Agentic Book-Sourcing Workflow for UK–Ukraine E-Commerce
70–80% time saved
Summary
We built an agentic AI workflow for an e-commerce book retailer that sources inventory in the UK and sells in Ukraine. The system uses store sales data to discover books worth sourcing, identifies price-attractive items, and pushes candidates into Trello for human review—so the team decides what to order instead of manually hunting for opportunities.
Domain: E-commerce Operations · Sourcing & Procurement
Process type: Data-driven discovery + human-in-the-loop decision (AI Agents, intelligent workflow)
The Client Situation
A book retailer operates across two markets: sourcing in the UK, selling in Ukraine. To stay competitive, they need to constantly find titles that are priced attractively in the UK and have demand (or potential) in Ukraine. Doing this manually—checking sales data, scanning sources, comparing prices, and deciding what to order—was slow and left margin on the table. The team needed the discovery and shortlisting automated, with a human still in the loop to approve what actually gets ordered.
What We Delivered
We implemented an agentic flow that automates the front end of the sourcing process and hands off to humans for the final call:
Discovery driven by your own sales data
The system uses sales data from the store to identify which kinds of books are relevant—so scraping and search are focused on titles that align with what already sells or fits the assortment.Scraping and price-attractiveness logic
Agents scrape sources for interesting books (e.g. by category, author, or demand signals) and apply logic to flag items that are price-attractive—i.e. where UK cost and Ukraine selling potential make the item a good sourcing candidate.Trello as the handoff layer
During discovery we proposed Trello as the handoff so the team could review and approve or reject without leaving their existing workflow. Candidates that pass the automated filters are added as cards; each card represents a potential order with title, price, and context in one place.Human review before ordering
The flow stops before any purchase: a person reviews whether it makes sense to order each item. That keeps quality and assortment control with the business while the AI handles the heavy lifting of finding and shortlisting.Agentic implementation (CrewAI, LLM-powered)
The pipeline is built as an agentic workflow (CrewAI, LLM-based agents). Agents decide what to scrape and how to interpret sales data and demand signals—so the process adapts to new categories or sources without rewriting code. A fixed script would require developer changes for every new source or rule; here, the process is autonomous up to the review step and can be extended by adding steps or sources.
Outcome
We delivered an agentic book-sourcing workflow that turns “find and shortlist” into an automated, data-driven pipeline and leaves “order or not” to the team in Trello. The client gets a steady stream of sourcing candidates (e.g. dozens per week) with no extra headcount, grounded in their own sales data, with a clear handoff so humans stay in control of ordering decisions. The same pattern—agents + data + human review—applies to other e-commerce operations (e.g. restocking, assortment expansion, or supplier discovery).