Problem
India's 5,000+ BPO and in-house contact centers handle hundreds of millions of calls annually, yet quality auditors manually review fewer than 5% of calls — meaning compliance breaches, poor CX, and script deviations go undetected at scale. Manual QA costs ₹25,000–40,000 per QA agent per month and still misses the vast majority of interactions. No affordable, India-first solution handles Indian English accents and regional language code-switching out of the box.
Solution
A SaaS platform that ingests call recordings (via S3/SFTP/API), transcribes them using a fine-tuned ASR model optimised for Indian English and Hindi-English code-switch, then runs LLM-based scoring against configurable rubrics (compliance checklist, sentiment arc, script adherence, forbidden phrases). The dashboard surfaces agent leaderboards, flagged calls, and weekly coaching reports — replacing spreadsheet-based QA workflows. V1 integrates with Exotel, Ozonetel, and Knowlarity (the dominant Indian CCaaS platforms) for automatic call ingestion.
Why Now
Cresta — an AI conversation-intelligence platform backed by a16z and Sequoia — hit $100M ARR in April 2026, validating that enterprises pay for AI call QA at scale. In India, Bolna's $6.3M raise in January 2026 from General Catalyst signals that voice AI infrastructure for India-specific accents and languages has matured enough to build on commercially. Meanwhile, rising DPDP (Digital Personal Data Protection) Act compliance requirements in 2026 are forcing Indian BPOs to audit call recordings for data-handling disclosures — a regulatory pull that did not exist two years ago.
Target User
First 1,000 customers are mid-sized Indian BPOs (100–1,000 seats) in Bengaluru, Hyderabad, Pune, and NCR serving BFSI, telecom, and healthcare verticals. Operations managers and QA heads earning ₹15–35 LPA who currently manage teams of 5–10 manual auditors and are under pressure from clients demanding SLA-level QA transparency. Purchase trigger: client audit or BPO contract renewal requiring documented QA coverage above 50%.
Business Model
Monthly SaaS subscription priced per seat: ₹800/agent/month up to 500 agents, ₹600 above 500. A 200-seat BPO pays ₹1.6L/month. One-time onboarding fee of ₹50,000 covers custom rubric setup and CCaaS integration. Gross margin is ~82% (primary COGS: ASR + LLM API credits at ~₹0.40/minute of audio). CAC via direct BD and BPO industry events (CII, NASSCOM) estimated ₹1.5–2L; LTV at 24-month average contract = ₹38L, giving LTV:CAC above 15:1.
Competitive Landscape
- Direct (India): Enthu.ai, Observe.AI India presence — both target large enterprises (2,000+ seats), leaving the mid-market underserved
- Direct (global reference): Cresta (US, $100M ARR), Balto (US) — pricing and complexity inaccessible to India mid-market
- Why we win: India-first accent and language support (Hindi-English, Tamil-English code-switch) built from day one; pricing 60–70% below global players; CCaaS integrations with Exotel and Ozonetel not available in global products
6-Month Plan
- Month 1 (₹1.5L): Fine-tune Whisper on Indian call-center audio corpus; build ingestion pipeline; hire 1 BPO domain expert as co-founder
- Month 2 (₹2L): Build LLM scoring engine with configurable rubrics; Exotel integration; internal QA dashboard v0
- Month 3 (₹2L): Pilot with 2 seed BPOs at no cost; iterate rubric templates for BFSI and telecom; ship agent leaderboard UI
- Month 4 (₹1.5L): Launch paid tier; Ozonetel + Knowlarity integrations; DPDP compliance report module
- Month 5 (₹1.5L): BD push at NASSCOM BPO Summit; target 5 paid customers
- Month 6 (₹1.5L): Coaching recommendation engine (auto-generated feedback per agent); target 10 paying BPOs
- Total capex: ₹10L | Remainder held as 2-month runway buffer
Risks
- Accent and language ASR quality (high likelihood × high impact): Indian English accent variance is wide; if transcription accuracy falls below 85% on regional accents, QA scores become unreliable. Mitigation: build accent-tagged fine-tuning corpus from day one using publicly licensed call datasets.
- Long enterprise sales cycles (high likelihood × medium impact): BPOs with 200+ seats require procurement approval and security review, extending sales to 90+ days. Mitigation: price onboarding at ₹50K so mid-level Ops managers can approve without CFO sign-off.
- Data privacy pushback (medium likelihood × high impact): Some clients may resist uploading call recordings to a third-party SaaS under DPDP rules. Mitigation: offer on-prem or VPC deployment at a higher tier from Month 5.
Score Breakdown
- Market (16/20): India's contact center workforce exceeds 1.3M agents; TAM for mid-market AI QA SaaS at ₹800/agent/month on 10% penetration equals ₹125Cr/year, comfortably above the ₹100Cr threshold.
- Capital (11/15): MVP requires ASR fine-tuning, an LLM scoring pipeline, and a React dashboard — buildable for ₹8–10L; just above the ₹5L ideal but well within the ₹20L ceiling.
- Team (7/10): Needs 2 engineers plus one BPO domain expert; sourcing a co-founder with BFSI or telecom call-center operations background is the primary hiring risk, so not a full 10.
- Trend (13/15): Two direct trend signals within 90 days — Cresta's $100M ARR (April 30, 2026) and Bolna's $6.3M raise (January 20, 2026) — provide strong global and India-specific validation.
- Moat (12/15): Each BPO's call history and custom rubric creates high switching cost; the language-tuned ASR model improves with every client's audio data, forming a genuine data flywheel. Not 15 because incumbents could eventually catch up on regional language support.
- Economics (12/15): 82% gross margin; LTV:CAC exceeds 15:1 at steady state; pricing is affordable enough for mid-market without deep discounting.
- Speed (7/10): Realistic MVP in 8–9 weeks given the ASR fine-tuning pipeline; the model training loop adds time beyond standard SaaS development, preventing a faster rating.