Product

QA Tools That Don't Understand Hinglish

19 October 2026

A short call where the customer and agent switch between Hindi and English in the same sentences

“Mera recharge ho gaya, but data abhi tak nahi aaya.”

“Sir, main abhi check karti hoon. Can you confirm your number?”

Nobody on this call thinks they’re speaking two languages. It’s just how people talk. Across India, customers and agents move between Hindi and English mid-sentence, often without noticing.

For a person listening, this is easy to follow. For many QA tools, it isn’t.

Where generic tools break

Most call analysis tools are built for one language at a time. When a call mixes two, things go wrong in predictable ways.

Words get dropped or guessed. A tool expecting English hears Hindi words and replaces them with the nearest English-sounding word, or leaves them out.

Meaning gets lost. “Kal bhi yahi bola tha” — “you said the same thing yesterday” — is the most important line on the call. If it’s mistranscribed, the frustration behind it disappears.

Scoring goes wrong. If the greeting was “Namaste, main Priya bol rahi hoon”, a tool listening only for English may decide there was no greeting at all.

Calls get skipped. Some teams simply leave mixed-language calls out of automated review because the results can’t be trusted. That’s often most of their calls.

What it costs

  • Reviews that are wrong. Agents get marked down for steps they actually did, in Hindi.
  • Agents who stop trusting QA. If the tool keeps missing what they said, they stop taking the scores seriously.
  • Blind spots. The calls left out of review are often the ones with your most local, most frustrated customers.
  • Manual QA comes back. Teams end up reviewing mixed-language calls by hand, at the same small sample rate as before.

What good looks like

A QA tool built for Indian contact centres needs to handle the way your customers and agents actually speak:

  • Hindi, English and Hinglish on the same call, without anyone choosing the language first.
  • Transcripts split by speaker, so you can see exactly who said what.
  • Scoring that understands the step, not just the English phrase — a greeting in Hindi is still a greeting.
  • People who check the results, because no transcription is perfect, and mixed-language calls are where it matters most.

Workspace configuration in The Heartbeat app for a prepaid support team: primary language set to Other, with Hinglish flagged words such as “galat recharge” and “paisa wapas”

From The Heartbeat app: a prepaid support team taking Hindi, English and Hinglish calls. Its primary language is set to “Other”, which covers Hindi and code-switched Hinglish, and its flagged words are in Hinglish too: “galat recharge”, “paisa wapas”, “complaint karunga”.

A Hinglish call transcript in The Heartbeat app: “Mera recharge ho gaya, but data abhi tak nahi aaya”, with agent and customer lines and a time on every line

From The Heartbeat app: the call from the top of this post, Hindi and English together, exactly as it was said. It’s split by agent and customer with a time on every line, and “Kal bhi yahi bola tha” is marked as the customer’s negative moment.

Questions to ask any QA tool

1. Can it handle a call that switches language mid-sentence? Test it on your own calls, not a demo recording.

2. Does it score a Hindi greeting as a greeting? Check a few calls where the opening was in Hindi.

3. Who checks the output? Fully automatic results on mixed-language calls need a human check before anyone is coached on them.

4. What happens to calls it can’t handle? If they’re skipped, find out how many.

How The Heartbeat handles this

The Heartbeat reviews 100% of your agents’ calls, in any department. Analysts + AI go through every recording, and every AI flag is checked by a trained analyst before it reaches you.

Calls in Hindi, English and Hinglish are transcribed and scored. Every call is split by speaker, with a time on every line, and scored on the same ten soft skills whichever language the agent used. Your data stays in India, hosted on AWS ap-south-1 (Mumbai).

Your customers don’t pick one language. Your QA shouldn’t need them to.