Open any real work group in a factory in Nashik, a CA office in Pune or an agency in Indore and read twenty messages. You will not find twenty English sentences, and you will not find twenty Hindi ones either. You will find something like *“Sharma sir ka GSTR-3B main aaj shaam tak file kar dungi, challan mil gaya hai.”* — one sentence, two languages, one deadline, one owner. That is not a degenerate case to handle later. That is the median message.
What the mixing actually looks like
It's worth being precise, because “supports Hindi” is not the same claim as “reads what your team types”. There are at least four distinct patterns in ordinary operational chat, and a system has to handle all of them to be useful.
| Pattern | Example | What's hard about it |
|---|---|---|
| Devanagari Hindi | कल सुबह तक भेज देता हूँ | Different script, same meaning |
| Romanised Hindi (Hinglish) | Kal subah tak bhej deta hoon | No standard spelling — subah / subeh / sube |
| Code-mixed in one sentence | Payment gateway key rotate karke deploy kar deta hoon by 6 | English nouns, Hindi verbs, English time |
| Marathi, romanised | Udya sakali pathvto | Third language, same group, no warning |
There is no language toggle in anyone's head. The same person writes all four in one afternoon, choosing whichever gets the thought out fastest. Any system that expects a declared language has already lost.
Time is the hard part
Identifying that a sentence contains a commitment is the easier half. Working out *when* is where most of the difficulty sits, because operational time expressions are relative, vague and culturally loaded all at once.
- “Shaam tak” — by evening. Not a clock time, but a real deadline, and everybody in the group agrees roughly what it means.
- “Kal subah” — tomorrow morning, which in a factory running a 7 AM shift means something different than in an agency that starts at 10.
- “Lunch ke baad” — after lunch, anchored to a shift pattern rather than a clock.
- “Truck aane se pehle” — before the truck arrives, anchored to an event that itself isn't scheduled precisely.
- “EOD” — imported English office shorthand, used freely in otherwise Hindi sentences.
A useful system doesn't need to resolve every one of these to a precise minute. It needs to be right about the *day part*, and honest when it isn't sure. Being wrong in a small way is survivable; being confidently wrong about when something is due is how people stop trusting a nudge.
The softness problem
Here is the subtlety that decides whether a tool is pleasant or exhausting to live with. Not every forward-looking sentence is a promise, and Indian workplace language is full of carefully calibrated softness.
| What was said | Is it a commitment? |
|---|---|
| “Kar deta hoon” | Yes — I'll do it |
| “Kal tak bhej dunga” | Yes, with a deadline |
| “Dekhta hoon” | No — I'll look into it |
| “Try karta hoon” | Softened on purpose; not a firm promise |
| “Ho jayega” | Ambiguous — often a reassurance, not an undertaking |
Treat “dekhta hoon” as a hard commitment and you will nudge somebody about a thing they never agreed to do. Do that twice and they will start resenting the tool, which is a far worse outcome than missing a task. The bias has to be towards under-catching, because a missed catch costs one task and a false catch costs trust.
A tool that chases people about things they didn't promise gets uninstalled faster than one that occasionally misses something.
Why “just write in English” isn't an answer
It gets suggested in every one of these conversations, usually by someone who doesn't have to do it. It fails on three separate counts, any one of which would be enough.
- It taxes the busiest people most. A supervisor mid-changeover types in whatever comes fastest. Asking for a translation step is asking him to slow down at the worst moment.
- It degrades the message. Forced-English versions of operational instructions are shorter, vaguer and lose the nuance that made the original useful to the team.
- It's backwards. The software exists to serve the team. A team that has to change how it speaks so the software can cope has bought a second job, not a tool.
What we're honest about
Reading mixed-language operational chat well is genuinely hard, and we'd rather say where the edges are than pretend there aren't any. Very terse replies — a bare “ok”, “haan”, “done” — depend entirely on what they're replying to, which is why a quoted reply carries so much weight. Regional vocabulary varies more than any test set can cover. And the softness scale above is a judgment call that different managers would make differently.
What we do commit to: your team never has to change how it writes, a missed commitment is a bug we want to hear about, and we would rather catch nine out of ten things quietly than catch eleven and invent one.
Written by Team Scoop, Founding team, Scoop. General guidance on running a team, not professional or legal advice for a specific situation.