What sits inside a voice AI platform
Every AI phone call passes through the same layers. Some platforms give you all of them ready to use; others give you an API and expect you to plug in your own telephony, models and logic. Knowing the layers makes it much easier to compare vendors honestly.
| Layer | What it does | What to check | In Outpero |
|---|---|---|---|
| Telephony | Places and receives calls over the phone network and carries audio | Indian numbers, DLT registration, call quality | DLT-registered line; dedicated number ₹649/month |
| Speech recognition (ASR) | Turns the caller's speech into text | Indian accents, code-switching, noisy lines | Seven Indian languages with English switching |
| Turn-taking | Detects when the caller has finished or interrupts | Barge-in, no talking over people | Part of the 0.8 to 1.3 s caller-stops-to-agent-speaks loop |
| Language model and logic | Decides what to say and which action to take | Stays on brief, admits what it does not know | Configured by describing the job |
| Voice (TTS) | Speaks the reply | Natural Indian voices, speed of first audio | Swara voice engine, two voice tiers |
| Actions | Does things mid-call | Calendar, sheets, APIs, messaging | Calendar booking, spreadsheet, custom API, WhatsApp |
| Campaigns and triggers | Decides who to call and when | Instant triggers, lists, retries, windows | 30-second lead calling; CSV campaigns with retries and DND screening |
| Analytics and data | Stores and exports what happened | Transcripts, summaries, webhooks, export | Recording, transcript, summary, sentiment, custom fields, webhook, CSV |
Developer platforms versus managed platforms
Developer-first voice AI platforms such as Vapi, Retell AI and Bland AI give engineering teams building blocks and fine control. They suit companies with developers who want to choose their own models, write custom logic and run their own telephony. Pricing is usually in US dollars, and the total bill depends on the components you plug in.
Managed platforms aim at the business user. Outpero is one: you describe the job, pick a language and voice, connect a lead source or upload a list, and calls start. You give up some low-level control in exchange for not having to build or maintain the stack. If your team has strong engineers and an unusual use case, a developer platform may genuinely be the better choice.
The hidden cost of the developer route is everything around the model: buying Indian numbers, handling DLT and DND, tuning speech recognition for Hinglish or Tanglish, building retries and calling windows, storing recordings, and wiring results into a CRM. None of it is impossible, but it is ongoing engineering work. A managed platform has already made those choices, which is faster when the goal is simply to get every lead called this month rather than to own a voice stack.
Why India changes the platform question
Three things make Indian voice AI different. First, language: callers mix English into Hindi, Telugu or Tamil constantly, and a platform built for English-only calls tends to mishear them. Second, telephony and regulation: commercial calls fall under TRAI's TCCCP 2018 regulations, with DLT registration and DND screening for promotional calls. Third, price sensitivity: a voice AI minute has to compete with a telecaller costing roughly ₹15,000–₹25,000 a month.
Hear it on a real call before you read further?
How Outpero's platform runs a call
- 01A trigger arrives — a new Meta or Google Ads lead, a website form, a CRM update, or the next row in a campaign list.
- 02The platform dials from your line within 30 seconds for instant leads, or inside your calling window for campaigns.
- 03Swara holds the conversation in the chosen language, at 0.8 to 1.3 seconds on average end-to-end latency, and switches to English with the caller.
- 04Actions you enabled run mid-call: a calendar booking, a spreadsheet lookup, an API call or a WhatsApp message.
- 05The recording, transcript, summary, sentiment and custom fields are saved and pushed to your webhook.
Concurrency is sold as channels: each is one call at a time, the number includes 1 and each extra is ₹649/month, for instant, bulk and inbound calls. The platform carries a 99.9% uptime SLA.
Pricing models you will see
- —Per connected minute, all-in: one rate that covers telephony, speech, model and voice. Outpero works this way at ₹5 or ₹6.8 per minute by voice tier.
- —Per minute plus components: a platform fee plus separate charges for the model, voice and telephony you choose — common on developer platforms.
- —Seat or licence pricing: a monthly fee per agent or per channel, sometimes with a setup charge and minimum term.
- —Per call or per dial: charged even when nobody answers, which punishes low pickup rates.
Outpero has no setup fee, no monthly base fee and no lock-in. Credits are prepaid at ₹1 each, never expire, and the first 50 are free at signup.
A checklist for choosing a voice AI platform
- —Call the demo agent in your customers' language and answer in mixed English.
- —Time the pause after you stop talking; compare vendors with the same stopwatch.
- —Interrupt mid-sentence — the agent should stop and listen.
- —Ask how DLT and DND are handled and whose line the calls go out on.
- —Check what reaches your CRM or webhook and in what shape.
- —Work out a monthly bill from your real call volume, including the number and any minimums.
- —Ask what is not supported — languages, live transfer to a person, integrations — and get it in writing.
Limits of Outpero as a platform
Outpero can transfer a call to your team when a person should take it; otherwise it captures the request and flags it in the summary and webhook for a callback. It covers seven languages, not every Indian language. And because it is managed rather than a raw API, teams that want to swap individual models or host the stack themselves will find a developer platform, or building their own, more flexible.