Glossary

Custom Actions

Custom actions are operator-defined functions or API calls that extend an AI support agent's default action set, enabling the agent to perform operations specific to a particular store's systems — such as applying a loyalty discount, pausing a subscription, or triggering a custom fulfillment workflow — that aren't available in the platform's standard action library.

Also covered on this page: Function Calling, Tool Use.

Book a demo See pricing

What it means

Key insight

Custom actions are how you teach your AI the things only your store does — the operations that make you distinct from every other ecommerce brand.

Every AI support platform ships with a set of standard actions: look up an order, initiate a return, process a refund. These cover the common cases for most Shopify stores. But many brands have support workflows that go beyond the standard set: applying store-specific loyalty credits, pausing or modifying a subscription managed by a third-party platform, triggering a custom reshipment workflow, applying a brand-specific goodwill coupon. Custom actions let merchants define these operations as callable functions that the AI can invoke based on conversation context, extending the agent's resolution capabilities to match the brand's specific fulfillment and customer service model. Implementation typically involves defining the action as a function call with specified parameters, connecting it to the relevant API endpoint, and configuring the conditions and permissions under which the AI is allowed to invoke it.

Why it matters

The most impactful AI support deployments are ones where the AI can resolve the full range of issues customers actually bring — not just the standard subset. For Shopify stores with subscription products, loyalty programs, or custom fulfillment workflows, custom actions are what enable the AI to handle those interactions end-to-end rather than always escalating them to a human. Each custom action added expands the AI's resolution scope, directly increasing the auto-resolution rate and reducing human ticket volume.

Related concepts, explained

These terms are part of the same idea, so they live here rather than on pages of their own.

Function Calling

Function calling is an AI model capability that allows a large language model to recognize when an external tool or API call is needed to fulfill a request, generate a structured call specification for that function with the appropriate parameters, and incorporate the function's return value into its response.

Without function calling, a large language model can only generate text — it cannot fetch live data or execute operations in external systems. Function calling changes this by giving the model a declared set of available functions (each described with a name, purpose, and parameter schema) and the ability to signal "I need to call this function with these arguments" rather than generating a text response directly. The calling application then executes the actual function, returns the result to the model, and the model incorporates that real data into its final response. For ecommerce support, function calling is what enables an AI to actually look up a customer's real order status (not generate a plausible-sounding one), check actual inventory levels, and initiate real refund transactions. It's the bridge between the AI's language reasoning and real-world systems, and it is foundational to the difference between a chatbot and an action-capable AI agent.

Function calling is why modern AI support can resolve tickets rather than just respond to them. A model without function calling can only discuss what it would do if it had access to systems — "I would look up your order" — while a function-calling-enabled model actually does it. For Shopify merchants, this matters directly: every support function the AI can call (order lookup, return initiation, refund processing) is a category of ticket that doesn't require a human to handle, which directly translates to reduced support cost and faster customer resolution times.

Tool Use

Tool use is the capability of an AI agent to select and invoke external tools — such as search engines, calculators, databases, or APIs — during the process of fulfilling a request, extending the agent's abilities beyond what its underlying language model can accomplish with text generation alone.

Tool use is the general capability of which function calling is a specific implementation. An AI agent equipped with tools can reach outside its own parameters to access real-time information, perform computations, or execute operations in external systems — all capabilities that text generation alone cannot provide. In ecommerce support, the most important tools are: order lookup (retrieving live order status from Shopify), knowledge retrieval (searching the knowledge base for relevant policies), carrier tracking (fetching shipment status from carriers), and action tools (initiating returns, processing refunds). Each tool the agent can use expands the range of issues it can resolve. Tool selection is itself an AI reasoning task: the agent must determine which tool is appropriate for the current step in the conversation, what parameters to pass, and how to interpret the result. This reasoning layer is what makes tool-using agents powerful and also where the complexity lies — poorly designed tool use results in agents that call tools unnecessarily, with wrong parameters, or that misinterpret results.

Tool use is what makes AI support operational rather than merely conversational. The most common customer support issues in ecommerce — "where is my order," "can I return this," "why was I charged this amount" — all require access to live data that the AI's training doesn't contain. Tool use is the mechanism that makes these resolvable by AI rather than requiring human lookup. Brands that deploy agents with deep tool access handle a fundamentally different class of customer issues autonomously than brands running knowledge-base-only chatbots.

How Bookbag helps

Custom Function Definition

Bookbag allows merchants to define custom callable functions — with specified inputs, API endpoints, and parameter mappings — that the AI can invoke when the appropriate customer scenario arises.

Permission-Gated Execution

Custom actions in Bookbag are configured with execution permissions — some can run autonomously, others require a human confirmation step — so high-stakes operations always have appropriate oversight.

Third-Party Platform Actions

Bookbag supports custom actions that call third-party platforms — subscription management tools, loyalty platforms, custom ERPs — extending AI resolution to cover the full ecommerce tech stack.

Go deeper

Guides & benchmarks

See it in the product

Frequently Asked Questions

Defining custom actions typically requires some technical configuration — specifying API endpoints, parameter schemas, and authentication. However, once defined, support managers can configure which scenarios trigger them without further developer involvement.

The AI uses intent detection and conversation context to determine which action is appropriate. Custom actions are described to the AI with natural language documentation about what they do and when to use them, enabling the AI to match situations to actions correctly.

Custom actions should be configured with execution conditions (what context must be true for the action to fire), parameter validation (ensuring input values are valid before calling the API), and rollback or confirmation requirements for irreversible operations.

A regular API integration is a hard-coded connection where a specific event always calls a specific endpoint. Function calling is dynamic — the AI model decides at inference time whether a function call is needed, which function to call, and what parameters to pass based on the conversation context.

The agent reasons from the conversation context and its goal. Each tool is described with a natural language specification of what it does and when it's appropriate. The agent selects the tool whose description best matches the current need — this is the same reasoning process it uses to select words in generating a response.

See Bookbag in action

Join the ecommerce teams resolving more tickets, answering 24/7, and turning support into a revenue channel with Bookbag.