A chatbot is only useful if it answers the question the customer actually asked. Rule-based bots with fixed menus handle the top 20 questions well and fall apart on the 21st. Large language model (LLM) bots handle open questions but, without guardrails, invent fee structures and delivery dates. The right answer for most Indian businesses is a hybrid: structured flows for booking, payments and data collection, plus an LLM layer trained on your documents for everything else.
AICLEX Technologies designs and builds exactly that. As an AI-first product studio we have shipped chatbots on WhatsApp through our WhatsPilot platform and on websites, apps and Instagram for coaching institutes, developers, clinics and D2C brands. This page explains how we decide between rule-based and AI, how training on your data works, how handoff to a human is done properly and what it costs.
Rule-based vs LLM-based: an honest comparison
Rule-based bots follow a tree you design. They are predictable, cheap to run, need no AI credits and are ideal for tasks with a fixed shape: pick a slot, choose a plan, enter a PIN code, confirm an order. Their weakness is language. A customer typing 'kal subah free ho kya?' does not match a button labelled 'Book appointment'.
LLM-based bots understand that sentence and reply naturally in Hinglish. They can read your brochure, fee sheet, refund policy and product catalogue and answer from them. Their weakness is confidence: without retrieval and constraints they will guess. Our standard build uses retrieval-augmented generation (RAG) so the model answers only from your approved content, says 'I will connect you to the team' when it is unsure, and never quotes a price that is not in your documents.
- Rule-based only: FAQ menus, appointment booking, order tracking, lead forms. Budget-friendly, no per-message AI cost.
- Hybrid (our default): structured flows plus an AI fallback trained on your data. Handles 80 to 90 percent of conversations end to end.
- Full agentic: AI that can look up a booking, raise a ticket or update a CRM record mid-conversation. For businesses with mature systems.
Training the bot on your business data
Training does not mean the model memorises your data. We collect your source material: website pages, PDF brochures, price lists, policy documents, past WhatsApp chats (with personal data removed), and a list of the 100 most common questions your team answers. We chunk and index this content so the bot retrieves the relevant passages for every question and answers from them, with a citation visible to your team in the review console.
We then run a test set of 150 to 300 real questions, including messy ones in Hindi, Hinglish, Tamil or whichever languages your customers use, and grade the answers with your team. Anything wrong is fixed by adding or correcting a source document, not by hoping the model improves. When your prices change, you update one document and the bot changes the same day.
Handoff to humans: the part most vendors skip
The bot must know when to stop. We configure handoff triggers: the customer asks for a person, the customer expresses anger or urgency, the question involves money disputes or medical advice, or the AI's confidence score drops below a threshold. On WhatsApp, handoff assigns the chat to a named agent in the WhatsPilot inbox and sends them a push notification with the full transcript. On a website, the widget shows 'Priya from the team has joined'.
Outside working hours the bot collects the details, tells the customer when to expect a reply, and creates a task. If you use Zonvo, a high-intent lead can be called by the AI voice agent within minutes, and a human joins only if the caller asks. Nothing falls through because a chatbot went quiet.
Sample conversation: real-estate developer website bot
Visitor: 'sector 150 wale project me 3bhk ka rate kya hai'. Bot: 'Sector 150 project mein 3 BHK units 1,650 sq ft aur 1,850 sq ft mein available hain. Current base price sheet ke hisaab se starting Rs X.XX crore (all-inclusive nahin, PLC aur GST alag). Kya aap price sheet PDF chahte hain ya site visit book karna chahenge?' Visitor taps 'Site visit'. Bot collects name, phone and preferred day, checks the site team's calendar, confirms Saturday 11 am and pushes the lead to Siteboard CRM with source 'website bot'.
The price in that answer came from the developer's approved rate sheet, not the model's imagination, and the bot explicitly named what was excluded. That is the standard we build to.
Build process and timeline
Week 1: discovery, gathering documents and question lists, agreeing the flows that need to be structured. Week 2: build flows, index content, set up handoff rules and connect channels (website widget, WhatsApp via WhatsPilot, Instagram DM). Week 3: testing with your team's question set, tone adjustments, language checks. Week 4: soft launch to a portion of traffic, review transcripts daily, then full launch.
Rule-based WhatsApp bots go live in 7 to 10 working days. Hybrid AI bots take 3 to 5 weeks. We provide a review console where your team can read conversations, flag wrong answers and add knowledge without calling us.
Measuring whether the bot works
We report containment rate (conversations resolved without a human), handoff rate and reason, average first-response time, answer accuracy from weekly transcript sampling, leads or bookings created, and customer thumbs-up ratio. A bot with 90 percent containment and 70 percent accuracy is worse than one with 70 percent containment and 95 percent accuracy. We tune for accuracy first.
Who this is for
- Coaching institutes and edtech companies fielding hundreds of fee, batch and syllabus questions daily
- Real-estate developers whose site team cannot answer website and portal leads fast enough
- Hospitals, clinics and diagnostic chains that want triage, booking and report queries automated
- D2C brands needing product recommendations, order status and return handling on WhatsApp and web
- SaaS and B2B companies wanting a support bot trained on their documentation
- Colleges and universities managing admission enquiries in multiple languages
Why choose AICLEX™
Product studio, not a prompt shop
We build and run WhatsPilot and Zonvo, so your chatbot sits on infrastructure we maintain rather than a chain of third-party subscriptions.
Answers grounded in your documents
Retrieval with citations means every fact the bot states can be traced to a file you approved. Wrong answers are fixed by editing content, not by guesswork.
Indian languages tested, not assumed
Hindi, Hinglish and regional-language test sets are part of every build, including transliterated typing like 'kitna time lagega'.
Handoff and voice in the same system
Escalation to a human inbox or a Zonvo AI call is configured on day one, so the bot never becomes a dead end.
Pricing
Rule-based WhatsApp or website chatbot flows start at Rs 15,000 one-time plus WhatsPilot from Rs 1,599 per month. Hybrid AI chatbots trained on your data start around Rs 50,000 and go up with integrations, languages and channels. LLM usage is billed at cost or bundled in a monthly plan. GST invoice included. Fixed-price proposal after a discovery call.
Frequently asked questions
How much does an AI chatbot cost in India?+
A rule-based bot on WhatsApp or your website starts at Rs 15,000 one-time. A hybrid AI bot that answers from your documents starts around Rs 50,000 and typically lands between Rs 50,000 and Rs 1,50,000 depending on integrations, channels and languages. Monthly costs are the WhatsPilot plan plus any AI usage, which we quote transparently.
Can the chatbot work on my existing WhatsApp number?+
Yes if that number moves to the WhatsApp Business API. A number cannot run the WhatsApp Business App and the API at the same time, and chats do not migrate. We plan the switch with you, or set the bot up on a new number while the old one stays on the app during transition.
Will the AI give wrong information to my customers?+
Any AI can make mistakes, which is why we constrain it to answer only from your approved content and to escalate when unsure. Before launch we test hundreds of real questions with your team and fix gaps. After launch, your review console lets you flag and correct any answer within minutes.
Does the bot need a green tick or Meta verification?+
WhatsApp bots need Meta Business verification to reach full messaging limits, which we help complete. The green tick is a separate badge granted at Meta's discretion and is not required for a bot to function. Website bots need neither.
How long does it take to build and launch?+
Rule-based bots take 7 to 10 working days. Hybrid AI bots trained on your data take 3 to 5 weeks including testing and a soft launch. Timelines stretch if your content is scattered or if CRM and calendar integrations are complex, and we flag that at discovery.
Where is my data stored and do I get a GST invoice?+
Your documents and conversation logs are stored on India-hosted servers with access limited to your accounts. We sign an NDA on request and remove personal data from any chat history used for training. All payments to Aiclex Solutions Pvt. Ltd. carry a GST invoice.