Evaluating voice AI platforms for an Indian deployment is different from evaluating SaaS software. The language dimension, the compliance requirements, and the infrastructure realities of India create a set of questions that many global platforms simply can't answer well.
Here are the 10 questions to ask every platform before signing anything.
1. Where are your servers located?
This is the most important question for Indian deployments. If the answer is "US East" or "EU West," you have a DPDP Act compliance problem and a latency problem. Voice data of Indian citizens must be processed and stored in India. Ask for the specific data center region and AZ.
Good answer: "India (Mumbai / Hyderabad), ISO 27001 certified, no data leaves Indian jurisdiction."
2. Is your per-minute pricing all-in, or does it exclude telephony, LLM, or STT?
Many platforms advertise a low per-minute rate and then bill separately for the language model, text-to-speech, and the phone call itself. Ask for the total cost per minute for a 3-minute Hindi outbound callvoice, LLM, TTS, and telephony included.
3. Do you handle Hinglish natively?
Not "do you support Hindi?"Hinglish. Ask them to demo a call where the customer switches mid-sentence between Hindi and English. If the system struggles with code-switching, it will fail on 40% of your real calls.
4. Which regional Indian languages do you actually support well?
Most platforms claim "30+ Indian languages"but many of these are low-quality, synthesis-only support without genuine STT accuracy in the language. Ask for STT accuracy benchmarks for each language you need. Ask them to make a test call in Telugu, Kannada, or Tamil and evaluate honestly.
5. How do you handle DPDP consent capture?
Every call must open with a consent disclosure and capture a consent signal. Ask how this is implemented, whether it's in the customer's language, whether it's recorded, and where the consent log is stored.
6. Is RBI Fair Practice Code compliance automated?
For BFSI use cases: are call windows enforced at the infrastructure level (can't be overridden) or at the campaign level (can be misconfigured)? How are calls logged for RBI audit? Where are recordings stored?
7. What is your actual latency for India calls?
Ask for the p50 and p95 TTFB for Hindi outbound calls routed through Indian mobile networks. If they don't have this data, that itself is informative. Target: p50 < 300ms, p95 < 500ms.
8. How long does deployment take, and who owns it?
Some platforms are infrastructure-onlyyou build everything. Others are managed deployments where the platform team handles configuration. For non-technical buyers, this is critical: do you need a developer to go live?
9. What does your NDNC scrubbing and TRAI compliance process look like?
Any outbound campaign must scrub against the NDNC registry and use TRAI-registered telephony infrastructure. Ask specifically how this is handledand whether you or the platform holds responsibility for compliance.
10. Can I see real outcome data from Indian deploymentsnot demos?
Demos are optimized environments. Ask for case studies with specific metrics from Indian deployments in your industry. Connection rate, completion rate, outcome rate (payment commitment, site visit booked, etc.). If the platform can't show you Indian deployment data, they don't have meaningful Indian deployments.
Agni's answers to all 10: India servers (Mumbai), all-in pricing, native Hinglish, 30+ languages with real STT benchmarks, automated DPDP consent, RBI enforcement at infrastructure level, p50 240ms latency, no-code deployment in 7 days, TRAI-registered telephony, and real Indian case studies with verifiable metrics.
