WhatsApp is where Pakistani customers already are, which makes a chatbot on it an easy thing to want and a surprisingly easy thing to get wrong. The businesses that succeed with one are not the ones with the best natural language handling. They are the ones that were honest about which conversations should never reach a bot.
Here is what a WhatsApp chatbot genuinely fixes, what it quietly breaks, and what it costs to run properly.
What you are actually building on
Three things get called a WhatsApp chatbot and they are not the same product.
- WhatsApp Business app. Free, phone-based, with quick replies and away messages. Not a chatbot. A set of canned responses. Fine for a single-person business.
- WhatsApp Business API via a provider. The real option. Multiple agents, automation, CRM integration, message templates, and outbound notifications. This is what “WhatsApp chatbot” means commercially.
- An unofficial automation on a personal number. Cheap, and a reliable way to get your number banned. Businesses do this, and then lose the number their customers have saved for six years.
Everything below assumes the official API.
What it genuinely fixes
Repetitive questions, at volume
Business hours, location, price lists, delivery charges, “do you have this in stock”, “where is my order”. For a retailer or clinic, these can be 60 to 80% of incoming messages, and they are identical every time. Automating them is not a novelty; it removes most of the queue.
After-hours response
A significant share of Pakistani consumer messaging happens after 8pm. A bot that answers the common questions at 11pm captures intent that would otherwise be gone by morning. This is often where the measurable revenue effect actually is.
Order status and tracking
A bot connected to your order system turns “where is my parcel” from a staff task into a self-service lookup. High volume, entirely structured, zero judgement required. The ideal automation candidate.
Lead qualification
Four questions before a human takes over (what do you need, what size, what timeline, which city) routes the conversation to the right person with context already collected. This works well and is undersold.
Appointment booking and reminders
For clinics, salons and services. Reminders in particular have a direct, measurable effect on no-show rates, which is one of the few chatbot benefits that shows up cleanly in a spreadsheet.
What it quietly breaks
Complaints
An annoyed customer meeting a menu becomes an angrier customer. Anger should be detected and escalated immediately. Every bot needs a rule that routes frustration to a human on the first signal, and most deployments add it only after a public complaint.
Negotiation and anything unusual
Bulk pricing, custom orders, special arrangements. These are exactly the high-value conversations, and a bot standing between a serious buyer and a salesperson is losing money quietly. Nobody reports this in a dashboard.
The escape hatch
The most common single failure: no obvious way to reach a person. Every flow needs a visible route to a human, at every step. A bot the customer cannot get past is worse than no bot.
Promising what your systems cannot do
A bot that says “your order will arrive tomorrow” when your courier data is 24 hours stale creates a complaint that costs more than the automation saved. The bot is only as truthful as the system behind it.
Menu-driven or AI-driven?
Menu-driven (numbered options) is predictable, cheap, easy to fix, and works in Urdu, English and Roman Urdu without any language handling at all. Customers understand it immediately because it resembles a phone IVR.
AI-driven (the customer types freely and a language model answers) feels better and is genuinely stronger for varied questions. It also requires grounding in your real data, guardrails around what it may state, and a review process for what it actually said.
Our recommendation for most Pakistani SMEs: start menu-driven, add AI for the open-ended part. Menus for order status, hours, location and pricing; AI for product questions where the answers are documented. This gets 90% of the benefit at a fraction of the risk.
The Roman Urdu point deserves emphasis: a large share of customers write in Roman Urdu, and mixed-script handling is where generic bot platforms most often disappoint. Test this before you buy anything.
What it costs
- WhatsApp conversation charges. Meta charges per conversation category, with service conversations cheaper than marketing ones. For a business handling a few thousand conversations a month, budget in the tens of thousands of rupees.
- Platform or provider fee. 15,000 to 80,000 PKR per month depending on volume and features.
- Build. 150,000 to 400,000 PKR for a menu-driven bot with basic integration; 400,000 to 1,200,000 for AI-driven with real system connections.
- Maintenance. Ongoing. Flows need adjusting as products and policies change, and a bot nobody has updated in a year is confidently telling customers things that stopped being true.
Getting it right
- Read 200 real conversations first. Not what you imagine customers ask. What they actually asked last month. This is the single highest-value hour in the project and almost nobody does it.
- Automate the top five questions only. They will be most of your volume. Resist the ambition to handle everything.
- Put “talk to a person” in every menu. Every step, visible.
- Connect it to real data. A bot that cannot see your inventory or orders can only recite what a pinned message could have said.
- Detect frustration and escalate. Repeated messages, all caps, a complaint keyword. Hand over immediately.
- Review transcripts weekly for the first two months. You will find the flows that fail within days, and you will find questions you did not know customers had.
The honest test
If your business receives fewer than a hundred messages a week, a chatbot is not your bottleneck and building one is a distraction. Fix the answers first. A well-written FAQ page and a pinned message may resolve most of it. Above a few hundred a week with repetitive content, the economics turn quickly.
Related reading: where AI agents genuinely absorb operational work, and the AI tools small businesses actually keep using.
Frequently asked questions
How much does a WhatsApp chatbot cost in Pakistan?
150,000 to 400,000 PKR to build a menu-driven bot with basic integration, plus 15,000 to 80,000 per month in platform fees and Meta’s per-conversation charges. AI-driven bots with real system integration start around 400,000.
Can a WhatsApp bot handle Urdu and Roman Urdu?
Menu-driven bots handle any language trivially because customers pick numbers. AI-driven bots vary considerably in Roman Urdu quality. Test with real customer messages before committing to a platform.
Will WhatsApp ban my number for using a bot?
Not if you use the official WhatsApp Business API through an approved provider. Unofficial automation on a personal or Business-app number carries a genuine ban risk, and losing an established customer-facing number is an expensive way to save a subscription.
Do chatbots actually increase sales?
The reliable gains are response speed, after-hours capture and reduced no-shows. Businesses that automate their high-value conversations often lose more than they gain: the wins come from removing repetitive load, not from replacing salespeople.
Ezitech builds WhatsApp and AI automation connected to real business systems (inventory, orders, CRM), and will tell you when a pinned message would do the job instead. See our AI solutions or describe your message volume.