Let the model understand the request. Let application rules decide what it may do.
The model handles language. The application still owns permissions, confirmation, tenant separation and whether an action really succeeded.
Nexfic · ADR Group
Customer conversations that start from your approved business information.
Nexfic is a business-support AI platform. It looks up answers from the organisation’s approved websites, documents and help content, remembers what the customer is talking about, and hands the conversation to a person when an exception or business action needs human ownership.
Knowledge kept separate by customerContextual conversationsLead captureHuman handoffConversation insights
Customer conversationApproved knowledge connected
Where is my order and can I exchange the size after delivery?
I can help with order tracking and post-purchase questions. Your order is in transit and scheduled for delivery tomorrow. If the size does not fit, I can explain the exchange policy and prepare the request. Grounded in approved order, shipping and returns knowledge
Order trackingShipping policyReturns & exchanges
Can I change the delivery address if the parcel has not yet arrived?
Yes. I can capture the updated address and, if the shipment is already with the courier, route the case to a support agent with the full conversation context. Case capture + human handoff
Ask a follow-up…
Why Nexfic
Stop writing a new bot rule for every way customers ask the same question.
Customers do not use the exact words a bot designer expects. Nexfic uses the language model to understand the request, then retrieves relevant business information. Actions such as changing an order or submitting a request stay behind explicit workflow rules.
Approved knowledge
Use the organisation’s approved website pages, documents and FAQs as the source material for answers.
Conversation memory
If the customer asks “what about returns?” after discussing an order, keep the order context instead of starting over.
Human judgement
Hand off when the assistant is uncertain or the customer needs an exception, approval or action.
Learning loop
See repeated questions and the places where the available business information is incomplete.
Platform architecture
What happens between the question and the next step.
The flow is deliberately split: receive the question, understand the conversation, retrieve the relevant source, then answer or hand off. Business actions sit behind a separate workflow.
Customer channel
Webchat is the primary customer-facing channel today, with the architecture designed to extend to additional channels.
Conversation intelligence
Interpret intent, follow-up questions and conversation state instead of matching isolated phrases.
Organisation knowledge
Retrieve only from the approved sources assigned to that customer organisation.
Response or handoff
Answer grounded information requests or route the conversation to a person when needed.
Information requests: AI can answer from sufficient approved evidence. Business actions: confirmation and successful execution must happen before saying the action is complete.
Capabilities
The customer chat is only one part of the product.
Behind the chat, teams need to manage knowledge, review handoffs, capture leads and understand what customers are asking.
Knowledge & RAG
Retrieve relevant business content from approved tenant knowledge before composing the answer.
Grounded informationConversation context
Maintain topic continuity so follow-up questions do not start from zero.
Natural dialogueLead capture
Collect contact details and relevant enquiry context when the visitor is ready for business follow-up.
Structured next stepAgent handoff
Move a conversation to a person with the intent and conversation history available for follow-up.
Human controlConversation analytics
Understand recurring questions, conversation themes and knowledge gaps.
Operational insightCustomer separation
Keep each organisation’s knowledge and conversation context separate from every other customer.
Tenant isolationBusiness use cases
Pick a conversation that already consumes time.
Starting with one focused use case is usually more useful than trying to automate every customer interaction at once, especially in ecommerce and service environments with repeatable questions.
Customer supportAnswer repeatable product, service and business-information questions from approved knowledge.
Ecommerce supportHandle order tracking, delivery questions, exchange policies and return guidance from approved knowledge.
Sales enquiriesExplain offerings, qualify intent and collect the information required for human follow-up.
Enterprise knowledgeMake approved internal or customer-facing knowledge easier to find through a conversational interface.
Enterprise controls
The assistant should know when it has reached its limit.
In production, three boundaries should be obvious: which sources the system may use, which actions it may take, and when a person owns the next step.
Tenant isolation
Knowledge and conversation context remain scoped to the appropriate organisation.
Controlled actions
Security-sensitive actions and completion claims require explicit workflow control.
Audit & observability
Design the operating layer so conversations, handoffs and system behaviour can be reviewed.
AI cost awareness
Model usage, context size and operating cost should be measurable as the platform scales.
See it with your business knowledge
Bring one real customer conversation.
For a useful demo, bring the questions customers actually ask, the knowledge your team uses to answer them and the points where human judgement matters.