Home Blog AI voice agents for customer service in India: inbound vs outbound
Workplate Journal / Voice guide

AI voice agents for customer service in India: inbound vs outbound

Choose a first AI calling workflow with an inbound versus outbound comparison, a worked example and a pilot checklist.

An AI voice agent handles a phone conversation using a defined objective, business information and rules about what to do next. Start with inbound calls if the problem is unanswered enquiries. Start with scheduled outbound callbacks if people already request a call and your team struggles to make it on time. Choose the direction that matches the problem you can measure.

Inbound vs outbound AI calling

Choose a first workflow by who starts the call
Call directionGood first job and success measure
Inbound: the customer calls youAnswer opening-hours or service enquiries, capture the reason for the call and route exceptions. Check whether the caller got the right answer or handoff.
Outbound: your agent calls the customerReturn a requested callback or follow up on an agreed appointment. Check whether the right person was reached and whether the next action was recorded.

Choose one job first, even if you eventually need both. If most calls require judgement, use the agent for intake and route the decision to a person. For an Indian audience, test actual caller phrasing, including Hindi and English where relevant. A language option alone doesn’t tell you how reliably the agent will hear a name, date or address.

Choose the call direction from the work you already do

Which customer service calls should you automate first?
What your team seesWhere to start
Missed calls and repeated questions about hours, locations or servicesStart inbound. Use a small set of approved answers and capture the enquiry when the agent cannot resolve it.
Website visitors request callbacks, but the team cannot reach them at the agreed timeStart with scheduled outbound callbacks. Confirm the reason for the call and make it easy to agree another time.
Most calls involve complaints, exceptions or decisions that need judgementKeep a person responsible for the decision. Use the agent to collect context and route the request.

Use recent call notes to identify repeatable questions. Choose one workflow where you can write the correct response, specify when to hand over, and tell from the call record whether the job was done.

Worked example: intake at a small consultancy

A fictional consultancy gets calls about its services while the team is with clients. It starts with inbound intake. The agent answers from an approved service sheet, asks what the caller needs, and confirms a name and preferred callback time. If the caller asks for a custom quote or raises a complaint, the agent routes the request according to the firm’s handover rules. The firm reviews whether the call record gives the next person enough context to follow up.

Sample call: a customer switches between Hindi and English

This fictional exchange shows what to test with the languages you configure. The agent can explain the next step and capture a request, but the adviser owns the quote.

Caller: Namaste, mujhe accounting service ke baare mein poochna hai. Can someone send me a quote?

Agent: Namaste. I can note what you need and ask an adviser to call you. Kya aapko monthly bookkeeping chahiye?

Caller: Yes, monthly bookkeeping. Kal dopahar teen baje call kar sakte hain?

Agent: Ji, I’ll note tomorrow at 3 p.m. as your preferred time. An adviser will confirm. May I have your name?

Check that the call record captures monthly bookkeeping, quote requested, preferred callback at 3 p.m. tomorrow, awaiting confirmation. In a live workflow, read back the actual calendar date as well as the time. If the caller corrects the name or date, verify that the final record reflects the correction.

A copyable pilot brief

First call type:
What the agent should resolve:
Approved answers and sources:
Languages and caller phrases to test:
Details to confirm back to the caller:
When to hand over and to whom:
What to do if nobody is available:
What a successful call looks like:
Who reviews exceptions and updates the script:

Test a routine enquiry, a noisy call, an unfamiliar question, an incorrect date, and a request for a person. Review the outcomes before expanding the script. When reporting results, keep completed actions, unanswered calls and human handoffs separate.

Review the pilot with a simple scorecard

Score a sample of calls against the job you assigned
CheckWhat to count and review
Correct answersCalls answered correctly out of reviewed calls that asked a question covered by your approved material. Log any incorrect or unsupported answer.
Usable handoffsHandoffs with the caller’s request, a clear owner and a next action out of reviewed calls that needed a person.
Failed or incomplete callsTrack dropped calls, repeated misunderstandings and missed callbacks separately. For outbound calls, record no-answer attempts separately from conversations that failed after connection.

For example, if eight of ten reviewed calls that needed a person had a usable handoff, look at what was missing from the other two before expanding the pilot. A completed conversation is only useful if the customer got the answer or the next person can act on the record.

AI calling in India: what to check before going live

Check two things separately: whether you can place the call under telecom rules, and how you handle the caller’s personal data. Permission to return an enquiry does not automatically cover promotional calls or a new use of a recording or transcript.

TRAI rules: classify outbound calls by purpose

  • Inbound support: the customer starts the call. Identify your business, tell the caller they are speaking with an automated assistant, and explain any recording or transcription before collecting information. Assess any later outbound call separately.
  • Requested callback: keep a verifiable record of when and how the customer requested it and what they asked about. Keep the call to that purpose. For commercial communication relying on a customer enquiry, TRAI’s September 2026 amendment summary describes a seven-day window and requires the enquiry to be made in writing or digitally and kept in verifiable form. A callback request does not replace the checks your telecom provider requires.
  • Promotional outreach: review the consent and customer preference (DND) rules before calling. Keep sales pitches out of service scripts; under the TCCCPR amendment, mixing promotional content into a commercial call makes it promotional. Don’t use a support callback as permission to add the caller to a campaign.

TRAI’s 2026 amendment requires entities using application-to-person (A2P) calls to declare their use and the calling numbers to their telecom service provider in advance. Before enabling automated outbound calls, confirm the call category, declaration, applicable numbering series, sender or DLT registration, and preference or consent checks with your provider. The setup can differ by sector and call type.

Check where promotional contacts came from

A phone number in a CRM, spreadsheet or public directory does not show permission to send promotions. TRAI describes consent as permission given to a sender for a specific purpose, product or service. Where you rely on consent, verify that it covers your business, the promotional purpose and the calling channel.

  • Contacts you collected directly: keep the wording shown to the person, their affirmative choice, the date, the source and any withdrawal. A number supplied for support, a quote or a callback does not establish promotional permission.
  • Contacts supplied by a third party: ask for the original notice and verifiable evidence that the person agreed to the relevant sharing and promotional calls from your business. Permission to hear from the supplier, or a generic assurance that the list is “opted in”, is not proof that your business is covered.
  • A vendor calling on your behalf: distinguish that role from a third party using the list for its own marketing. Confirm who the registered sender is, what data the vendor may use and how it will honour withdrawals.

Confirm the applicable consent registration and customer preference checks with your telecom provider, including how it verifies or registers consent collected outside its platform. Leave contacts out if you cannot establish that their source and the proposed use are permitted.

DPDP: plan for recordings, transcripts and CRM notes

A phone number, recording, transcript or CRM note can be personal data if it identifies a person or can be linked to one. India’s Digital Personal Data Protection Act and Rules are being phased in. The official commencement notification schedules most core processing duties and individual rights for 18 months after publication in November 2025. Check the official Rules and enforcement timeline for what applies when you launch, alongside any current privacy and sector rules.

  • Explain the purpose. Say what you collect and why. If you rely on consent, make the choice clear, keep a record and provide a way to withdraw it. A recording announcement alone does not establish valid consent.
  • Capture only what you need. A callback may need a name, number and reason for calling. Keep passwords, OTPs and full card details out of ordinary support scripts.
  • Follow the data. Map who can access audio, transcripts and summaries across the voice provider, CRM and exports. Set retention and deletion rules, and agree how providers will help with security incidents and privacy requests.
  • Test the exceptions. Decide what happens if a caller declines recording, asks for a person, withdraws consent or asks you to stop calling. Check that the right person receives the request and that any applicable suppression carries through to the next campaign.

Before the pilot: have your telecom provider confirm the outbound setup, and have your privacy or legal owner review the script, data flow and opt-out handling. This is general guidance; the requirements depend on your workflow and sector.

Where Workplate Voice fits

Workplate Voice lets you configure inbound or outbound agents with business knowledge, capture fields and handover rules. Start with one call type and review the records before you broaden the workflow.

Start with one call workflow.

Explore how Workplate Voice handles enquiries, scheduled callbacks and follow-ups.

Explore Voice