A concept, not a delivered project: the pattern we would build for a UAE flower shop that takes orders by WhatsApp, phone and web form, and loses the customer somewhere between them.
An order comes in by WhatsApp, phone or a simple web form. Whoever is on shift takes a first name, maybe a phone number, and moves to the next chat. Staff turn over often and nobody is paid to keep a customer record, they are paid to get the order out the door. No last name. A digit wrong. No note of what the customer likes or dislikes. By the time the owner wants to run a promotion, the "customer list" is a phone book with no names attached and no way to tell who is a regular.
A repeat customer is treated like a stranger every time, because nothing in the shop remembers the last order. There is no "same as last time," no acknowledgement that this number has ordered from the shop for years. The owner cannot market to people who have already proven they spend, because the proof is buried in chat history nobody reads. The cost is not the missing fields. It is the relationship the shop keeps forgetting.
Reads incoming WhatsApp messages, calls and order confirmations, cross references them against past chats and delivery details, and builds one record per customer instead of one fragment per order. Conflicts are flagged, never guessed silently.
A note like "no red flowers," mentioned once in a chat, or a delivery address used repeatedly, becomes a default the shop can see on the next order.
Instead of quietly rewriting anything, the system surfaces a short list of records that look wrong or contradictory, for the owner or a trusted staff member to confirm.
All processing can run on a local model inside the shop, so customer chats never have to leave the country.
Staff still take the order. The owner still decides what to do with a cleaned list. The system only observes and proposes corrections. It never messages a customer without approval and never fills in a name it is not confident about. Judgment about a valued customer stays with a person. The archaeology of digging through last year's chats to work out who someone is becomes the machine's job.
Today, most records are incomplete or contradictory, and nobody can say with confidence who the regulars are. Once this is built, records converge on a consistent name, contact and preference history, and the owner can see, for the first time, who actually buys from the shop and what they tend to order.
Customer data collected this way is personal data under the UAE's PDPL, and the same law that protects it also gives customers the right to have inaccurate details corrected, which is exactly what this system is built to surface. Running the model locally, on hardware inside the shop, keeps chat content and phone numbers inside the country instead of passing through a cloud service abroad.
A flower shop loses its customer between three channels and a staff turnover. Another business loses the same customer to a different gap. The pattern is the same: find where the record breaks, build the tool that keeps it whole, leave the judgment with a person.
Book a free scoping call