Conversational AI · Blog
Introduction
WhatsApp is where many Indian customers already are — so putting sales and support there feels obvious. The hard part is not connecting an API. The hard part is designing a bot that answers the right questions, stays within template and opt-in rules, and hands off to a human before the conversation becomes a liability.
At AVYRION, we build WhatsApp bots for retail, logistics, and service teams out of Gurugram. This guide explains how we scope those projects: which journeys belong on WhatsApp, what to leave out of v1, and how to avoid the common failure mode where a demo looks clever and production support quietly hates the bot.
The problem
Teams often treat WhatsApp as a free channel for unlimited free-form AI chat. That approach collides with Business API rules, brand risk, and the reality that most customers want status, booking, pricing, or a human — not a long chat that never resolves.
Without a clear journey map, bots become dumping grounds for every FAQ and every sales pitch. Response quality drifts, template approvals stall, and agents still live in a parallel inbox for anything important. Leadership then concludes “WhatsApp bots don’t work,” when the real issue was an unscoped product.
The second failure mode is skipping human handoff. When a bot cannot escalate with context, customers repeat themselves, agents lose trust in the channel, and your brand pays the cost in public chat threads.
The solution
Start with three high-volume journeys that have clear success states: order or ticket status, appointment or demo booking, and a short lead capture with qualification questions. If a journey cannot end in a definite outcome or a clean human handoff, it does not belong in v1.
Use approved message templates for outbound notifications and keep inbound flows structured — buttons, lists, and short forms — before you introduce free-form language models. Structured flows are easier to test, audit, and improve. Free-form AI can help later for FAQ paraphrasing once the happy paths are solid.
Design handoff as a first-class feature: pass chat history, customer identifiers, and the last intent to your helpdesk or CRM. Agents should never start from zero. Measure containment rate, handoff rate, and time-to-resolution — not vanity message volume.
Finally, treat WhatsApp as one surface on a system of record. Order status, inventory, and CRM truth should live in your backend. The bot is a channel, not a new database of customer promises.
Best practices
Map opt-in and quiet hours before you write copy. Customers who never consented, or who get midnight promotional blasts, will block the number and poison the channel for transactional messages you actually need.
Write template copy for the real failure cases: delayed shipments, cancelled appointments, payment pending. Happy-path templates are easy; exception templates are what reduce support load.
Keep session logic short. Long multi-step trees that expire mid-flow create more support tickets than they prevent. Prefer deep links and “continue where you left off” patterns for complex tasks.
Log intents, template failures, and handoff reasons. Without those signals, you cannot tell whether a bad week was product, content, or upstream API instability.
Budget for ongoing content ops. WhatsApp bots are not a one-time build — catalogs, FAQs, and offer seasonality change, and someone must own those updates.
Run a closed pilot with real agents before a public launch. Agents will surface wording gaps and edge cases that never appear in stakeholder demos, and fixing those early protects your brand number.
Document the escalation matrix in plain language: which intents go to sales, which to support, which to on-call ops, and what SLA customers should expect after handoff.
Examples from real delivery
For a logistics team drowning in “where is my shipment” chats, we shipped a status bot wired to their visibility layer, with WhatsApp milestone alerts and a clear escalate path for damaged or delayed parcels. Status-only contacts dropped because customers could self-serve the boring questions.
For a D2C brand, we built a sales-assist flow that answered sizing and shipping FAQs, captured intent, and booked callbacks during business hours. The bot did not try to close complex custom orders — it protected sales time for conversations that needed a human.
For a clinic network, appointment reminders and reschedule flows on WhatsApp reduced no-shows without putting clinical advice in free-form chat — a boundary we kept explicit for safety and compliance.
In each case, the winning pattern was the same: one system of record, a short list of journeys, and a handoff path agents trusted. Fancy generative answers came later, if at all, once containment metrics were stable.
Common mistakes
Launching an open-ended AI chat with no grounding in your policies, prices, or inventory. Hallucinated promises are expensive on WhatsApp because the transcript is the customer’s evidence.
Ignoring template approval lead time in the project plan. If go-live depends on approved templates, start that process in discovery, not in the last week.
Building the bot in isolation from CRM and helpdesk. Without handoff and identity, every conversation resets and agents hate the channel.
Measuring success only by messages sent. High message volume with low resolution is noise, not automation.
Skipping a kill switch and fallback. When the upstream API fails, customers need a honest “we’re checking” path and a human route — not a broken loop.
Letting marketing own the number with no ops owner. When campaigns change offers weekly but nobody updates bot copy, the channel becomes a contradiction engine.
Conclusion
A useful WhatsApp bot is a product with boundaries: a few journeys, clear outcomes, approved templates, and respectful handoff to humans. Treat the Business API as a channel into systems you already trust, not as a place to invent a second source of truth.
AVYRION builds WhatsApp bot development for sales and support teams that need measurable containment — not demo theater. If you want a scoped v1 for your catalog, logistics, or booking flows, talk to us and we will propose the lightest milestone that still changes daily ops.
Bring your top three chat topics and your systems of record to a discovery call. That is enough for us to tell you whether WhatsApp is the right first channel — and what a sober four-to-eight-week MVP can realistically automate.
