PROJECT

AI customer service for faster answers and lower support effort

Custom software development, Customer support chatbot, FAQ and knowledge-base management, Product information retrieval, Website chat widget, Business-system integrations, Conversation analytics

  • Industry: E-commerce
  • Spain Location: Spain
  • Delivery time: 10 weeks
  • Technologies: Python, JavaScript / TypeScript (Node.js)
3 months

This client spends support time answering repeated product, account, policy and order questions. The AI customer service software is an embedded chatbot for the retailer’s website: it reads product and support information, responds in natural language, and sends uncertain requests to human agents. Visitors use it on desktop or mobile, while the support team manages answers, connected data and conversation quality in one place.

Summary

Problem

An online retailer's support team spends time answering routine questions and redirecting complex inquiries.

Solution

The solution became a branded website chatbot connected to the retailer's knowledge sources and business systems.

Result

The software is designed to reduce the support team's manual workload while preserving access to a human for uncertain inquiries.

Client story

The client is an online retailer from Spain. It assists website visitors who need information about products, order tracking and support.

The retailer's support team works with product data, account information, FAQs, policies and technical instructions. Repetitive questions take up time, while unhandled or complex inquiries still require human attention.

The request was straightforward: connect the assistant to business data, embed it on the existing website, record conversation quality and prepare a documented production-ready solution.

The solution was built as four connected modules: a visitor chat widget, an editable knowledge area, integration and escalation settings, and an analytics dashboard. The stages included analysis, structuring, design and development, with presentations and feedback from the client's team.

The client asked for

  1. Connections to products, orders, CRM and other business services
  2. Tracking of conversations, usage, fallback responses, escalations and response quality
  3. Tested implementation guidance, source code and secure configuration
  4. Embedding a branded, responsive chat widget on the existing website

Why off-the-shelf software did not fit

Off-the-shelf AI chatbot products support conversational responses and integrations, but their descriptions also point to feature limitations, integration issues, AI inaccuracies or training complexity. This client needed one branded website assistant with editable commercial information, connections to business systems, escalation rules and analytics in a single solution.

Project overview

The solution was designed around the retailer's website, support content and business services. The embedded chatbot answers visitor questions about products, FAQs and policies.

To support changing information, an editable knowledge and response area was provided. Content managers can manage FAQs, product specifications and policy documents without changing the conversation interface.

Integration settings connect product and account databases, API endpoints for order tracking, and CRM or support systems. Escalation destinations include email, webhooks and live chat.

The solution was also designed to record conversation history, usage and frequently asked questions. It logs fallback responses, escalations and response quality for support analysis.

The chatbot uses natural language understanding and AI tools to generate responses, while Markdown, API and database sources provide a clear and maintainable knowledge structure. Production handover included documentation and secure API key management.

Key features delivered

  • Embedded branded chat widget for desktop and mobile websites
  • Natural language responses based on product, FAQ, policy and support content
  • Editable FAQ, product and knowledge base content
  • Connections to databases, APIs, CRM and support systems
  • Responses about order tracking and account information
  • Human escalation via email, webhooks or live chat
  • Conversation history, fallback responses and escalation logs
  • Analytics for usage, frequently asked questions and response quality

Can be added

+ Product publishing + Website data scraping + Response debugging + Dashboard integration + Additional support system integrations + Personalized multi-step conversations after authentication

Who else can use this solution

Businesses with a similar information and support task can adapt the same solution to their websites.

A customer-support team

Task It needs first-line automation while keeping complex requests with human agents.
How the solution helps The solution answers routine questions, records context and routes unsupported requests to human support.

A business website with FAQ and policy content

Task It needs to give visitors reliable answers from changing information.
How the solution helps The solution combines editable knowledge sources with a branded conversational website widget.

A digital-product and service business

Task It wants conversational guidance for leads, bookings or purchases.
How the solution helps The solution connects website conversations with relevant business data and escalation paths.

Technology stack

Python

Conversation and data services

Python supports the conversation logic and supporting data services.

JavaScript / TypeScript (Node.js)

Widget and API connections

JavaScript and TypeScript support the website widget and service connections.

PHP

Analytics and API services

PHP supports selected analytics and Google API connections.

rasa

Natural-language dialogue management

rasa (an open-source framework for understanding messages and managing dialogue) handles natural-language understanding and conversation flows.

rasa-sdk

Conversation integration

rasa-sdk supports the Rasa conversation framework used for natural-language understanding and dialogue management.

Dialogflow

Intent and language understanding

dialogflow (a Google conversational-AI service) supports intent detection and dialogue management.

OpenAI

AI-generated responses

openai (the official library for connecting to the OpenAI API) supports AI responses.

LangChain

Composable AI application flows

langchain (a toolkit for connecting language models with prompts, data and tools) links conversational steps and knowledge sources.

Markdown, APIs and database data

Knowledge and business information

Markdown, API data and database data provide the product, FAQ, policy, account and order information used by the chatbot.

Integrations

Product and account databases

The chatbot reads product details and account information from connected databases.

Order-tracking APIs

Order-tracking endpoints provide responses about the customer’s order status.

CRM and support systems

The chatbot connects to CRM platforms and customer-support systems when those integrations are required.

Email, webhooks and live chat

Unsupported requests are sent to an available human-support destination with conversation context where required.

Existing website

The branded widget is embedded into the website with a JavaScript snippet or iframe.

The result

Visitors get answers inside the website

The branded widget answers questions about products, FAQs and policies without sending visitors to a separate tool.

When the assistant lacks a suitable answer, the same conversation exposes a human-support handoff instead of ending at an unsupported reply.

How it works

  1. The widget receives visitor questions and shows contextual replies in the conversation area.
  2. The chatbot searches FAQ and product information to produce natural-language answers.
  3. The widget shows clarification or escalation messages and provides the human-support handoff action.
  • Responsive desktop and mobile presentation
  • Product, FAQ and support answers
  • Escalation messages for unsupported requests
Tip: The widget can be embedded with a JavaScript snippet or iframe.
Embedded customer chat widget

Embedded customer chat widget

Functions on this screen
Create Chatbot Functionality Chatbot Integration Service Search FAQs Answer questions

Content managers keep answers current

The knowledge editor gives the retailer one place for FAQs, product specifications, policies, manuals and reference data. Editable answers keep the assistant aligned with the information visitors receive.

The content structure was shaped around the retailer’s own sources, so product and support information remain manageable as the catalog and policies change.

How it works

  1. The editor lists knowledge sources and shows FAQ, product and policy content.
  2. The system stores product specifications, reference data and editable answers.
  3. Saved updates refresh the information used for FAQ and product responses.
  • Knowledge-source list
  • FAQ and product-information fields
  • Policy or manual content editor
Knowledge and answer editor

Knowledge and answer editor

Functions on this screen
Product FAQs Edit Knowledge Base Manage FAQs Efficiently

Business data and escalation paths stay connected

The integration screen brings product data, account information and order tracking into the conversation. It also defines where unsupported requests go, from email and webhooks to live chat or a support system.

Authentication settings support personalized multi-turn conversations when identity-provider data is available.

How it works

  1. The system connects database settings and API endpoints for product, account and order information.
  2. It exchanges customer details with CRM or support-system connections.
  3. It sends unsupported requests to configured email, webhook or live-chat destinations.
  • Database and API connection fields
  • CRM or support-system connection
  • Email, webhook and live-chat handoff
Note: Connected services depend on the retailer’s available APIs, databases and authentication data.
Integration and escalation settings

Integration and escalation settings

Functions on this screen
Integrate Existing Systems Integrate Customer Support Systems Transfer calls to live agents

Support managers see where the assistant needs attention

The analytics dashboard turns conversation records into a practical view of usage, common questions, fallbacks and handoffs. Response-quality information helps the team identify content or conversation rules that need tuning.

The dashboard also keeps conversation volume visible, so the support team can review how the assistant is being used.

How it works

  1. The dashboard shows usage analytics and conversation volume.
  2. The system groups common questions and displays fallback and handoff reports.
  3. The response-quality view presents question and answer quality analytics.
  • Usage summary
  • Common-question report
  • Fallback and handoff report
Conversation analytics dashboard

Conversation analytics dashboard

Functions on this screen
Chat Performance Analytics

What the solution gives

With this solution, the retailer can:

  • Answer product, account, FAQ, policy and support questions on the website
  • Keep product and support information editable in one knowledge area
  • Connect conversations with databases, APIs, CRM and support systems
  • Route uncertain requests to human agents with relevant context
  • Review common questions, fallbacks, handoffs and response quality
  • Manage a branded chat experience across desktop and mobile browsers

Could ready-made systems do the job?

Before the design, the task was compared with ready-made products by their own descriptions.

What the client needs AI chatbot software Google Cloud Dialogflow, Zoom Virtual Agent, Joule Conversational design tools Voiceflow, Letter AI Customer support automation Kapture CX Custom solution
Connect business data and services partly
Track conversations and quality partly
Provide documented production setup — — —
Embed branded responsive website widget — — —
Fits the client's own rules: stages, roles, reports — — —
All steps in one place — — —
No manual steps between programs — — —

By the public descriptions of the products and their category.

Ready-made products cover conversational answers, integrations and support automation. A ready-made product is enough when only one part of this task is needed.

This client needed its own software because the requirements combined editable retail knowledge, business-system connections, escalation rules, analytics and a branded website experience in one place. The ready-made options also left gaps around the client’s own rules and the connection between all steps.

In the end

In the end, the solution was made to take manual support work out of the path from a visitor’s question to a useful answer or human handoff.

FAQ

Thinking about building a similar system? Read the answers below to see how the chatbot, knowledge sources, integrations and support handoffs fit together.