IVR Guide for Businesses

Interactive Voice Response (IVR) is an automated phone system that greets callers, offers menu options (e.g., Press 1 for Sales), collects input via keypress (DTMF, RFC 2833 out-of-band) or speech, and routes the caller to the right destination — information, queue or agent. For Indian businesses handling inbound at scale, IVR reduces manual transfers, supports multilingual (Hindi + English + regional), and can be on-premise for RBI-regulated workloads (EnableX) or cloud via providers like Exotel/MyOperator.

DTMF (touch-tone) is universal — works on feature phones, landlines, smartphones without data — critical for India. Speech (ASR/NLP) enables natural language ('I want to check my balance') but requires ASR/TTS and NLP, and may struggle with accents/noise. Many Indian IVRs use DTMF primary with speech as optional, supporting both RFC 2833 and in-band.

Inbound call → DID/Virtual Number → SIP gateway (e.g., Exotel UL VNO) → Media server (FreeSWITCH/Asterisk) → IVR logic (menu tree, TTS prompts) → CRM dip via webhook/API (Salesforce/Zoho) → Routing decision → ACD queue / agent / self-service. Edge: Mumbai PoP (ap-south-1) with TLS/SRTP, static IP whitelisting, G.711 PCMA/PCMU, SRTP for BFSI.

IVR collects intent (e.g., Press 2 for Support), optionally dips into CRM via webhook (caller ID → Salesforce record), then ACD picks the best agent. Example: Caller selects 'Billing' → webhook fetches account → if VIP, route to dedicated queue; else, skill-based ACD. EnableX pattern: every IVR event triggers webhook to your server for personalization.

Route by time: 9 AM–6 PM to live agents, after-hours to voicemail/callback or alternate location. MyOperator and Edesy examples include time-based routing. Offer callback: 'Press 1 to request a callback' reduces abandonment — queue callback / virtual hold is standard in Ozonetel.

Indian IVRs often start with language selection (Hindi, English, regional). MyOperator supports 10 Indian languages, Edesy 40+. Design: 'For Hindi, press 1; for English, press 2' as first menu, then subsequent menus in that language. Keep prompts under 30 seconds per Zingle.

Self-service: balance inquiry, order status via API dip without agent. Authentication where needed: OTP via SMS, or ask for account number + DTMF. For sensitive flows, consider on-premise deployment (EnableX for RBI/SAMA) and encrypted SIP.

Callback: offer at queue entry or after wait threshold. Queue escape: 'Press * to return to main menu' or 'Press 0 for agent'. Voicemail: if no agent, record message and create CRM ticket. Fallback: on invalid input, repeat twice, then route to agent.

Heuristic (not law): 2–3 levels, 4–5 options per level (Nesark: unlimited levels supported, but design should limit). Order by frequency (Sales first). Use verb + noun: 'For sales, press 1' vs jargon. Keep prompts concise, consistent voice/volume, and announce estimated wait.

Level 1: Welcome to Agilis (EN/Hindi) → Press 1 Sales, 2 Support, 3 Billing, 4 Regional Language, 0 Agent, * Repeat. Level 2 (Sales): Press 1 New Purchase, 2 Existing Order, 3 Callback, 0 Agent. Support: CRM dip → if VIP → priority queue else skill-based. After-hours: 'We are closed 6 PM–9 AM, press 1 to leave voicemail, 2 for callback.'

**Good (30 sec):** 'Welcome to Agilis. For Sales, press 1. For Support, press 2. For billing, press 3. To repeat, press *.' → immediate routing, callback offer. **Bad:** 45-sec marketing message before menu, 7 options, no human exit, inconsistent TTS voice, no repeat, deep 5-level tree — caller abandons.

Track: abandonment after IVR (target <5%), containment/deflection (% resolved without agent via self-service), transfer rate, time to resolution, repeat rate. MyOperator analytics: call traffic, missed calls, routing performance, menu usage. No universal law for '2–3 levels' — present as heuristic, not guarantee.

India requires Hindi + English + regional (Tamil, Telugu, Marathi, Gujarati, Bengali per Zingle). Use TTS for 40+ languages (Edesy) or pre-recorded prompts. Test from Jio/Airtel/BSNL/VI as Zingle suggests, as carrier audio differs.