What it means
Conversation design is UX design for dialogue — the same craft applied to words instead of screens.
Building a capable AI chatbot and building a good customer experience with that chatbot are not the same problem. Conversation design is the craft that bridges them. A conversation designer maps out the paths a customer might take ('they want a refund → do they have their order number? → is the order eligible? → confirm resolution preference'), writes the prompts that guide customers through those paths, crafts the fallback messages for when the AI doesn't understand, establishes the escalation phrasing that feels helpful rather than abandoning, and ensures every interaction reflects the brand's voice. In LLM-based systems, explicit conversation design is still critical even though the model can generate flexible language: designing the scope boundaries, the error recovery behaviors, the escalation points, and the confirmation patterns for high-stakes actions is not something to leave entirely to model judgment. The best AI support experiences combine LLM flexibility with deliberate conversation design.
Why it matters
Two AI systems with identical underlying capabilities can produce dramatically different customer experiences depending on their conversation design. The Shopify store whose AI confidently apologizes and offers a clear next step when it can't help retains customers; the store whose AI produces a generic 'I didn't understand that' message loses them. For merchants investing in AI support, conversation design is the leverage point that turns technical capability into actual customer satisfaction improvement. It's also the fastest fix for poor customer experience when the underlying AI is already capable — often a design issue, not a model quality issue.
Related concepts, explained
These terms are part of the same idea, so they live here rather than on pages of their own.
Dialogue Management
Dialogue management is the AI component responsible for tracking the current state of a conversation, determining what information has been exchanged, and deciding what action to take next — whether to answer, ask a follow-up question, execute an action, or escalate — based on the conversation history and current context.
A conversational AI system has many components — intent classification, entity extraction, knowledge retrieval, response generation — but dialogue management orchestrates them all. At each turn of the conversation, the dialogue manager assesses the current state: What has the customer told us? What do we still need? What action is appropriate next? It maintains a conversation state machine that tracks progress through workflows (has the slot been filled? has eligibility been confirmed?), handles interruptions (customer changes topic mid-return-flow), and manages conversational repair (customer provides clarification to a previous question). In simple rule-based systems, dialogue management is a rigid decision tree. In modern LLM-based systems, dialogue management is largely emergent from the model's contextual reasoning — but production support AI often combines LLM flexibility with explicit state tracking for structured workflows like return processing or account updates.
Poor dialogue management is responsible for the most frustrating chatbot experiences: bots that ignore what you just said, reset to the beginning when you change topics slightly, or get stuck in loops asking for information you've already provided. For Shopify merchants, the impact is direct: a customer who gets trapped in a confusing bot flow abandons the interaction and either contacts a human (adding cost) or gives up entirely (reducing retention). Good dialogue management makes the AI feel like a coherent, purposeful conversation partner rather than a non-deterministic question generator.
Chatbot Flow
A chatbot flow is a structured, branching sequence of conversation steps — including prompts, conditional branches based on customer inputs, data collection points, and resolution actions — that guides an AI or scripted chatbot through a specific support scenario toward a defined outcome.
Chatbot flows formalize what a skilled support agent does instinctively: understand the customer's need, collect any required information, check eligibility or conditions, and execute a resolution. In practice, a flow for 'process a return' might look like: (1) confirm the customer wants to return an item, (2) collect or verify the order number, (3) ask which item is being returned, (4) check return eligibility against your policy, (5) if eligible: confirm return method and generate return label; (6) if not eligible: explain why and offer alternatives. Flows vary in rigidity. Rule-based chatbots follow rigid flows where every branch must be pre-authored. LLM-based systems can follow flexible flows where the model handles language variation naturally within a defined structure. Hybrid approaches — LLM for language, explicit flow logic for process control — combine the best of both, ensuring structured processes execute reliably while keeping conversations feeling natural.
Chatbot flows encode your support best practices in a repeatable, scalable form. Every customer who contacts about a return gets the same well-structured experience as if your best agent handled it — because the flow embodies what your best agent would do. For Shopify merchants, the highest-value flows to build are for your highest-volume, most structured scenarios: order status lookup, return initiation, refund status check, and discount code application. Each well-built flow directly reduces handling time, increases first-contact resolution, and creates a consistent brand experience.
Decision Tree Bot
A decision tree bot is a rule-based chatbot that guides conversations through a predefined hierarchy of branching choices — presenting customers with button options at each step and following the corresponding scripted path — rather than interpreting free-form natural language.
Decision tree bots were the dominant chatbot architecture through the 2010s and remain common today in simple deployments. The merchant pre-authors a tree of conversation branches: the customer is presented with a set of options ('I have a question about: [My Order] [Returns] [Products] [Other]'), selects one, is presented with sub-options, and progresses until they reach a pre-written answer or a human escalation. Decision tree bots are predictable, auditable, and completely within the merchant's control — every path the customer can take was authored in advance. Their limitations are equally predictable: they fail when customers want to ask something not in the tree, frustrate customers who must navigate multiple levels of menus to find their answer, and require significant ongoing maintenance as policies and scenarios evolve. In ecommerce, decision tree bots are a reasonable starting point for very high-volume, simple scenarios, but they cannot scale to handle the full range of customer questions without becoming unmanageably large.
Understanding where decision tree bots fall short helps Shopify merchants make better architecture decisions. A decision tree bot can deflect a high percentage of simple, anticipated queries at low cost — but it will also fail publicly and visibly when customers ask anything unanticipated. The frustration of being forced through irrelevant menu options or told 'I didn't understand that' repeatedly drives customer abandonment and negative sentiment. For stores with even moderate question variety, the step up to AI-powered NLU dramatically outperforms a well-maintained decision tree on customer satisfaction.
Rules-Based Chatbot
A rules-based chatbot is a conversational system that responds to customer inputs by matching them against a set of predefined rules — keyword triggers, button selections, or pattern matches — and returning a pre-authored response, without any machine learning or natural language understanding.
Rules-based chatbots operate on explicit if-then logic: if the customer's message contains the word 'return,' respond with the return policy. If the customer clicks the 'Track Order' button, show the order tracking prompt. Every response is pre-authored; the system matches inputs to rules and retrieves the associated output. There is no language understanding, no contextual reasoning, and no ability to handle inputs the rules don't anticipate. Rules-based chatbots dominated the first wave of ecommerce chatbot deployments because they were simple to build, predictable in behavior, and required no machine learning infrastructure. Their limitations are well-known: they fail on any phrasings not covered by the rule set, frustrate customers with irrelevant keyword matches, and require continuous manual maintenance as products and policies change. In the modern landscape, rules-based systems are typically used for very narrow, controlled scenarios (button menus, structured forms) rather than as the primary support AI.
Many Shopify merchants started with rules-based chatbots and continue using them because they're familiar and deployed. Understanding their limitations — and when they're causing measurable support failures — is the first step to making the case for AI-powered alternatives. The key signal is containment rate: if your rules-based bot is escalating 60–70% of conversations to humans, that's a signal the rules aren't covering actual customer queries, and an AI approach would likely contain a much higher percentage. The transition to AI is also operationally simpler than maintaining an ever-growing rule set.
Hybrid Chatbot
A hybrid chatbot is a conversational AI architecture that combines scripted rule-based flows for specific, high-control scenarios with AI-powered natural language understanding for open-ended or unpredictable customer queries — delivering the precision of rules where control matters and the flexibility of AI where variety is high.
Pure rule-based bots are predictable but brittle. Pure AI bots are flexible but less controllable. Hybrid chatbots combine both: they use pre-authored flows with explicit logic for scenarios where exact execution matters (processing a refund, confirming a cancellation, collecting required information in a defined order) and delegate to AI natural language understanding for everything that doesn't fit a predefined path. In practice, the hybrid architecture often looks like: the AI handles the initial intent classification and entity extraction; if the detected intent maps to a structured workflow, the dialogue manager hands off to a scripted flow; if the intent is informational or doesn't have a structured workflow, the AI generates a grounded response from the knowledge base. This architecture is the current best practice for production ecommerce support because it delivers both the reliability merchants need for consequential actions and the flexibility customers need for natural conversation.
For Shopify merchants, the hybrid model directly addresses the two biggest risks in chatbot deployment: a purely scripted bot that fails publicly on unanticipated questions, and a purely AI bot that occasionally does unexpected things in high-stakes workflows. The hybrid architecture uses scripted control where the stakes are high (return processing, refund authorization) and AI flexibility where coverage is the goal (general inquiries, product questions, policy lookups). Most of the market's best-performing ecommerce support tools are hybrid systems, even if they don't always advertise it as such.
How Bookbag helps
Visual Flow Builder
Bookbag's flow builder lets merchants map structured conversation paths for key support scenarios — returns, order lookups, discount applications — without writing code, with AI handling flexible language within the structured paths.
Tone and Persona Controls
Merchants set the chatbot's personality — formal, friendly, empathetic, concise — and Bookbag applies that voice consistently across all conversation types, ensuring the AI sounds like a natural extension of the brand.
Fallback and Escalation Design
Bookbag provides configurable fallback messages and escalation prompts, letting merchants craft the exact language for moments when the AI reaches its limits — ensuring those moments feel helpful rather than frustrating.
Go deeper
Guides & benchmarks
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.