What it means
Getting the intent right on the first message is what separates a support AI that resolves issues from one that frustrates customers.
Intent detection is the decision-making backbone of a support AI. Every incoming message must be classified: what does this customer want? In traditional chatbot frameworks, intents were hand-labeled categories with manually curated example phrases. Modern LLM-based systems infer intent from context and meaning, making them much more robust to the variety of ways customers express the same underlying need. Accurate intent detection directly drives downstream quality: get the intent right and the rest of the conversation flows logically; misclassify it and every subsequent step is wrong. In ecommerce, the most common high-volume intents — order tracking, return initiation, refund status, product inquiry — can be detected with high accuracy because they have clear linguistic signals, even when customers phrase them in unexpected ways.
Why it matters
Intent detection accuracy is the single biggest lever for improving AI support quality. Each misclassified intent is a broken support interaction. For high-volume Shopify stores handling thousands of conversations daily, even a 5% improvement in intent detection accuracy prevents hundreds of failed interactions per day — reducing escalations, cutting handle time, and improving CSAT scores directly.
Related concepts, explained
These terms are part of the same idea, so they live here rather than on pages of their own.
Intent Classification
Intent classification is the AI task of automatically assigning an incoming customer message to one of a predefined set of intent categories — such as 'order tracking,' 'refund request,' or 'product question' — based on the message's meaning, enabling downstream routing, response selection, and action triggering.
Every incoming customer message must be understood before it can be handled. Intent classification is the formal AI task of assigning that understanding: given this message, which category of customer need does it represent? In ecommerce support, intents map to operational workflows: a 'return initiation' intent triggers the return eligibility check workflow; a 'shipping status' intent triggers an order lookup; a 'product question' intent retrieves relevant product catalog content. Classification can be binary (is this a refund request or not?) or multi-class (which of 20 intent categories does this belong to?) and can handle multi-intent messages where a single customer message expresses more than one underlying need. Modern LLM-based classifiers greatly outperform the older intent-slot models by understanding meaning and context rather than keyword patterns — enabling much higher accuracy on the wide variety of ways customers express the same underlying need.
Intent classification accuracy is the single most impactful variable in AI support quality. Misclassify a refund request as a product question and the customer receives a response about the product instead of about their money — a frustrating experience that typically escalates. For Shopify merchants, the most important intents by volume are consistent across stores: order tracking, return initiation, refund status, shipping delay, product inquiry, and discount code use. Getting classification right for those five or six categories handles the vast majority of support volume and delivers the largest impact on resolution rates.
Entity Extraction
Entity extraction is the AI process of identifying and pulling specific, structured pieces of information from a customer's free-form message — such as order numbers, product names, email addresses, delivery dates, or monetary amounts — making that data available for downstream actions like order lookups, ticket creation, or personalized responses.
When a customer contacts support, their message often contains specific data points the support system needs: an order number, a product SKU, a shipping address, a date, an amount. Entity extraction is the task of automatically identifying and structuring those data points from free-form text. In traditional support workflows, agents manually copy order numbers from messages into lookup tools. AI entity extraction does this automatically: the system reads 'I never received order 1023-B that I placed on March 12' and extracts {order_id: '1023-B', event_date: 'March 12'} as structured data, immediately usable for an API call to Shopify to retrieve that order. The combination of entity extraction with intent classification is what enables true end-to-end AI resolution: the intent tells the system what to do, the extracted entities provide the data needed to do it.
Every question an AI has to ask a customer to get information it could have extracted from the original message is unnecessary friction. 'Can I get your order number?' when the order number was in the first message is a failure of entity extraction that delays resolution and erodes trust in the AI. For Shopify merchants, strong entity extraction means the AI can immediately look up the specific order, product, or account the customer is asking about — delivering a personalized, accurate response rather than a generic one that requires a follow-up exchange.
Slot Filling
Slot filling is the conversational AI process of systematically collecting all required pieces of information (slots) needed to complete a specific task or workflow — asking follow-up questions for each missing piece until all required data has been gathered and the action can be executed.
Many support actions require multiple pieces of information to execute. Processing a return requires: order number, the item being returned, the reason for the return, and optionally a preferred resolution (refund vs. exchange). Slot filling is the structured process of collecting these required data points across a conversation. The AI defines a 'frame' — a set of required slots for this action type — and tracks which slots are already filled (from entity extraction or prior conversation turns) and which still need to be collected. It then asks only for the missing pieces, in natural language, one at a time. Done well, slot filling feels like a natural conversation rather than a form: the AI adapts its questions based on what it already knows, skips questions whose answers are already clear from context, and confirms the full set of collected information before executing the action.
Slot filling is what makes multi-step ecommerce processes automatable through natural conversation. Without it, AI support is limited to answering questions — it can't execute workflows that require gathering multiple data points. With robust slot filling, a customer can initiate and complete a return entirely through a chat conversation, without navigating a portal, finding a form, or waiting for a human. For Shopify merchants, this translates directly to a higher percentage of support interactions that resolve fully without human involvement, lowering cost per resolution.
Customer Intent
Customer intent is the underlying goal or request a customer is expressing in a support message — what they actually want done — as determined by AI classification of their words and context.
In ecommerce support, intent classification is the process of reading a customer's message and determining what they want: to know their order status, to initiate a return, to request a refund, to cancel an order, to change a shipping address, to report a defective item, to leave feedback, and so on. This sounds straightforward, but natural language is messy. 'I need to send this back' means the same as 'how do I return this?' and 'I don't want it'. 'My package is taking forever' could be an order status request or a frustration expression that needs an empathetic response before any data lookup. Intent classification sits at the top of the resolution pipeline: get it right, and the rest flows efficiently; get it wrong, and every downstream step is solving the wrong problem. Good intent systems handle multi-intent messages ('I want to return item A and also update my address for the reshipment of item B'), low-signal messages ('I'm very unhappy'), and novel phrasings not in the training set.
Every support routing, prioritization, and resolution decision depends on knowing what the customer wants. For ecommerce AI, intent accuracy directly drives resolution accuracy and customer satisfaction. A misclassified intent that routes a refund request to the order status flow means the customer gets tracking information when they wanted their money back — a frustrating and trust-eroding experience. High intent accuracy is the foundation everything else is built on.
How Bookbag helps
Shopify-Tuned Intent Library
Bookbag ships with a pre-built intent library tailored to ecommerce support — covering order tracking, returns, refunds, product questions, shipping issues, and more — with high out-of-the-box accuracy.
Custom Intent Training
Merchants can add custom intents specific to their store — subscription management, loyalty points, product customization questions — and provide example phrases for Bookbag to learn from.
Fallback on Low Confidence
When intent confidence falls below threshold, Bookbag asks a clarifying question rather than proceeding with a guess, ensuring misclassifications don't cascade into wrong resolutions.
Go deeper
Frequently Asked Questions
See Bookbag in action
Join the ecommerce teams resolving more tickets, answering 24/7, and turning support into a revenue channel with Bookbag.