Why Roman Urdu is the real test
Most Pakistani customers do not write to businesses in formal Urdu script, and they rarely write in clean English either. They write Roman Urdu: Urdu in Latin letters, with English words dropped in and spelling that changes from person to person. "Kal 5 baje appointment mil sakti hai?" "Rate kya hai bridal ka?" "Dr sahab aaj baithein ge?" A Roman Urdu WhatsApp chatbot has to read all of that, plus the voice notes customers send when typing feels like too much effort.
We put together the kinds of messages Pakistani clinics, salons and academies receive every week and walked through how a language-model assistant like NextReply handles them. This post shares those sample messages, the patterns that work, the ones that need guardrails and a simple test you can run on your own number before you trust any tool, ours included.
If you want the product overview first, see the NextReply Pakistan page.
The message set: 12 messages every Pakistani business gets
Use a set like this when you evaluate any assistant. It covers spelling drift, script switching, time expressions and voice notes:
- "kal 5 baje mil jaye ga?" (Will I get 5 o'clock tomorrow?)
- "kl shaam ko koi slot hai" (any slot tomorrow evening, with "kal" shortened to "kl")
- "Rate kya hai keratin ka? Lambe baal hain" (keratin price, long hair)
- "Dr sahab aaj baithein ge? Kitne baje tak?" (Is the doctor in today, until when?)
- "parson subah 11 baje theek hai" (day after tomorrow, 11am)
- "آج شام چھ بجے" (6pm today, in Urdu script)
- "Bhai booking cancel kar do, emergency aa gayi" (cancel)
- "Sat ko shift kar dein same time" (move to Saturday)
- "Mehndi wale din ka package hai?" (wedding-day package)
- "Payment Easypaisa pe bhej di, screenshot dekh lein" with a photo
- A 30-second Urdu voice note asking for a Sunday appointment for an elderly mother
- A voice note in Punjabi-flavoured Urdu with traffic noise in the background
These are representative messages, not a scientific benchmark. Your own customers will have their own habits, which is why the last section of this post tells you how to run the same check on your chats.
What a language-model assistant reads well
Older WhatsApp bots matched exact keywords. Type "rate" and you got the price list; type "rates kya hain" and you got "Sorry, I didn't understand". A language model reads meaning, which changes the picture for Roman Urdu:
- Spelling variation. "Kal", "kl" and "kall" are the same word. "Mil jaye ga", "mil jayega" and "mil jae ga" too.
- Time expressions. "Saarhay 5" is 5:30, "pone 4" is 3:45, "shaam" sets pm, "parson" is the day after tomorrow.
- Script switching. A customer who switches to Urdu script mid-chat is still understood, and the reply can follow in the same script.
- Two questions in one message. "Rate kya hai aur parking hai?" gets both answers.
- Politeness. You can set the AI to use "aap", "ji" and "sahab" and avoid casual forms.
The key safeguard is the confirmation line. NextReply names the weekday, date, time, service and staff member in every confirmation ("Saturday 18 October, 5:00pm, keratin with Sana"), so any misreading surfaces before it reaches your calendar.
Where it can trip, and the fix for each
"Kal" means yesterday and tomorrow
In Urdu "kal" covers both. In a booking chat it almost always means tomorrow, and the AI reads it that way, but a complaint such as "kal aap ne bohat wait karwaya" refers to yesterday. Fix: the confirmation always names the weekday, and complaints go to staff.
Wedding vocabulary and local package names
"Mehndi package", "baraat makeup", "walima look" are specific to your price list. If you have not listed them, the AI should not invent a price. Fix: add your packages with these exact names and the alternative spellings customers use.
Payment screenshots
The AI can see that a customer sent an Easypaisa or JazzCash screenshot, but it cannot confirm money reached your account. Fix: set a rule that payment screenshots go to staff, and the AI says "Shukriya, hamari team payment confirm kar ke aap ko bata degi."
Noisy voice notes
Rickshaw horns, wedding music or a crowded bazaar can blur a word. Fix: when the time or date is unclear, the AI asks one short question ("Sunday 11 baje, theek hai?") instead of guessing.
Regional languages
Customers in Peshawar, Multan or Hyderabad may switch into Pashto, Saraiki or Sindhi. NextReply replies in the customer's own language, but quality varies by language and accent. Test with samples from your actual customers.
- Unusual request: Complaint, refund or special case
- AI receptionist: Knows when to step back
- Answers from your FAQs: Services, hours and prices / Hands the chat over: Your team is notified
- Your team replies: Same chat, full history
Urdu voice notes: what happens behind the reply
Voice notes are how many Pakistani customers, especially older ones, prefer to talk to a business. An Urdu voice note bot worth using does three things: it understands the audio as part of the conversation, it replies in text the customer can reread, and it asks for confirmation when something is unclear.
Take message 11 from the set above. A son records: "Assalam o alaikum, ammi ke liye Sunday ka appointment chahiye, Dr Farah ke saath, subah ka time ho to behtar hai." NextReply reads the request, checks Dr Farah's Sunday morning slots in Google Calendar, offers two times and asks for the patient's name. The son replies "Shamim Akhtar, 10:30" and gets a structured confirmation. If the clinic has asked for it, the AI also adds a fixed line about bringing old reports.
What the AI does not do: it never interprets symptoms described in a voice note. Clinics set handoff rules so that anything medical goes to staff with an instant notification.
Replying in the customer's style
The default that works best in Pakistan is mirroring: Roman Urdu in, Roman Urdu out; Urdu script in, Urdu script out; English in, English out. A customer who writes "Assalam o alaikum" should get the greeting back. A corporate client who writes in English should not suddenly receive Roman Urdu.
You choose the personality: warm and informal for a salon on MM Alam Road, formal and brief for a specialist clinic in F-8, friendly but exact for an academy in Gulshan-e-Iqbal. The AI keeps that tone across languages.
- مرحبا، عندكم موعد؟ / Olá, tem horário? / Hola, ¿hay cita? / Hi, any slots?
- Detects the language: From a WhatsApp text or voice note
- Replies in that language: You set it up once
How to test it on your own chats in 30 minutes
- Export 20 real messages from last month: 12 typed in Roman Urdu or mixed English, 3 in Urdu script, 3 voice notes and 2 complaints.
- Set up NextReply with your services, prices, timings and FAQs, and connect Google Calendar. The how it works page shows the steps.
- Send each message to your number from another phone, exactly as the customer wrote it.
- Score each reply: correct meaning, correct price or slot, right tone, and a handoff where one was needed.
- Fix the knowledge base for every miss: add the package name, the spelling variation or the handoff rule. Then send the message again.
Most misses come from missing information, not language. Once your service list uses your customers' words, the replies improve quickly. You can see what the AI can do on the features page and how bookings work on appointment booking.
Cost, channels and where to start
NextReply plans are priced in US dollars, the same in every country: Starter $29, Pro $59 and Business $99 a month, and every plan has a 7-day free trial. At checkout, your card or bank may show an estimate in rupees, and local taxes may apply. Plans cover up to 500, 1,500 or unlimited bookings. Details are on the pricing page. Meta charges separately for any template messages you send.
Roman Urdu does not stay on WhatsApp. Instagram DMs to boutiques and salons look just the same, and many families still message clinics through Facebook. Instagram, Messenger and Telegram are coming soon to NextReply and will book into the same calendar. Today the assistant works on WhatsApp, and Roman Urdu voice notes and photos are handled there. See the channels page. Most owners start with nights and weekends, as explained on after-hours replies, and you can find examples by business type on industries or more guides on our blog.