Try for free
  1. Home
  2. /
  3. Blog
  4. /
  5. Anish Korat

Anish Korat

How chatbots work · Vatdi team

Anish explains how retrieval-based chat works in plain words: what the assistant reads, why it can say "I don't know", how languages are detected, how to test accuracy, and where it still fails. He cites the research and vendor documentation rather than borrowed statistics.

Comparison of scripted chatbot behaviour and conversational AI answering from store content
Chatbot fundamentals
8 views

Conversational AI vs chatbot is largely a vocabulary argument, but three real differences sit underneath it: whether the software follows a script or reads your content, whether it handles wording nobody anticipated, and whether it holds context across a conversation. This guide explains the terms plainly, gives four test questions that reveal which you are looking at, and says which differences matter for a store.

Read More
Diagram separating what belongs in chatbot instructions from what belongs in store content
Chatbot fundamentals
72 views

AI chatbot instructions are short by design. They set tone, scope and escalation, and nothing else: facts belong in your content, where they can be updated without touching the assistant. This guide covers what to include, a template you can adapt in ten minutes, the additions that quietly make answers worse, and how to tell instructions problems from content problems.

Read More
Comparison of a question-answering assistant and an action-taking agent for an online store
Chatbot fundamentals
79 views

Chatbot vs AI agent is mostly a marketing distinction, but one real difference sits underneath: whether the software only answers or also takes actions on your systems. This guide defines both terms plainly, shows where a store genuinely benefits from action-taking, where it adds risk for little gain, and the questions that cut through a vendor's wording to what the product actually does.

Read More
Checklist contrasting when to pick a no-code chatbot builder and when to pick an AI trained on your own content
Chatbot fundamentals
167 views

A no-code chatbot builder lets you draw the conversation: menus, branches, forms. An AI trained on your own data reads your content and answers whatever is asked. They solve different problems, and picking the wrong one means either weeks of flow-building or a bot that wanders. This guide shows what each is good at, the real effort to build and maintain each, prices as of September 2026, and a decision list.

Read More
Decision flow between an AI chatbot and live chat: content and order questions to the bot, judgement calls to a person
Chatbot fundamentals
203 views

Chatbot vs live chat is a false choice for most stores: the questions with an answer in your content or order data belong to a retrieval-based chatbot, at any hour, and the ones needing judgement belong to a person. This guide compares the two on availability, cost unit, answer quality and exceptions, and sets out the handover rules and pricing that make running both practical.

Read More
Large statistic tile: 39.6 percent of store chat conversations started outside 9 to 6 UTC, from 884 real conversations
Chatbot fundamentals
279 views

Most ecommerce chatbot statistics online cannot be traced to a dataset. These can: 884 real conversations across 63 stores on Vatdi over 90 days, eval traffic excluded, with the method stated. They show when shoppers ask, what they ask about, how long conversations run, how often a person is needed, where shoppers are, and how often the answers were good, plus the one external figure we consider sourced.

Read More
Flow of a store chatbot: question, retrieval from catalogue and policies, answer with product cards, grade, content fix
Chatbot fundamentals
382 views

Nine techniques that move a store chatbot from "answers something" to "answers correctly": grounding every reply in retrieved content, structuring the catalogue so attributes are findable, writing policy pages a retriever can use, steering with quick replies, handover rules, per-language testing, coupon conditions, guardrails on prices and promises, and a 30-question quality loop you re-run after every content change.

Read More
Pipeline of a RAG chatbot: question, retrieval over your content, generation from the passages, answer or a decline
Chatbot fundamentals
961 views

A RAG chatbot answers in two steps: it retrieves the passages of your own content that match the question, then writes an answer from those passages only, declining when nothing matches. This guide walks one question through both steps, explains chunks, embeddings and keyword search in plain words, lists the ways retrieval still fails on a real store and the content fix for each, and shows how to test it.

Read More

Ready to Add AI Chatbot to Your Store?

Join thousands of ecommerce stores using Vatdi for 24/7 customer support, lead capture, and sales automation.

Try for free Book a Demo
Free plan available No credit card required 5-minute setup