Conversational software systems that engage users through text or voice, answering questions, guiding experiences, and completing tasks without requiring human intervention for each exchange. For agencies, chatbots are a client deliverable, a media channel, and an operational tool, all three of which require different design disciplines.
Also known as conversational bots, chat agents, virtual assistants
Chatbots interact with users through natural language, interpreting inputs and generating responses either from predefined rules or from generative language models. Rule-based chatbots follow decision trees: if a user says X, respond with Y. They are predictable and easy to audit but fail when users phrase things unexpectedly. Large language model-powered chatbots generate responses dynamically, handling more varied inputs but requiring more careful guardrails to prevent off-topic or inaccurate outputs.
Most production chatbots combine both approaches: a language model for natural language understanding and flexible response generation, layered over structured logic that constrains what the bot can do, which data it can access, and when it escalates to a human agent.
Chatbots span a wide deployment range: a simple FAQ bot on a client’s contact page, a lead qualification bot in a campaign landing experience, an internal agency tool that retrieves brand guidelines on request, and a customer support system handling hundreds of concurrent sessions. Each requires a different design, different data access, and different success metrics.
Agencies encounter chatbots in three distinct roles. They build and recommend them for clients as customer-facing experiences. They manage media buys that interact with chatbot-driven landing pages. And they use them internally as productivity tools. Fluency in all three contexts is part of the modern agency skill set.
Brand voice extends to conversational interfaces. A chatbot deployed under a client’s brand speaks in that brand’s voice, makes commitments the brand is responsible for, and shapes the client’s reputation with every exchange. Agencies that treat chatbot copy as a technical deliverable rather than a brand deliverable produce experiences that feel off-brand even when they are technically functional. Brand voice guidelines must explicitly address conversational channels.
Failure modes are public and fast. When a chatbot produces an unhelpful, inaccurate, or inappropriate response, the user is already in the conversation. There is no editorial review between generation and delivery. Agencies deploying chatbots for clients need pre-launch coverage testing, clearly defined escalation paths, and a monitoring protocol, not just a QA pass before go-live.
LLM-powered chatbots require ongoing governance. A rule-based chatbot does what it was programmed to do. An LLM-powered chatbot can drift when the underlying model is updated, when the prompt context is manipulated, or when users discover edge cases the design team did not anticipate. Agencies recommending LLM chatbots to clients need to build governance and monitoring into the engagement, not just the launch.
An agency builds a lead qualification chatbot for a B2B software client’s campaign landing page. In the first two weeks, the bot handles 80% of interactions successfully. A review of the remaining 20% reveals two patterns: users asking detailed pricing questions the bot deflects poorly, and users from a specific industry vertical using terminology the bot does not recognize. The agency updates the escalation logic for pricing questions to connect directly to a sales calendar, and rewrites the intent classification training examples for the industry vertical. The fix requires understanding both the language model behavior and the business rules, which is why chatbot quality work is not a pure engineering task.
The automations and agents module of the workshop teaches you how to build AI workflows that compress the busywork without taking the craft out of the studio.