Turning fragmented commercial data into one usable view.

Commercial information often lives in separate systems. Customer details may sit in a CRM, purchases in an e-commerce platform, and new leads in a chatbot.

Each system can work correctly on its own while still leaving the business without a clear answer to simple questions: Who is this customer? What have they bought? Are they already a lead? What is happening to revenue and pipeline?

I wanted to explore how a lightweight MVP could connect these sources without attempting to replace the systems themselves.

The CRM and e-commerce system do not share customer IDs. A person might be crm_001 in one source and shop_101 in another.

For the MVP, I normalised email addresses and used them as the matching key. Once a customer is resolved, their orders, deals and chatbot events reference one internal customer record.

During polling tests, an order was imported more than once. Application-level duplicate checking was not enough to guarantee data integrity.

I fixed the issue at the database level by adding unique constraints to external identifiers such as order IDs, CRM deal IDs, chatbot conversation IDs and customer emails.

That changed the architecture from “the importer should avoid duplicates” to “the system cannot store the same source record twice.”