Phone calls still matter in ecommerce. A customer may want to know where an order is. A subscriber may need help understanding a renewal. A shopper may abandon checkout because they are unsure about a product. A payment may fail even though the customer still wants to buy.
RevCent AI Voice Agents are built for those moments. They let ecommerce businesses accept inbound calls or place outbound calls with AI, while keeping the conversation connected to the RevCent context behind the store: customers, sales, payments, shipments, subscriptions, notes, events, and follow-up workflows.
For custom ecommerce stores and WooCommerce stores, this can create an instant inbound support layer without waiting to build a call center workflow from scratch. A store can route phone calls to an AI Voice Agent that understands the store’s RevCent context, answers common order and account questions, collects the right details, creates notes, and escalates when a customer needs a person.
That is the difference between a generic phone bot and a RevCent Voice Agent. The agent is not just answering from a script. It can be configured around the actual work happening inside the business.
How Voice Agents Work in RevCent
In RevCent, an AI Voice Agent is the configured phone automation. It defines how the agent should behave, when it should be available, what instructions it follows, what voice it uses, which tools it can access, and what limits protect the customer experience.
An AI Voice Call is the actual phone call created by that configuration. One Voice Agent can create many calls over time, and each call can be reviewed as part of the larger customer or operational history.
RevCent Voice Agents can be used in two major ways:
- Inbound: the agent answers calls from customers or prospects.
- Outbound: the agent calls someone after a RevCent event or on-demand trigger.
For inbound calls, RevCent can attempt to match the caller’s phone number to an existing customer. When that match is available, the agent instructions can adapt to the caller’s context. For outbound calls, the agent can be triggered from events such as a declined payment, a customer entering a segment, or another workflow deciding that a call should happen.
Before calls begin, the store connects the phone and AI pieces inside RevCent. In the current RevCent setup path, that means linking a Twilio voice integration for the phone number and an OpenAI integration for the realtime voice model, then choosing those integrations on the agent. After that, the agent lives inside RevCent with the rest of the store’s workflows, events, tools, and customer data.
The RevCent Settings That Matter
A good AI Voice Agent is more than a prompt. The setup around the prompt is what makes it safe and useful.
Call Method
The call method determines whether the agent is inbound or outbound. Inbound agents answer calls routed to the store’s connected phone number. Outbound agents place calls from a RevCent event, API request, or approved workflow.
Inbound agents are good for support coverage, shipment questions, subscription questions, and basic sales help. Outbound agents are good for payment recovery, post-purchase follow-up, abandoned checkout assistance, and proactive customer care.
Active Window
Active windows help control when an agent can answer or place calls. For inbound support, a store can choose what happens outside the active window, including forwarding to another number. For outbound calls, active windows help prevent awkward timing, such as a recovery call landing late at night.
Trigger Settings
Outbound agents need a reason to call. In RevCent, that can come from an event trigger or an on-demand trigger.
Event triggers are useful when the store wants to react to something that happened in the business. A declined sale, failed renewal, customer group change, or high-intent customer event can become the signal for a call.
On-demand triggers are useful when another workflow, API call, or approved AI tool should start the call directly.
Trigger settings can also include delays, filters, filter Functions, max runs per item, and minimum run intervals. These details matter because not every event should become an immediate call. A customer may fix a failed checkout on their own. A prospect may need time before follow-up. A customer may trigger the same event more than once.
For example, a declined-payment agent can wait a few minutes, confirm the sale is still pending, and limit the run count so the same failed sale does not create repeated calls. A prospect follow-up agent can wait longer, check that the customer is still in a no-sale group, and start with one campaign before expanding.
Agent Settings
Agent settings control the experience of the call. This includes the agent voice, optional pre-agent Function, and the agent instructions.
A pre-agent Function can prepare extra context before the call begins. For example, it might return a product list, campaign name, approved offer, customer segment, or other store-specific details that should appear in the instructions. This is useful when the agent should know more than the base event provides, but the store still wants that context calculated inside RevCent before the call starts.
Agent Instructions
Agent instructions are written in Markdown. They can include clear sections such as role, tone, context, tools, rules, conversation flow, and escalation logic.
RevCent instructions can also use dynamic content with Handlebars. For example, an instruction can greet a verified customer by first name, reference item context from the event that triggered the call, include a pre-agent Function response, or insert a reusable snippet.
Handlebars Compilation
Handlebars lets a RevCent Voice Agent start with one reusable instruction template and compile it into a call-specific set of instructions before the conversation begins. The template can pull in safe context from the agent, trigger item, customer record, pre-agent Function, and reusable snippets.
This is useful because the same agent may need to behave slightly differently from call to call. An inbound support agent can know whether the caller is already matched to a customer. A declined-payment agent can include the current sale status. A prospecting agent can include only the products or offers approved for that campaign.
The best use of Handlebars is practical and controlled: include facts the agent should know, hide empty values behind conditionals, and use snippets for shared policy language. The compiled instructions should still read like clear human-authored guidance, not a data dump.
Snippets
AI Voice Snippets help stores reuse instruction content across agents. This is useful for shared rules such as identity verification, refund boundaries, card-handling policy, transfer language, brand tone, or closing language.
Instead of rewriting the same policy in every agent, the business can keep reusable content consistent.
Call Limits
Call limits protect cost and customer trust. A store may limit calls per day, calls per customer, max call duration, active windows, or how often the same item can trigger a call.
These limits are especially important for outbound agents. A helpful recovery call can become annoying if the customer receives too many calls or receives them too quickly.
Built-In Call Handling
RevCent Voice Agents also include built-in handling for common phone situations, such as inactive callers, voicemail, robocalls, transfers, and ending calls.
That matters because real calls are messy. People pause, machines answer, customers ask for a person, and some calls should end quickly. The agent should be prepared for those paths.
How Agents Use RevCent Tools
RevCent AI Voice Agents can be given access to approved system actions. Think of these as the tools the agent is allowed to use during the call.
Those actions are opt-in. The business enables the exact actions an agent needs, one by one, so a support agent, recovery agent, and sales agent can each have different capabilities.
Tool access is what turns a voice conversation into a useful ecommerce workflow. A support agent can look up a customer. A recovery agent can review a failed sale. A sales agent can inspect eligible product context. A follow-up agent can save the outcome of the call.
Examples of RevCent tool-enabled workflows include:
- Searching for a customer when the inbound caller is not automatically verified.
- Looking up a sale or shipment before answering a support question.
- Reviewing a pending sale after a declined payment.
- Sending an approved follow-up email.
- Adding notes or metadata so the team can report on outcomes.
- Triggering a custom Function when the store needs business-specific logic.
- Transferring the call when the issue requires a person.
The important rule is scope. A Voice Agent should only have the system actions required for its job. A shipment support agent does not need payment-processing tools. A payment recovery agent does not need refund authority unless the business has a very clear reason and strict instructions.
For high-impact actions such as refunds, card updates, sale creation, payment processing, cancellations, and subscription changes, the store should use narrow permissions, explicit instructions, testing, and escalation rules.
Common RevCent Tool Sets
The right tool set depends on the job the agent is doing. These examples are starting points, not universal permissions:
- Inbound support: customer search, customer lookup, sale lookup, shipment lookup, note creation, and transfer.
- Declined payment recovery: customer lookup, sale lookup, checkout follow-up, outcome metadata, and pending-sale processing only when the workflow explicitly allows it.
- Prospect or sales assistance: customer lookup, campaign context, estimate sale, approved follow-up, note creation, and sale creation only after clear customer consent.
- Subscription support: customer lookup, subscription context, billing or renewal status, note creation, and escalation for cancellations or exceptions.
Inbound Voice Agents
Inbound agents answer calls from customers or prospects. They are a good fit when the store receives repeat questions that need fast, consistent answers.
Order and Shipment Questions
Shipment questions are often the easiest place to start. Customers call because they want confidence: Was the order received? Has it shipped? Is the package delayed? Can the tracking link be sent again?
An inbound RevCent Voice Agent can greet the caller, use customer context when available, look up the relevant order or shipment, explain the status, send an approved follow-up, and create a note.
If the issue is sensitive, confusing, or outside policy, the agent can transfer or escalate.
Subscription and Account Support
Subscription ecommerce businesses can use inbound agents for renewal questions, billing timing, active subscription status, overdue accounts, and account support.
The agent does not need to be a retention machine. The best version is helpful and clear. It explains what is active, what happened, what the customer can do next, and when a human teammate should step in.
Product and Pre-Sale Questions
Some stores sell products that benefit from a conversation. A customer may ask about sizing, bundles, subscriptions, compatibility, shipping timing, or product differences.
An inbound sales-help agent can answer approved product questions and guide the customer toward the right next step without requiring every call to reach a human.
Outbound Voice Agents
Outbound agents place calls after a signal. In RevCent, that signal can come from an event, a customer group, or another workflow.
The best outbound agents are restrained. They call for a specific reason, respect timing, avoid over-contacting the customer, and make it easy for the customer to decline.
Declined Payment Recovery
A declined payment does not always mean the customer is gone. The card may have failed, the checkout may have timed out, or the customer may need help finishing the order.
In RevCent, an outbound recovery agent can be triggered after a failed or declined sale event. The business can add a short delay so the customer has time to fix the issue on their own, use filters to confirm the sale still needs attention, and limit calls so the customer is not contacted repeatedly.
The agent can then call with context, explain that the order did not complete, and guide the customer toward the approved payment update or checkout path.
In a RevCent setup, this might use a declined-sale event, a pending-payment filter, and only the system actions needed to review the customer, inspect the sale, send a safe follow-up, or process a pending sale when the workflow explicitly allows it.
Abandoned Checkout Follow-Up
Some abandoned checkouts deserve more than an email. If a shopper is high-intent, high-value, or tied to a specific campaign, a Voice Agent can call to ask whether they need help.
RevCent can support this with event triggers, customer groups, delays, and filters. A pre-agent Function can prepare campaign-specific context or a product list so the agent knows what it can discuss.
This is where RevCent’s operating context matters. The agent can be tied to the same customer groups, campaign data, and sale status the store already uses instead of calling every shopper the same way.
Outbound Sales Assistance
For some stores, outbound Voice Agents can help prospects who entered the customer journey but have not purchased. This should be used carefully.
A good sales-assistance agent should be limited by customer eligibility, campaign scope, call frequency, and clear product guidance. If offers are allowed, they should be defined in the instructions or prepared by a pre-agent Function, not invented during the call.
For example, a no-sale customer group can become the signal, a delay can give the shopper time to finish checkout on their own, and a pre-agent Function can return the approved products or offers for that campaign.
Proactive Customer Care
Outbound agents can also support customer experience. A store might call after a delayed shipment, a first subscription order, a VIP purchase, or a support event.
These calls are less about immediate revenue and more about confidence. The agent can check whether the customer has questions, provide context, save notes, and route issues to a person when needed.
Example Instructions
AI Voice Agent instructions are where the business defines the role, tone, goal, boundaries, tools, and escalation rules.
The examples below use labeled Markdown sections so the agent has a clear role, context, tool policy, conversation flow, and safety path. They also show where RevCent dynamic content can be used without turning the post into a technical setup guide.
Inbound Customer Service Agent
These instructions frame the agent as a general support entry point. They focus on caller verification, safe use of RevCent customer and order context, clear answers for common support questions, call notes, and escalation when a human should review the issue.
# Role & Objective
You are the inbound customer service voice agent for our ecommerce store.
Your objective is to help callers with order status, shipment questions,
payment questions, and simple account support. A successful call gives the
customer a clear answer or routes them to the right human teammate.
# Personality & Tone
## Personality
- Friendly, calm, and practical.
- Confident without sounding robotic.
## Tone
- Warm, concise, and clear.
- Do not repeat the same phrase every call.
- Use plain English.
# Context
- RevCent may provide a verified customer record when the caller's phone number
matches an existing customer.
- Customer context:
{{#if customer.verified}}
Verified caller: {{customer.first_name}} {{customer.last_name}}
Email: {{customer.email}}
{{else}}
No verified customer is attached yet.
{{/if}}
- RevCent may provide customer, sale, payment, shipment, subscription, and note
context.
- Use RevCent context to avoid making the caller repeat information the store
already has.
- Shared support policy:
{{{snippets.support_policy}}}
# Tools
## Search customers
Use when: the caller is not automatically verified and provides an email or
phone number.
Do not use when: the caller only asks a general store question.
## Get customer
Use when: you need customer details before answering account-specific questions.
## Get sale or shipment
Use when: the caller asks about an order, delivery, tracking, or payment tied
to a sale.
## Create note
Use when: the support team should see the call outcome.
## Transfer call
Use when: the issue requires human review.
# Instructions / Rules
- If the caller is verified, you may greet them by first name:
{{#if customer.verified}}Hi {{customer.first_name}}, thanks for calling.{{/if}}
- If the caller is not verified, ask for enough information to find and verify
the account before sharing order-specific details.
- Do not promise delivery dates that are not present in RevCent.
- Do not approve refunds, replacements, cancellations, or policy exceptions.
- Before using a tool, say a short natural phrase such as "Let me check that."
# Conversation Flow
## 1) Greeting
Goal: Welcome the caller and learn why they called.
How to respond:
- Identify as customer support.
- Ask what they need help with.
Exit when: The caller explains the issue.
## 2) Customer Details
Goal: Find or confirm the customer record.
How to respond:
- If verified, continue with the available customer context.
- If not verified, search by email or phone and confirm the result before
sharing account details.
Exit when: The customer is verified or verification cannot be completed.
## 3) Diagnose
Goal: Determine the type of issue.
How to respond:
- For order questions, review the relevant sale.
- For shipment questions, review the relevant shipment.
- For payment questions, review safe payment context and escalate if needed.
Exit when: The correct path is clear.
## 4) Resolve
Goal: Give the best store-approved answer.
How to respond:
- Explain the status in simple language.
- Send approved follow-up information when available.
- Create a note if the team should know what happened.
Exit when: The customer understands the answer or needs escalation.
## 5) Close
Goal: End cleanly.
How to respond:
- Summarize the answer or next step.
- Ask whether the caller needs anything else.
- Thank the caller and end the call politely.
# Safety & Escalation
Escalate to a human when:
- The customer is upset or threatens a chargeback.
- The customer asks for a refund, cancellation, replacement, or exception.
- The package appears lost or damaged.
- The account or order context is unclear.Outbound Declined Payment Agent
These instructions are built for a narrow recovery workflow. The agent explains that a recent order did not complete, confirms whether the customer still wants help, uses customer and sale context, sends approved follow-up, and avoids sensitive payment handling unless the workflow explicitly allows it.
# Role & Objective
You are the outbound declined payment voice agent for our ecommerce store.
Your objective is to help customers complete an order after a recent declined
payment or failed checkout. A successful call either helps the customer return
to the approved payment path or records that they are not interested.
# Personality & Tone
## Personality
- Respectful, steady, and service-oriented.
- Helpful without being pushy.
## Tone
- Brief and clear.
- Never blame the customer for the decline.
- Do not sound like collections.
# Context
- RevCent may provide the customer, pending sale, payment attempt, campaign, and
decline context that triggered this call.
- Triggered sale:
{{#if item.id}}
Sale ID: {{item.id}}
Current status: {{item.status}}
{{else}}
No sale context was attached to this call.
{{/if}}
- Customer:
{{#if customer.first_name}}{{customer.first_name}}{{else}}the customer{{/if}}
- The customer may have already completed checkout after the original decline.
- Treat the call as checkout assistance.
- Recovery policy:
{{{snippets.payment_recovery_policy}}}
# Tools
## Get customer
Use when: you need customer context for the call.
## Get sale
Use when: you need to review the pending or failed sale.
## Send checkout follow-up email
Use when: the customer wants to finish later or wants the approved link.
## Process pending sale
Use only when: this workflow is explicitly enabled for the agent and the
customer has completed all approved payment steps.
## Create note or metadata
Use when: recording whether the call helped, was declined, or needs human
follow-up.
# Instructions / Rules
- If RevCent provides verified customer details, use the customer's first name
naturally:
{{#if customer.verified}}Hi {{customer.first_name}}, this is the store calling about your recent order.{{/if}}
- Explain only that the recent order did not complete.
- Do not ask the customer to say full card details aloud unless this agent has
been explicitly approved for that workflow.
- Do not create discounts, waive fees, or promise exceptions.
- If the customer says they were charged, stop recovery and escalate.
- Before using a tool, briefly explain what you are checking.
# Conversation Flow
## 1) Opening
Goal: Explain why you are calling.
How to respond:
- Identify the store.
- Say the recent order did not complete.
- Ask whether the customer would like help finishing it.
Exit when: The customer accepts help, declines help, or asks for details.
## 2) Check Intent
Goal: Confirm whether the customer still wants the order.
How to respond:
- If they want help, review the related sale context.
- If they already completed checkout, thank them and close.
- If they are not interested, thank them and do not push.
Exit when: The customer's intent is clear.
## 3) Help
Goal: Guide the customer to the safest next step.
How to respond:
- Send the approved checkout follow-up if requested.
- Guide the customer to the approved payment update path.
- Only process a pending sale when the workflow explicitly allows it.
Exit when: The next step is complete, sent, or declined.
## 4) Close
Goal: Record the outcome.
How to respond:
- Summarize what happened.
- Save the appropriate note or metadata.
- End politely.
# Safety & Escalation
Escalate to a human when:
- The customer says they were charged.
- The customer asks for a refund, discount, or exception.
- The customer is upset or confused.
- The payment context suggests fraud risk.
- The customer asks for something outside the approved recovery workflow.Outbound Prospect or Sales Agent
These instructions are for eligible prospects who showed buying intent but have not purchased yet. The agent stays helpful and restrained, uses only approved product or offer context, answers practical questions, and creates a follow-up or sale only when the customer clearly wants that next step.
# Role & Objective
You are the outbound sales assistance voice agent for our ecommerce store.
Your objective is to help eligible prospects who showed interest but have not
completed a purchase. A successful call answers questions, explains approved
product options, or records that the customer is not interested.
# Personality & Tone
## Personality
- Helpful, knowledgeable, and respectful.
- Sales-aware but not aggressive.
## Tone
- Conversational and brief.
- Do not pressure the prospect.
- Respect a clear no.
# Context
- RevCent may provide customer, campaign, customer group, and product context.
- Campaign context:
{{#if item.campaign_name}}
Campaign: {{item.campaign_name}}
{{else}}
No campaign name was provided.
{{/if}}
- A pre-agent Function may provide approved product options or offer details:
{{{toString pre_agent_function.response.product_list}}}
- Use only the products, offers, and claims provided in RevCent context or the
pre-agent Function response.
- Sales boundaries:
{{{snippets.sales_boundaries}}}
# Tools
## Get customer
Use when: you need customer or campaign context.
## Estimate sale
Use when: the prospect asks about total cost and the workflow allows estimates.
## Create sale
Use only when: the prospect clearly agrees to purchase and this agent is
approved to create sales.
## Send follow-up email
Use when: the prospect wants product details or checkout information in writing.
## Create note
Use when: recording interest, objections, or follow-up requests.
# Instructions / Rules
- Call only eligible prospects from the configured RevCent trigger or workflow.
- Do not invent products, discounts, bundles, shipping promises, or guarantees.
- If a discount is allowed, use only the discount provided in context.
- Do not continue persuading after the prospect clearly declines.
- Before using a tool, briefly explain what you are doing.
# Conversation Flow
## 1) Opening
Goal: Explain why you are calling.
How to respond:
- Identify the store.
- Say you are following up because they showed interest.
- Ask whether they have a quick question or need help choosing.
Exit when: The prospect engages, declines, or asks for details.
## 2) Understand Need
Goal: Learn what is blocking the purchase.
How to respond:
- Ask one short question.
- Listen for product uncertainty, shipping concerns, price concerns, or checkout
friction.
Exit when: You understand the main concern.
## 3) Recommend
Goal: Provide approved product guidance.
How to respond:
- Use only provided product context.
- Offer to send details if the prospect wants to review later.
- If the prospect wants to buy and the workflow allows it, proceed through the
approved sale path.
Exit when: The prospect has a next step.
## 4) Close
Goal: End respectfully.
How to respond:
- Thank the prospect.
- Save a note with the outcome.
- Do not call again unless future RevCent rules make them eligible.
# Safety & Escalation
Escalate to a human when:
- The prospect asks for a custom deal.
- The prospect has a complaint or support issue.
- The requested product or offer is not in the provided context.
- The agent is not approved to complete the requested action.What Makes a Good RevCent Voice Agent
A good RevCent Voice Agent is not trying to do everything. It has one job, the right settings, the right tools, and clear boundaries.
For most ecommerce stores, the strongest agents include:
- A specific call method: inbound or outbound.
- A specific role, such as order support, payment recovery, subscription help, prospect follow-up, or post-purchase care.
- Trigger settings that fit the workflow.
- A short list of required system actions.
- Call limits that protect cost and customer experience.
- Active windows or forwarding behavior when needed.
- Dynamic instructions for verified customers or event-specific context.
- Escalation paths for refunds, disputes, chargebacks, fraud risk, or upset customers.
- Boundaries around discounts, promises, card handling, cancellations, and policy exceptions.
The safest rollout is narrow. Start with one repeatable workflow, review the calls, tune the instructions, adjust the tools, and then expand.
What to Measure
AI Voice Agents should be judged by business outcomes and customer experience, not by call volume alone.
For revenue workflows, the store can look at recovered orders, completed checkouts, subscription saves, and follow-up conversion. For support workflows, the store can look at resolution rate, escalation rate, call duration, repeat contacts, and customer sentiment.
Because RevCent keeps voice calls connected to customers, sales, events, notes, and outcomes, teams can also use their AI chat of choice to generate reporting on Voice Agent conversions. With RevCent context available through approved tools or MCP, an operator can ask plain-language questions instead of manually stitching together exports.
Example questions a store could ask an AI assistant include:
- Which outbound Voice Agent calls led to completed orders this week?
- What was the conversion rate for declined-payment calls compared with email follow-up?
- Which campaigns produced the most recovered revenue after a Voice Agent call?
- How many inbound support calls were resolved without a human transfer?
- Which Voice Agent outcomes are most often followed by refunds, cancellations, or repeat purchases?
- Which call scripts or instruction versions are tied to the best recovery rate?
- Show me customers who received a Voice Agent call, did not convert, and may need human follow-up.
The Bigger Opportunity
The real promise of RevCent AI Voice Agents is not replacing every phone interaction. It is making the repeatable, time-sensitive, commerce-connected conversations easier to handle.
Inbound agents can make support more responsive. Outbound agents can recover revenue and follow up at the right moment. System actions let the agent do useful work during the call. Dynamic instructions, snippets, pre-agent Functions, and call limits help keep the experience specific and controlled.
The best RevCent AI Voice Agent feels like a helpful store representative with the context, tools, and boundaries needed to solve one specific problem well.
For stores already using RevCent as an operating layer, AI Voice Agents add a practical voice channel to the same customer, payment, subscription, shipping, event, and reporting context that already powers the business.