.svg)
From Answers to Actions: How an AI Assistant Expands a CS-Cart Store’s Capabilities
An AI assistant understands a customer’s request, searches the knowledge base and catalog, and can interact with business systems through configured APIs.
Customers do not visit an online store for its catalog or filters. They need to choose a product, understand the terms of purchase, or solve a specific problem.
However, visitors do not always know where to look for information or how to phrase a query correctly. They may not remember a product name, understand its specifications, or find an answer about delivery and returns. The more complex the assortment and site structure, the more steps stand between the customer and the result.
iisy is a platform for creating managed AI workflows. It connects a language model to a knowledge base, catalog, and external systems, while the business defines the sequence of steps and permitted actions.
iisy adds a conversational interface to a CS-Cart store. The customer asks a question in natural language, and the assistant determines what is needed: an answer from the knowledge base, a selection from the catalog, or an action in an external system.
This is not an unrestricted AI chat that responds at its own discretion. The store defines the information sources, workflow rules, and actions the assistant is allowed to perform.
Navigation
One Conversation Instead of Searching the Entire Site
Customers usually have to use several separate tools. Products are found in the catalog, delivery terms in the help section, compatibility in the specifications, and requests are submitted through a form.
A conversation brings these tasks together in one place. Depending on the request, the assistant can:
- find an exact answer in the store’s materials;
- clarify the customer’s needs and select products;
- contact an external system;
- add the selected product to the cart;
- suggest the next predefined action.
The customer does not need to understand the site structure or guess the right keywords. They simply describe the task in their own words.
Knowledge Base
Answer: A Knowledge Base for the AI Assistant
The store owner can upload a knowledge base to iisy: delivery and payment policies, return terms, warranty information, instructions, service descriptions, and other materials.
When a customer asks a question, AI recognizes its meaning and converts the wording into a query it can process. The answer is found in the store’s knowledge base rather than generated from the language model’s general knowledge.
For example, a customer asks:
“Can I return the product if the size does not fit?”
The assistant understands that the question concerns returns, finds the relevant policy, and answers using the business owner’s materials.
It works like a familiar FAQ, but the customer does not have to guess the section name or browse a list of questions. They can use natural wording, synonyms, or describe the situation in their own words.
For the business, this means control over the answer. If the policies change, the owner updates the knowledge base and the assistant uses the new information.
Product Selection
Select: From the Customer’s Task to Products in the Catalog
Standard search works well when a person knows the exact product name. But real queries often sound different:
“I need a lightweight drill for concrete under $200.”
“I’m looking for shoes for long walks, but not sports shoes.”
“What would work in a small bathroom without complicated installation?”
These queries describe a task, not a product. Simply matching words to names and categories may return results that are too broad or irrelevant.
The AI assistant first analyzes the request and asks a clarifying question if necessary. The workflow then queries the CS-Cart catalog and uses the available data: category, brand, price, stock, size, color, material, purpose, specifications, compatibility, images, and product-page links.
In a tool store, the conversation might look like this:
— I need a drill for concrete, but it must be lightweight and cost no more than $200.
— Will you use it occasionally at home or regularly for professional work?
— At home, a few times a month.
After clarifying the need, the assistant can show a suitable model, a more affordable alternative, and a more powerful option, briefly explain the differences, and provide links to the product pages.
The quality of the selection depends on the structure and completeness of the catalog. The more accurately the specifications are filled in, the more parameters the workflow can take into account.
Actions
Take Action: From Advice to Results
The assistant’s value does not end with an answer or recommendation. Through APIs, it can transfer data and trigger actions in store systems.
Adding a selected product to the cart is already implemented: the customer can move from a recommendation to an action directly in the conversation.
Other workflows are configured for the needs of a specific business. HTTP requests can connect iisy to a CRM, order management system, warehouse, spreadsheets, internal services, calculation modules, and third-party applications.
Once the relevant API is connected and the rules are configured, the assistant can:
- check an order status;
- verify stock availability;
- calculate a price or product bundle;
- send an inquiry to the CRM;
- create a request;
- place an order in the conversation;
- start another process in an external system.
Workflows
Workflows Tailored to the Store’s Needs
Product selection is just one example. The same platform can be used to build:
- an assistant for delivery, payment, returns, and warranties;
- an adviser on specifications and compatibility;
- a product configurator;
- a price or bundle calculator;
- request submission;
- order status checks;
- a workflow with authentication;
- an assistant for marketplace sellers;
- an internal tool for administrators;
- a sequence of actions through external APIs.
A workflow can use only the knowledge base, combine it with the catalog, or connect external services. Its capabilities are defined by the store’s needs, not by the rigid logic of a universal chatbot.
Integration
How It Works
The integration is implemented as a separate CS-Cart add-on. It connects the storefront to an iisy agent, while the workflow and connected resources are managed centrally.
The owner defines which data the assistant can access: the catalog, knowledge base, predefined HTTP requests, and other tools. AI understands the customer’s natural-language request, while the managed workflow decides which source to use and what to do next.
The process consists of individual steps. A workflow can ask a question, save the answer, check a condition, call an API, display a result, or start an AI agent. Some stages run without a language model, using rules, variables, buttons, and external requests.
The business logic is not embedded entirely in the client add-on. The workflow can be changed centrally without rewriting and reinstalling the add-on in every store.
Control
The Business Remains in Control
A language model is good at understanding natural language, but it should not define the business process on its own.
In iisy, the store specifies:
- required questions and data;
- information sources;
- conditions for moving between stages;
- available tools;
- permitted actions;
- fallback options in case of an error.
AI is used where human language must be understood, data must be processed, or an available tool must be selected. The workflow controls the sequence and state of the process.
This is especially important when working with a knowledge base and external systems. The assistant answers from connected materials and performs only the actions defined by the owner.
Channels
One Workflow Across Multiple Channels
A website widget, Telegram, MAX, and an internal testing channel can all be connected to one iisy agent. The logic, conditions, and resources remain shared.
The widget’s appearance is configured separately: the launch button, window, header, input field, legal links, and custom styles can all be changed.
Adding a channel does not require rebuilding the entire process. You only need to check the specifics of the new interface.
Analytics
Analytics Shows Where Customers Drop Off
Chat history shows the messages. iisy analytics makes it possible to reconstruct the entire journey:
- which channel the user came from;
- which steps the workflow completed;
- which resources were called;
- which branch the conversation followed;
- where an error occurred;
- which action completed the process.
This data helps improve the workflow based on facts. If customers often stop after a particular question, that stage can be simplified. Frequent transitions to a fallback branch show where logic or data is missing. If one path leads to cart additions more often, it can become the primary path.
This turns the conversation into a measurable process rather than just another support channel.
Pilot
What to Test in a Pilot
The first launch does not need to automate the entire store. A pilot is used to test a limited set of tasks with real requests and data.
Depending on the business priorities, this might include:
- knowledge-base consultation;
- product selection in one category;
- a combination of consultation and product search;
- adding the selected product to the cart;
- one workflow connected to an external system.
Before launch, define the metrics that will be used to evaluate the result:
- time from the first request to an answer or product;
- number of steps to the result;
- share of completed workflows;
- frequency of fallback branches;
- visits to product pages;
- cart additions;
- other target actions;
- paid orders.
| Metric | Standard Journey | iisy Workflow | Pilot Target |
|---|---|---|---|
| First response | Depends on navigation and user actions | The workflow starts immediately | Up to 3 seconds for the first step |
| Finding a suitable product | Categories, filters, and product-page comparisons | Conversation and clarification of the task | 1–3 minutes |
| Actions before the result | 6–12 transitions | 2–5 meaningful steps | 2–3 times shorter |
| Suitable options | A long list or a single product found | A short selection | 3–5 products |
| Completed workflows | Usually not measured | Visible in analytics | 60–75% of launches |
| Target action | Not linked to the original request | Recorded during the process | 10–18% of launches |
| Fallback branch | Empty or irrelevant results | Clarification or an alternative path | No more than 8–12% |
| Changing the process | Interface or code development | Draft, test, publish | Within one business day |
| New channel | Separate implementation | Connection to the existing workflow | No need to rebuild the logic |
Economics
Example Financial Model
Suppose the workflow is launched 1,000 times per month:
- 70% of users reach a result;
- 15% of launches end with a target action;
- 25% of target actions convert into a paid order;
- the average order value is $100.
The model gives us:
1,000 × 15% = 150 target actions.
150 × 25% = 37–38 orders.
38 × $100 = $3,800 in revenue associated with the workflow.
This amount is not guaranteed incremental revenue. Some customers might have placed an order without the assistant.
If we assume that only 25% of the calculated revenue resulted from the shorter, clearer journey:
$3,800 × 25% = $950 in additional monthly revenue.
This is an example model, not a forecast. After the pilot, the initial assumptions should be replaced with actual store data: traffic, conversion rate, average order value, and the number of paid orders.