AI reduced refund resolution time to under 1 minute.
This is a customer support chatbot adapted to the kalkhoff.com website — a fully automated flow that solves 73% of tickets without human input.
This AI assistant is live and demonstrates automated customer support
Instructions
Read before using the demo:
This demo uses mock customer records.
To test it without sharing real customer data, customer verification is simplified: the agent will accept any real email and phone number you own, so it can send you notifications during the demo.
How to test the demo:
To try order status checks, billing updates, or refund requests, use the test data below.
Order ID: 1236f
Name: John
Last Name: Smith
Email:YOUR OWN
Phone:YOUR OWN
Your contact details are only used to send demo notifications and aren't stored.
The problem it solves
/// Manual management of customer support tickets is time consuming, repeatable, and inefficient. During very busy periods it takes away the attention of your staff and limits your flexibility.
This is a kind of task that does not bring much benefit but prevents many problems so it has to be done regardless.
Specific issues targeted
Low concurrency of requests - human support can service only a limited amount of customers. Especially during sales seasons. AI resolves that issue.
Order status lookups consuming 60%+ of support time - your staff can focus on serious issues while AI handles all of the essentials swimmingly.
High costs of employing people - AI does not sleep, take breaks or call out sick. It works and supports you business 24/7.
Refunds taking too long - resolving a refund request manually can take ages — contact with the customer and collecting the right data requires time. AI can collect the right data directly from the customer within a couple of minutes and pass it onto the team.
Billing requests overload - Invoice lookup, VAT receipts, payment confirmations. AI handles it automatically with the highest security standards in place.
Slow or inconsistent replies - this damages the conversation and leads to unresolved tickets. AI answers instantly and on point 24/7.
Lack of real time order visibility - if your store's ecosystem consist of different tracking systems, multiple suppliers and fulfillment partners this can cause a headache when looking up information for customers. The production agent unifies all order data. The demo uses mock data, but the integration layer is ready for real systems.
Multilingual support - AI just does that. It speaks almost all languages in the world while people maybe 3 at the time.
Human errors - they happen. People misread policies, approve refunds they shouldn't, or send incorrect information. The agent enforces policies consistently and removes the risk of accidental over-refunds or miscommunication.
The solution
/// We created a modular customer support AI agent that uses Voiceflow's conversational capabilities and our own backend logic. Fully automated and conversational solution.
Legend (icons and core components explained)
///
- • Conversational logic (Voiceflow)
- • Dynamic order/refund/billing lookup
- • Policy-aware answers (RAG)
- • Human handoff flow
- • Analytics and logging
Smart Query Building
/// Understands user intent and routes requests to refund, billing, or troubleshooting workflows automatically.
Policy-Aware Answers
/// All answers are grounded in live refund and billing policies through retrieval-augmented generation (RAG).
The agent always follows the rules, which decreases mistakes and improves accuracy.
Real API Calls
/// The agent can be integrated to any system you are relying on and freely read the correct data to assist your customers.
This demo uses a mock database and verification services to simulate real life conditions.
Instant Handoff to Human
/// The agent can escalate to human when user's confidence drops, or user directly requests it.
AI can accurately judge user's frustration level, and offer to escalate automatically.
Metrics Logging
/// All conversations and data is tracked and stored. You can look at everything you need whenever you need. Additionally this data can be use to further improve and develop the system.
Performance
Behind the Scenes
/// This demo was designed using modular automation principles — every step can be customized or replaced.
Steps we've taken during the development:
Defining scope: 4 core functionalities in that the agent must fulfill.
Setting constraints: defining what agent must and must not do to comply with the highest safety standards.
Designing conversational logic in Voiceflow
Integrating the conversational logic with the database and identity verification tools via n8n
Creating analytics dashboard and connecting it to the agent
Deploying custom user interface.
What we learned
/// Our conclusions and key points for the future after building this demo
Consistency across different module's architecture is key.
The more consistent the database responses are in terms of structure, the easier it is for AI to follow the rules and policies. Consistent data means consistent prompting, which gives you consistent behaviour.
AI should not be given the responsibility of resolving sensitive issues, like changing billing data or resolving a refund on its own.
While we can make the agent super accurate, the probability of a mistake are not zero. Therefore AI has to assist people in high sensitivity cases within safe limits after high standard authentication.
Structured logic > conversation experience.
It is great when you can chat to AI like to a real human being, however in sensitive flows this might lead to missing key details. Therefore, we designed the agent to guide the user strictly through the processes to ensure accuracy, while leaving the agent conversational when the conditions are right.
Want this for your business?
/// We can deploy a custom support agent that is integrated to your systems and policies in less than 2 weeks.