI’s public per-minute commercial model supplies a useful starting point, but the rate and deployment scope still require a proposal.

01 / ProductFrom voice assistant to an agent-building environment

The PolyAI overview now describes an Agentic Dialog Platform for building, running and governing customer agents. The identity remains PolyAI at poly.ai; Agent Studio and the ADK are parts of the same company offer. The platform is relevant where the customer needs to complete a conversation-driven task, such as finding a booking or changing an account detail, rather than merely hear a recorded answer.

The Agent Studio page describes natural-language creation, testing and iteration alongside a route for developers. Its rapid-build messaging should be read as a vendor claim about authoring, not as a guarantee that an enterprise integration is ready in the same time. Connecting to reservation inventory, establishing permitted actions and checking data handling remain distinct implementation tasks.

The ADK documentation provides a more concrete view of the developer route. Teams can pull a project, edit it locally, validate it and push it back to Agent Studio for review and deployment. The documentation distinguishes self-serve account creation from enterprise accounts obtained through a PolyAI contact. A publicly available toolkit is therefore not evidence that every enterprise environment or production entitlement is immediately self-provisioned.

02 / AudienceTeams that need a maintained conversation service

PolyAI is a plausible fit for customer operations with substantial phone demand and well-defined business systems behind common requests. Hotels, service providers and retailers often have recurring questions mixed with transactions. The useful boundary is whether the agent can get the information and authority it needs to finish the specific job, while recognizing cases that need a person.

A team that wants to own its voice stack at a lower level should compare our Vapi blueprint. Our Retell AI blueprint offers another perspective on building and operating phone agents. These are contextual comparisons, not equivalent price bundles. The choice depends on whether you primarily need components and engineering flexibility or a broader enterprise dialog and service arrangement.

Before a demonstration, identify the people who would maintain the agent after launch. Service managers may own policy and conversational behavior; developers may own functions and integrations. The existence of two authoring routes is useful only if changes move through a review process both groups understand. Ask them to make one ordinary update together, using the same tools they would retain in production.

03 / WorkflowA proposed hotel reservation-change workflow

Consider an illustrative pilot for a hotel group handling reservation changes by phone. We have not deployed or tested this workflow in PolyAI. Start with one property, one reservation system and a narrow set of eligible changes. The agent should identify the reservation, understand the requested date change, retrieve live availability and explain any applicable price or policy difference before a customer commits.

The integrations page describes a Connect Portal with authentication, field mapping, health visibility and APIs. For this pilot, field mapping is not a cosmetic step. The reservation number, guest identity, property identifier and date format must have precise meanings. A connection that successfully returns data can still produce a wrong answer if the agent interprets a room-type code as an available room.

Keep quoted availability separate from a confirmed modification. Rooms can sell while the caller is deciding, and a displayed rate may have conditions that matter to the booking. The reservation system should validate the selected option at commitment time. The agent should then explain the confirmed result from that system rather than repeat an earlier offer as if it were still guaranteed.

Voice-specific tests should include a guest correcting a date, asking about two properties and changing the number of nights after hearing a quote. Do not judge these only by the transcript. Compare the final reservation state, any charge explanation and the confirmation delivered to the guest. A clear-sounding exchange can still contain a date or property mismatch that matters on arrival.

The ADK’s local edit and validation workflow creates a practical route for reviewing integration changes. In the proposed pilot, a developer can revise the reservation function while an operator reviews the conversational policy in Agent Studio. Require an explicit review before deployment, and test both the normal path and the case where the reservation service is unavailable. Local validation does not replace a realistic integration test.

Also rehearse a handoff in which the guest disputes the cancellation terms or requests an exception outside the agent’s authority. The human should receive the established booking context and the unresolved request. A graceful transfer protects the guest experience without turning the agent into an unofficial policy-maker. Track such transfers separately from failures caused by missing or incorrect integration data.

04 / PricingPer-minute pricing with the rate still to agree

PolyAI’s pricing page, consulted on 22 September 2026, states that ongoing voice-agent use is priced per minute and includes proactive performance improvements, maintenance and around-the-clock support. It does not show a universal monetary rate. The same page lists monitoring, upgrades and a phone-line uptime service-level commitment among plan features; contract definitions and exclusions still need confirmation.

A per-minute unit makes conversation design economically relevant. A longer call may be appropriate if it avoids an incorrect booking or a repeat contact. Conversely, excessive repetition can increase both caller frustration and usage. For the hotel pilot, compare completed eligible changes, accurate confirmations and repeat calls alongside total minutes. Do not optimize for shorter calls without checking whether the task actually finished.

Ask how setup, telephony, additional languages, new integrations and the chosen enterprise environment fit into the proposal. These are unresolved scope items, not claims that PolyAI adds separate charges for each one. Clarify how the public ongoing-service description relates to Agent Studio self-serve access and an enterprise agreement before using a trial experience to forecast production spend.

05 / DistinctionsTwo authoring routes can support shared ownership

The meaningful distinction in the current offer is the connection between a business-facing agent environment and a local developer workflow. An operator can explain the intended reservation rule while an engineer manages the integration that enforces it. That division can be productive when both changes are reviewed as one service release, with enough evidence to understand the customer-facing result.

PolyAI’s integration setup also makes the data contract visible. Managed authentication and monitoring may reduce routine connection work, but the buyer still needs to understand what a connector can read or change. In the hotel example, retrieving reservation details, offering room options and committing a new date should be treated as separate capabilities even if they share one provider connection.

The commercial inclusion of maintenance is another useful distinction. It creates a reason to ask how improvement actually happens: which recurring issues are reviewed, who approves behavior changes and how the team knows a correction helped. A service commitment is more valuable when it is tied to a process the customer can inspect than when it is summarized only as ongoing optimization.

06 / QuestionsWhat the public material cannot establish

The security page describes enterprise security and compliance arrangements. It is a starting point for examining the deployment’s data handling, not a substitute for checking the relevant environment, contractual commitments and access permissions. Reservation conversations can include sensitive account information, and the pilot should minimize what the agent collects and retains.

We did not test PolyAI’s speech quality, model performance or claimed deployment speed. Public pages also contain animated numerical displays that can extract incorrectly, so this guide does not turn those rendered counters into factual performance claims. The useful next evidence is a task-specific demonstration with realistic callers, actual system responses and a clearly identified production package.

07 / DecisionEvaluate a complete, maintainable dialog

PolyAI belongs on the shortlist when voice-service quality, enterprise integrations and ongoing improvement are central requirements. Begin with a narrow customer task and involve both its operational owner and its integration developer. Price the resulting minutes in the context of correct completed work, then assess whether the team can safely maintain the same service as policies and systems change.