The chatbot conversation changed: yesterday's menu-button robots became today's LLM-powered assistants that actually answer customers — in Hindi, Hinglish or English, at 2 a.m., about your specific products and policies. For small businesses, the technology finally matches the promise: 60–80% of routine queries handled automatically, leads captured while you sleep, at subscription costs under a part-timer's salary. Here is what works, what to buy, and how to deploy without embarrassing yourself.
The use cases that actually pay
- The 24×7 enquiry catcher: answering the ten questions every customer asks (price, timing, location, availability, process) — the after-hours leads currently dying in your unread inbox
- Lead qualification: intake conversations (requirement, budget, urgency, contact) delivering structured summaries to your team — sales mornings that start with qualified lists
- Order support: status checks, tracking, return initiation — the queries that swamp D2C sellers without adding a rupee
- Booking/appointment flows: slot display, booking, reminders — clinics, salons, consultants
- Internal copilots (the sleeper use case): staff asking 'what's our warranty policy?' against your documents — training and consistency without meetings
WhatsApp-first deployment
For Indian SMEs the chatbot's home is WhatsApp (where your customers already are), deployed via Business-API platforms with AI layers (Wati, AiSensy, Interakt and peers now ship LLM-powered bots; dedicated AI-agent tools bridge to WhatsApp too) — with website chat widgets and Instagram DMs as secondary skins of the same brain. Cost reality: ₹1,500–8,000/month for SME tiers (platform + AI usage + Meta conversation fees) — against which one saved staff-hour daily or a few recovered after-hours leads monthly already clears the bar. Setup effort: the honest work is content, not code — see the training section — with competent deployments live in 1–2 weeks.
LLM bots are only as good as what you feed them: the knowledge base (FAQs — real ones from your chat history, product/service details with prices, policies, process explanations) uploaded as documents the AI retrieves from; the boundaries (what it must not answer — medical/legal advice, discounts it can't give, competitor comparisons); the personality (tone matching your brand — warm Hinglish for retail, professional English for B2B); and the escalation triggers. The maintenance contract: review actual conversations weekly in month one (the bot's wrong answers are your knowledge-base gaps), monthly thereafter — an unmaintained bot slowly becomes a liability with a subscription.
The human-handoff rules (non-negotiable)
- Instant escape: 'baat karao', 'human', frustration signals or two failed answers → human queue with context transferred — bots that trap customers generate the anger that kills channels
- Honest identity: the bot introduces itself as an assistant — pretending to be human backfires legally and reputationally
- Business-hours choreography: off-hours bot commitments ('team will call you by 10 am') that your team actually honours
- The high-stakes carve-outs: complaints, refunds beyond policy, anything emotional — auto-route to humans; AI handles routine, people handle relationships
Measuring whether it's working
The dashboard that matters: containment rate (queries fully resolved without humans — 50–70% is realistic and excellent for SMEs), response coverage (after-hours conversations captured that previously died), lead volume and quality (qualified summaries per week), escalation quality (context-rich handoffs your team praises, not transcript dumps), and customer signals (conversation ratings, block rates on WhatsApp). The failure smells: containment below 30% (knowledge-base too thin), rising blocks (bot too pushy or trapped users), and staff ignoring bot leads (integration theatre). Iterate monthly against these and the bot compounds; deploy-and-forget and it decays into the digital equivalent of an IVR maze.
How Aidwish helps
Aidwish deploys AI assistants for client businesses — use-case selection, platform choice, knowledge-base building from your real conversations, handoff design and the monthly improvement loop — automation that answers like your best employee, not your worst menu.