AI automation
AI Automation and Business Process Automation
Dfeelings designs and builds automated workflows that move data between your systems, use AI to read, sort and draft, and keep a person in charge of the decisions that matter. Common work includes qualifying leads into your CRM, handling routine customer inquiries, syncing Shopify orders and stock, preparing business reports and extracting data from documents.
We build in n8n or the platform that fits your stack, connect the tools you already use through their APIs, and work in Arabic and English for businesses in Saudi Arabia, Jordan, the UAE, Oman, the wider Gulf and the United States. AI in these workflows can make mistakes, so every workflow is designed with checks, approval points and logs.
Who this service is for
Sales teams
Teams losing leads somewhere between web forms, WhatsApp, email and the CRM.
Support and operations teams
Teams answering the same questions and copying the same data every day.
Shopify stores
eCommerce businesses that still handle orders, stock and customer updates by hand.
Managers
People who rebuild the same weekly report from several spreadsheets and dashboards.
Careful AI adopters
Companies that want AI in daily work but need control over data, approvals and errors.
Where time is lost
Many delays in a growing business are not hard problems. They are handoffs: a lead waiting in an inbox, an order typed again into another system, a report rebuilt by hand every week, an invoice read line by line.
General AI chat tools can draft and summarize, but on their own they are not connected to your systems, not checked against your rules and not accountable for the result. Business process automation joins those steps into one workflow. AI handles the parts that need reading or classification, and a person approves anything that affects a customer, money or an official record.
What we automate
Each capability can be built on its own or combined into one workflow.
AI workflow automation with n8n
Workflows started by a form, a message, a schedule or a change in another system, built with clear steps, error handling and a record of every run.
CRM automation and lead qualification
Leads from forms, ads, WhatsApp and email captured in one CRM, scored against criteria your sales team sets, routed to the right person and followed up on time. AI can summarize each inquiry and suggest a score; your team confirms it.
Customer inquiry automation
Routine questions about orders, opening hours or published services answered from information you have approved, with anything unclear passed to a person along with the full conversation. WhatsApp flows run on the WhatsApp Business API.
API integrations
Systems that do not talk to each other connected through their APIs or webhooks: CRMs, ERPs, accounting tools, Google Workspace, Microsoft 365, helpdesks, eCommerce platforms and your own databases.
Shopify automation
Order routing, stock and price updates, low stock alerts, abandoned cart reminders, order status messages, and order data sent to accounting or a warehouse.
Reporting automation
Data collected on a schedule from sales, marketing, eCommerce and support tools into one report, with a short AI summary of what changed that a manager reviews before it is shared.
Document processing
Invoices, purchase orders, forms and contracts read by AI, key fields extracted into a structured record and checked against rules, with anything uncertain or mismatched sent to a person.
Human supervised AI agents
Agents that look up information and take a defined set of actions inside your tools, with approval before sensitive actions, limits on what they can access, and a log of each action. They support your team's judgement; they do not replace it.
Typical scenarios
These scenarios show how the pieces fit together. They illustrate the kind of workflows we design; they are not completed client projects or measured results.
Lead qualification for a B2B supplier
A quote request arrives through the website form or WhatsApp. The workflow creates or updates the contact in the CRM, AI extracts the product, quantity, city and deadline, the lead is scored against the sales team's criteria, and it is assigned to a salesperson with a short summary.
Where a person decides: The salesperson confirms the score and sends the first reply.
Order updates for a Shopify store
When an order is paid, shipped or delayed, the workflow updates the warehouse sheet, sends the customer an approved WhatsApp or email template in their language, and flags any order that has not shipped within the agreed time.
Where a person decides: Operations staff handle every flagged order.
Inquiry triage for a service business
Messages from the website, email and social media inboxes are collected in one queue. AI classifies each one, drafts an answer from the approved FAQ and service information, and marks anything that information does not cover.
Where a person decides: A team member reviews and sends each draft. By default, complaints and unusual requests are routed to a person, as confirmed in the proposal.
Supplier invoice intake
Invoices received by email are read, the supplier, amounts, tax and due date are extracted and matched against the purchase order, and a draft entry is prepared for the accounting system.
Where a person decides: Finance approves each entry. By default, nothing is posted without review, and mismatches are held for Finance.
Weekly management report
Each week, data from the CRM, store or website analytics and ad accounts is collected into one report, with a short summary of changes against the previous week.
Where a person decides: The manager reviews the summary before it is shared.
How an automated workflow works
Every workflow is built from the same parts, so it can be understood, tested and changed later.
Trigger
The event that starts the workflow: a form submission, a new message, a webhook from another system, a new file or a schedule.
Integrations
Connections to your tools through their official APIs or webhooks, each with its own credentials and only the permissions it needs.
Rules and data mapping
Routing, validation and formatting written as explicit steps rather than left to the AI.
AI steps
A language model reads, classifies, extracts or drafts, following instructions and examples from your business, and returns structured output the workflow checks before it continues. The model provider is agreed with you.
Human approval
Points where the workflow pauses until a person approves, edits or rejects the result, for example in email, Slack, Microsoft Teams or the CRM.
Logging and error handling
Runs are logged. A failed step retries where that is safe; otherwise it alerts a named person, so failures do not go unnoticed.
Documentation and change control
Workflows are documented and versioned, and changes are tested before they replace the live version.
Why n8n
n8n is a workflow automation platform with a visual editor, a large library of ready connectors and room for custom code where no connector exists. It can run in n8n's cloud or on infrastructure you control, which matters when data has to stay in a particular place.
We use it when it fits. If your business already runs another automation platform, or a task is better served by custom code, we use that instead.
Data handling, security and human oversight
Automation touches customer data and business records, so access, data flows and approval points are agreed in writing before anything is built. The points below describe how we design workflows; the exact arrangements for your project are set out in the proposal.
Limited access
Each workflow is given its own credentials with only the permissions it needs, kept in the automation platform's credential settings rather than in documents or chats.
Only the data a step needs
Workflows are designed to send an AI model only the fields a step requires, and to remove or mask personal data the task does not need.
AI providers you approve
Before customer data reaches an AI model, we go through the provider's terms on data use, retention and processing location with you, and use only providers and plans you have approved.
Where data is processed
When data must stay on infrastructure you control or in a particular region, n8n can be hosted there, and AI steps use providers or models that meet the same requirement.
Approvals and audit trail
Actions that affect customers, money or official records wait for a person's approval, and runs and approvals are logged.
Your accounts, your control
Automation platform, API and AI provider accounts are opened in your name wherever possible, and you can remove our access at any time.
Legal obligations
Data protection laws in the countries where you operate can apply to automated processing. We design to the requirements your legal adviser confirms; we do not give legal advice.
Costs billed by platforms and AI providers
Automation platform plans, AI model usage and any hosting are billed by those providers, separately from Dfeelings fees. Each proposal lists them and explains what drives AI usage costs.
Limits we are open about
Not fully autonomous
We do not build workflows that run your business unattended. People stay responsible for decisions that affect customers, money and records.
AI makes mistakes
Language models can misread, invent or misclassify. Workflows are designed so errors are caught by rules or by a person before they reach a customer or a record.
No promised savings
We do not promise time or cost savings in advance. We record how the process runs before automation, so any change can be measured afterwards.
Not every process should be automated
If a task is rare, unclear or needs fresh judgement each time, we will say so.
How an automation project runs
Process mapping
We map the current process with the people who run it: steps, systems, data, exceptions and where time is lost.
You receive: A process map and a list of automation candidates
Prioritization
We choose the workflows where automation is reliable and worth the effort, and decide where people stay in the loop.
You receive: A scope with every approval point marked
Data and access review
We confirm which data each step needs, where it is stored and processed, and which accounts and permissions are required.
You receive: A data flow and access list
Build and test
We build in a test environment with sample or anonymized data, including failure cases and both Arabic and English inputs.
You receive: A tested workflow
Supervised pilot
The workflow runs on real work with a person checking its output before it takes effect. Checks are reduced only where results stay consistently correct, and only with your agreement.
You receive: A pilot review
Launch and handover
The workflow goes live with logging and alerts, documentation and training for your team, and an agreed support scope.
You receive: Documentation and a handover session
AI automation: frequently asked questions
What is business process automation?
Business process automation uses software to carry out the repeatable steps of a process, such as moving data between systems, sending notifications and creating records, so people spend their time on the steps that need judgement. AI automation adds language models that can read, classify and draft text, which extends automation to emails, messages and documents.
What is n8n and why do you use it?
n8n is a workflow automation platform. Workflows are built visually and can include custom code, it connects to many business tools and APIs, and it can run in n8n's cloud or on a server you control. That last option helps when data must stay in a specific place. We use other platforms or custom code when they fit better.
Will AI reply to our customers on its own?
Only where you decide it should, and only from information you have approved, such as published opening hours or order status. Everything else is drafted for a person to review, and by default complaints or unusual requests are routed to your team.
How is our data handled?
Before anything is built, we agree with you which data each workflow uses, which AI providers process it, where it is processed and who can approve actions. Workflows are designed with limited permissions, minimal data sent to AI models and a log of each run. The proposal records these arrangements for your project.
Does AI automation work in Arabic?
Yes, with limits. Current language models read and write Arabic, but quality varies by task and dialect. We test each AI step on real Arabic and English samples from your business before launch and keep a review step wherever accuracy matters.
How much does an automation project cost?
It depends on the number of workflows, systems and AI steps, how much human review is needed, and the support you want after launch. Platform, AI usage and hosting fees are billed by those providers. We send a written proposal after the process is mapped.
What happens when a workflow fails?
A failed step retries where that is safe, then alerts a named person with the details. Workflows are designed to stop cleanly rather than create partial or duplicate records.
Start here
Start with one workflow
Tell us which process takes the most manual time, which systems it touches and who approves the result. We reply with the next step and, once the scope is clear, a written proposal before any commitment.
- Built for businesses in Saudi Arabia and Jordan
- Arabic and English, measured separately
- A written proposal before any commitment
Prefer to talk first?