Dialpad AI Agent uses conversation credits, but its billing rules are based on what the agent does during a session—not simply the number of minutes on a phone call. That distinction matters when you compare AI Agent with per-minute voice products or estimate how many credits a pilot will consume.

Reviewed 7 October 2026. Dialpad does not publish a universal AI Agent credit price on its public pricing page. Confirm your account’s rate card, included credits and recharge settings with Dialpad before building a budget.

How Dialpad describes AI Agent pricing

Dialpad’s public AI Agent pricing page describes conversation-based pricing: customers buy a pool of AI Agent Credits, and credits are used when the AI delivers value. Dialpad says a conversation is billable when the AI retrieves information or executes a real action, such as scheduling, routing or an order lookup. The page invites buyers to speak with sales rather than listing a public per-credit rate.

This is different from assuming every call minute is charged as AI Agent usage. Still, “only when value is delivered” needs to be interpreted through Dialpad’s detailed session criteria. The Dialpad Help Center credit guide provides those criteria for voice and digital sessions.

When a voice AI Agent session is billable

Dialpad’s Help Center says a voice session is billable when all of these conditions are met:

  1. There is at least one interaction between the customer and the AI Agent, including at least one customer response.
  2. The AI Agent looks up information or performs a task at least once.
  3. The session is closed—because inactivity times out, the customer hangs up, the AI closes it, or a live agent ends it after a transfer.

A session can still count after a transfer to a human if a skill was called and the other requirements are met. So “the agent handed off” does not automatically mean “no credit used.” Conversely, a call with no customer response, no lookup or task, or no closed session does not meet the listed voice criteria. Check Dialpad’s current documentation because platform behavior and contract rates can change.

Digital sessions follow a similar rule

For digital AI Agent sessions, the documented criteria also require an interaction with a customer, an information lookup or task, and a closed session. Dialpad defines an interaction as at least one response to the AI Agent. A transfer to a live agent can remain billable where the required skill was called and the session meets the other criteria.

The practical implication: a useful test plan should include both voice and digital examples if you plan to use both channels. Test a quick hang-up, a FAQ answer with a knowledge lookup, a scheduling action, a transfer, and an abandoned digital thread. Then compare the observed outcomes with the usage report and credit balance.

Usage alerts and credit controls

Dialpad says AI Agent usage alerts are emailed at 80% and 90% of credit usage, followed by a 100% balance alert that indicates whether auto-recharge ran or agents were paused. The alert contains links to Plan Management for changing recharge settings. Dialpad also says AI Agent credits are used by the primary office, which centralizes usage for the company.

Before launch, decide who receives these alerts, whether auto-recharge is enabled, what amount it adds, and what should happen when the balance reaches zero. A credit pool can make the cost harder to estimate than a simple per-minute price if your call mix or task completion rate changes, so set a monthly budget and review actual usage during the pilot.

What to request in a Dialpad AI Agent quote

  • Price per credit or session and how many credits each type of action consumes.
  • Included or committed credit volume, expiry or rollover rules, and overage rate.
  • Whether voice, digital, transfers, tool calls and AI Agent bundles use the same pool.
  • What qualifies as a “skill” call, lookup or completed task for billing.
  • How usage reports separate billable AI Agent work from ordinary telephony charges.
  • Whether auto-recharge is enabled by default and how agents behave at zero balance.

For a fair comparison, model a month from your own call mix: number of calls, share with a customer response, share that need a lookup or action, transfers, and digital sessions. Ask Dialpad to map those cases to expected credits in writing. Sysevo’s voice agents, call analysis and CRM workflows provide examples of the kinds of actions and records to include in a vendor pilot checklist.

Questions buyers ask

Does Dialpad AI Agent charge by the minute?

Dialpad describes AI Agent as conversation-credit based, with credits consumed when the agent retrieves information or performs an action. Regular phone carrier charges may be separate, so review both the AI Agent rate card and calling rates.

Is an AI Agent call billable if it transfers to a person?

It can be. Dialpad says a voice session remains billable after a transfer if a skill was called and the other session criteria are met.

Does Dialpad publish the AI Agent credit price?

Dialpad’s public pricing page explains the credit model but directs buyers to sales for AI Agent pricing. Request the rate card and usage assumptions for your channels and plan.

What happens when AI Agent credits run out?

Dialpad documents a 100% usage alert that tells you whether auto-recharge ran or agents were paused. Confirm the setting and fallback behavior on your account before routing live calls.

Estimate Dialpad credits from a call mix

Dialpad’s public documentation describes the conditions that make an AI Agent session billable, but it does not give every buyer a universal dollar cost per conversation. Ask for the applicable AI Agent rate card and whether your plan uses a dedicated AI Agent bundle before Dialpad Credits. The current help article says AI Agent usage uses the AI Agent bundle first and then Dialpad Credits if needed; verify this for your account, especially if you also use calling, messaging or older AI products.

To build a useful forecast, classify calls by whether the customer responds, whether the agent performs a lookup or task, and whether the session closes. A voice session that meets all three documented conditions can consume a credit even if it transfers to a live employee. A conversation that never gets a customer response or never uses a skill may not meet the listed criteria. Do not equate a phone call, a minute or a credit without confirming the plan’s unit and the account’s rate.

Take a representative sample and label each session: FAQ with knowledge lookup, booking action, routing, transfer after a skill, caller hang-up before action, no-response call, and digital conversation. Apply the written billing criteria, then compare your classification with Dialpad’s actual usage report. If you cannot explain why a session consumed a credit, ask the administrator or customer success contact to map the usage entry to the documentation.

Set controls before the balance is depleted

Dialpad documents email warnings at 80% and 90% of AI Agent credit usage and a final 100% alert indicating whether auto-recharge ran or agents were paused. Each alert includes links to plan management. Confirm who receives those messages, whether a shared finance or operations mailbox should be added, and what happens if the primary administrator is away. A warning nobody monitors is not a spend control.

Decide what should happen at the limit. Auto-recharge can avoid an abrupt interruption, but it can also continue spend unless the recharge amount and authority are clear. Pausing agents can protect a budget but leave callers without the intended service. Define the fallback phone path, who can approve a recharge, how quickly an administrator can act and what customers hear while the account is paused.

Dialpad says AI Agent credits are used by the primary office. For a multi-office organization, check whether that central pool matches your internal cost allocation. If each region owns a budget, agree on a reporting method that separates usage by agent, channel, task or office. Do not assume that one shared balance automatically gives finance a useful business-unit breakdown.

Ask for a written credit model

Request a clear example using your expected workload. The quote should show the price per credit or session, how a voice or digital action consumes credits, included volumes, expiry or rollover, additional-credit pricing, recharge rules and whether transfers are billable when a skill has been called. Ask how costs change when a session uses multiple skills and whether the rate differs by channel or package.

Also ask how ordinary telephony and AI Agent charges appear on an invoice. A voice call can create phone usage and AI Agent usage as separate charges. A transfer may stop the AI interaction while the telephone call continues. Digital channels may use a distinct service allocation. Match each quote line to the corresponding usage report so the organization can reconcile actual cost after rollout.

Use a pilot to learn the billable rate for your work

Run a controlled pilot with a known set of test calls. Include common cases and edge cases, record the number of sessions in each category, then compare the credit balance before and after. This does not replace the written rate card, but it helps estimate which interaction patterns drive consumption. Run the pilot for enough volume to include typical variation rather than drawing conclusions from a handful of calls.

Alongside credits, measure correct outcomes, transfer completion, false confirmations, caller repeat effort and employee cleanup. Divide AI Agent and telephony cost by verified tasks. If a task requires a person to redo the lookup or booking, count it as incomplete for the pilot scorecard. This encourages the team to improve useful completion rather than suppressing credits by making the agent less capable.

Review usage weekly during the pilot. If one skill consumes far more credits than expected, inspect whether it is being called repeatedly, whether a prompt needs clearer inputs, or whether the business workflow should route that request to staff. Change one variable at a time and keep the call sample consistent.

Before expanding, retain the current Dialpad credit documentation, the quote and the alert configuration in your budget record. Pricing and plan access may differ by account; recheck the service allocation and zero-balance behavior before increasing live call volume.

Give each credit-consuming action a business owner

List every skill or task the Dialpad AI Agent can call and name the team responsible for the connected data. A lookup that uses an outdated product catalog, calendar or CRM field can consume credits while still producing a bad customer outcome. Assign an owner to the source data and a review cadence for each skill.

For higher-risk actions, require a clear confirmation step and a reliable system response before the agent tells the caller the action is complete. Review a sample of transcripts and corresponding records. If the AI and system of record disagree, treat the event as a failed action and resolve the integration before expanding volume.

Reconcile credit usage with outcomes

At month end, compare the credit report with the number of completed bookings, lookups, transfers and digital tasks. Identify high-credit workflows with low completion, then decide whether the issue is repeated skill calls, an unclear prompt, a poor source record or the wrong task for automation. This keeps optimization focused on useful completion instead of reducing activity without understanding the cause.

Share the review between the AI owner, operations and finance. Record the credit balance, alerts, recharge, telephony charges and verified outcomes in one summary. This gives the company a repeatable basis for budget changes and makes it easier to spot when a new call flow changes the pattern of billable sessions.

Related Sysevo guides

Structuring Internal Accountability Before Credits Are Spent

When an AI Agent handles calls across voice and digital channels simultaneously, the credit pool belongs to the organisation but the decisions driving consumption belong to specific teams. Before any session reaches a caller, it is worth mapping which department owns which interaction type, so that when a usage review reveals unexpected consumption, there is a named person who can investigate rather than a shared inbox that defers indefinitely.

Consider separating ownership along three lines: the business workflow that triggers a skill, the data source the skill queries, and the customer experience standard that defines a successful outcome. A booking skill, for example, may be triggered by a sales process, rely on a calendar managed by operations, and be evaluated against a customer service standard. Each of those three owners can take a distinct action when something goes wrong, and none of them can reasonably act in isolation. Without that separation, a failed lookup produces a credit charge and a poor caller experience, but no one is clearly positioned to fix the root cause.

Designing a Fictional Enquiry to Test Ownership in Practice

To illustrate how accountability might work in a real evaluation, consider a fictional scenario. A caller contacts an AI Agent to reschedule an appointment. The agent greets the caller, receives a response, queries an availability system, confirms a new time and closes the session. Under the documented voice billing criteria, that session would meet the conditions for a billable credit: there was a customer interaction, an information lookup, and a closed session. The credit is consumed regardless of whether the downstream calendar was actually updated or whether the caller received a confirmation they could rely on.

In this scenario, three questions arise that a business evaluation should answer before go-live. First, whose responsibility is it to verify that the calendar system accepted the change? Second, if the caller calls back because the appointment was not updated, does the second session also consume a credit? Third, who reviews the transcript to catch that pattern before it recurs? Assigning those questions to named roles in advance turns a theoretical accountability model into a practical operating procedure.

A transfer adds a further layer. If the AI Agent calls a scheduling skill and then routes the caller to a human agent who completes the booking, the session may still be billable under the documented criteria. The human agent may not know a credit was consumed, and the caller may not know two interactions occurred. The business needs a record that connects the AI session, the transfer and the final outcome, so that cost and completion can be assessed together rather than in separate reports.

Comparison Criteria Worth Including in a Vendor Evaluation

When assessing AI Agent pricing against alternative approaches, the criteria that matter most are not always the ones that appear first in a sales conversation. The following considerations are worth building into any structured comparison, with reasons they are relevant to credit-based billing specifically.

  • Unit of charge versus unit of value: A credit consumed on a session that produces no verified outcome is a cost without a benefit. Evaluation criteria should include how the vendor's billing unit maps to your definition of a completed task, not just its definition of a delivered interaction.
  • Transfer billing behaviour: Where a session remains billable after a transfer to a human, the cost model differs from one where only fully automated completions are charged. This affects how you design escalation paths and whether you route borderline cases to staff earlier or later.
  • Reporting granularity: A shared credit pool is only useful for budget management if the usage report can be filtered by skill, channel, agent or time period. Before committing, request a sample report and confirm it contains the fields your finance team needs for allocation.
  • Zero-balance behaviour: Whether agents pause or continue when credits are exhausted has a direct effect on caller experience and on cost control. These are different risks, and the business should choose a fallback deliberately rather than accepting a default.
  • Rollover and expiry: A credit pool that expires at period end creates an incentive to consume rather than conserve. Evaluating rollover terms alongside the credit price gives a more accurate picture of the total cost across a contract period.

Exception Handling and Staff Responses Worth Defining Early

Certain interaction types sit at the boundary of what an AI Agent should handle without human oversight. These include situations where a caller disputes a previous AI-confirmed action, where the system of record returns an error during a skill call, or where a caller provides information that appears to contradict a known record. In each case, the agent may still close the session and consume a credit, but the appropriate staff response differs from a standard completed interaction.

For a disputed AI confirmation, the staff member handling the callback or complaint needs to know whether the original session transcript is accessible, what the system of record shows, and who has authority to resolve the discrepancy. Defining that escalation path before calls go live avoids a situation where an employee promises a resolution they cannot deliver or overrides a system record without authorisation.

For a skill that returns an error, the relevant question is whether the agent communicated the failure clearly to the caller or whether it proceeded as if the action had succeeded. Reviewing a sample of error-state transcripts during the pilot is a practical way to identify that distinction. If the agent confirms an action that did not complete, the business has both a credit charge and a broken customer expectation to resolve.

Practical Acceptance Exercise Before Expanding Live Volume

Before moving from a pilot to full deployment, a structured acceptance exercise can give operations and finance a shared basis for the decision. The exercise does not need to be technically complex. Run a defined set of session types, record the observable outcomes and compare them against the credit report.

Observable evidence worth collecting includes: the transcript of each test session, the corresponding record in the connected system, the credit balance before and after, the alert emails triggered if thresholds were reached, and the time between session close and system update. Where those elements are consistent and the skill performed as intended, the session passes. Where they diverge, the team has a concrete example to investigate before adding call volume.

Acceptance criteria should be agreed in advance by the AI owner, operations and finance rather than defined retrospectively once results are in. If a criterion for passing is that the system record updates within a defined window of the session closing, that window should be specified before the exercise begins, not chosen after the results show what the system actually did.

The completed acceptance record, including passing and failing sessions, the credit consumption observed and any configuration changes made during the exercise, forms the baseline against which future monthly reviews can identify drift. A deployment that performs well at launch but gradually shifts its pattern of billable sessions without a corresponding change in completed tasks is easier to catch when there is a documented starting point to compare against.