Delivered · anonymised
An AI agent for WhatsApp orders, with a custom CRM and human handoff
An integrated WhatsApp, AI agent and custom CRM system that handles personalized wall-art order conversations in parallel, keeps order status visible and hands complex cases to the team with their context intact.
- Client
- Personalized wall-art business
- Industry
- Custom prints and e-commerce
- Loop
- Enquiries to orders
- What we built
- WhatsApp intake + bounded AI agent + live order CRM + human handoff
- Status
- Delivered · anonymised
The challenge
What was getting in the way
Each personalized artwork takes several messages to get right. With many WhatsApp conversations running at once, the team needed to collect order details, track progress and know when a customer needed help.
A personalized wall-art business manages a high volume of customer conversations through WhatsApp. Each order can require several exchanges before the request is clear, while the internal team still needs one reliable view of order status and a direct way to enter the conversation when a customer asks for help.
Order details and customer questions were being handled inside many simultaneous chat threads. Without a shared operational record, the team had to reconstruct context, track progress manually and decide when to step in. A chatbot alone would not solve that problem: the business needed parallel conversation handling, explicit permissions, live order visibility and a clean human takeover path.
- A WhatsApp message is linked to the correct conversation and order context.
- The AI agent responds in natural language, collects missing order details and advances requests that remain inside its approved scope.
- The custom CRM updates the order state in real time so the team can see progress across active conversations.
- When a customer requests a person or the issue exceeds the agent's permissions, the conversation moves to the escalation interface with its context preserved.
Approach
How the system works
We connected WhatsApp to an AI order assistant and a custom CRM. The assistant handles routine exchanges, while the team sees order status and takes over complex conversations with their context intact.
- 01
WhatsApp conversation layer
Incoming messages are connected to the right order context and remain available as part of the operational history.
- 02
Bounded order agent
The smart agent handles several order conversations in parallel, uses the approved business tone and escalates when the next action falls outside its permissions.
- 03
Custom CRM and takeover interface
A purpose-built workspace shows live order state, conversation ownership and the context the team needs to continue an escalated discussion.
Keep every order connected to its conversation
Rules handle
- Connect each WhatsApp event to the correct conversation and order record.
- Persist customer messages, agent replies, order state and conversation ownership.
- Synchronize approved order updates with the custom CRM in real time.
- Open a human-review item when a customer requests a person or a permission boundary is reached.
- Keep the handoff reason and conversation context visible to the assigned team member.
- Record whether the active conversation is owned by the agent or a person.
AI handles
- Respond in natural language that follows the business's approved tone and order information.
- Identify which details are still missing before an order can move forward.
- Maintain the relevant context while several customer conversations progress independently.
- Answer and act only within the permissions defined for the order-processing role.
- Explain the next step or escalate the conversation when the request is ambiguous, sensitive or out of scope.
People decide
- Define the product information, conversation rules and actions available to the AI agent.
- Take over whenever a customer asks to speak with a team member.
- Resolve complaints, unusual customization requests and issues outside the agent's permissions.
- Approve consequential order changes and decide when a resolved conversation can return to automated handling.
n8n workflows
From WhatsApp message to tracked order and escalation
Use Previous and Next to follow the anonymised journey. The map explains the delivered system without exposing client data or production configuration.
Delivered system · anonymised view
WhatsApp orders, CRM updates and human takeover
A public, sanitized reconstruction of the delivered journey. It shows system responsibilities without customer data, prompts, credentials, internal URLs or production configuration.- Match the message to the correct order
The WhatsApp event is linked to its conversation, order and approved business context before any reply is prepared.
- Continue inside the approved role
The agent prepares the next response, then a permission check decides whether the action may continue automatically.
- Update the order state in the CRM
Approved steps are communicated to the customer and recorded in the operational view used by the team.
- Move an exception to the team
A customer request or permission boundary creates a complete review item, notifies its owner and puts the conversation under human control.
Stack
Built with
Every tool the system connects, and the job it does.
WhatsApp
Customer conversation intake, outbound replies and continuity across the order discussion.
E-commerce order context
Approved product and order information used to understand the request and keep the conversation grounded.
Custom CRM
Real-time order status, conversation history, ownership and internal tracking for the operations team.
Escalation workspace
A focused interface for reviewing the reason, taking ownership and continuing the WhatsApp conversation.
Limits
Limits of this case
- The public system map is an anonymised reconstruction; customer data, prompts, credentials, internal URLs and production configuration are omitted.
- Agent responses depend on the completeness and freshness of the approved product, policy and order information available to it.
- Ambiguous personalization details, complaints and actions outside the defined permissions require a team member.
- Conversation and order synchronization depends on the continued availability of WhatsApp and the connected commerce systems.
- No response-time, conversion or order-volume metric is published because no approved measurement was provided for this case study.
Possible next phase
Use the CRM history to review where handoffs occur, refine the agent's permissions and improve the order states only where the recorded conversations support the change.
Next case study
