Hindi-First Ops for Indian SMBs

Indian small businesses are not under-digitized because owners hate software. They are under-digitized because most software is an English admin panel designed for a full-time operator. Attendance lives in a register. Khata lives in a notebook. Payroll is a cousin's spreadsheet. GST is a separate panic at month-end. We built Arth Saathi for that stack, not as a translation of an HRIS, but as a Hindi-first (and 13-other-languages-first) ops layer that feels like sending a message.

The product is a sentence, not a menu

The sentence we kept coming back to: "sabko present kar do." Mark everyone in. If that action takes seven taps in a sidebar, the owner will go back to the register. Voice and chat are not decoration. They are how you remove the menu. The assistant has to do attendance, leave, and salary questions in the language the owner already uses with staff, including on WhatsApp, which is where slips actually get read.

Translating labels is a trap. "Attendance" written in Hindi in a layout designed in English is still an English product. We put 14 languages through the owner app, the employee portal, the assistant, and notifications. That is a data and copy-system problem, not a locale JSON problem.

One data model or it falls apart

The reason owners hate tools is not features. It is four tools that do not talk. Attendance that does not feed payroll. Payroll that does not know advances. Invoices that do not know khata. We started from the two daily jobs (who showed up, who gets paid) and layered ledger, GST, and inventory on the same records.

• Voice + GPS attendance as the source for the month, not a parallel sheet
• Leaves and advances rolling into a payroll close that is one approval
• GST invoices (HSN, CGST/SGST, IRN) generated from the same customer/job, not retyped
• Khata as running credit against those customers, not a second set of names

WhatsApp is the client, the portal is the exception

Employees will not log into your beautiful portal every morning. They will open WhatsApp. Slips, invoices, and approval pings belong there. The portal exists for the owner who wants a table, and for the staff member who wants to see their own leave balance without asking. Seat-based SaaS thinking ("everyone must be a user") dies on contact with a 12-person salon.

Pricing has to look like a shop, not like Salesforce

We made the first 10 employees free and put paid tiers in the ₹500–₹1,000/month range after that. That is not a growth hack. It is the only number that can sit next to a kirana margin. If your SMB AI product starts at a Western SaaS seat price, you are selling to a different country than the one in the pitch.

What 10 shops in three weeks actually taught us

When we wrote about the first wave of businesses automating ops with Arth Saathi, the lesson was not "AI went viral." It was: owners who already had a workaround (notebook, biometric, CA at month-end) will switch only if the daily motion is shorter. Onboarding that starts with a data migration workshop does not ship. Onboarding that starts with marking today present, in the language they already speak, does.

• GPS on attendance is a trust feature for the owner, not surveillance theater, explain it in one sentence
• Payroll mistakes are remembered for years; show the math, do not hide it behind a model
• GST is where amateurs get the business fined, generate from structured HSN, do not let the LLM invent rates
• Support is WhatsApp, because that is where the product already lives

AI that is not a chatbot as the product

The assistant is a way to trigger structured actions. It is not a companion. If the model is down, the owner still needs a form. That is the same rule we use on client MVPs: do not treat the LLM as the product. For SMB ops, the product is a closed month: attendance in, slips out, invoice sent, khata updated.

If you are building for Bharat, not for a deck

Ask, in this order:

1. What does the owner do every morning without opening a laptop?
2. Which language do they use with staff, not with investors?
3. Where does the output have to land (WhatsApp, print, CA's email)?
4. What is the one number that makes the month close cheaper than the notebook?
5. What happens when the shop has no reliable data connection at 11am?

If those answers are vague, you are designing an English dashboard. See Arth Saathi, how we scope in 30 minutes, or inquire with the workflow and the language constraint. We will not start with a persona slide.