- Retail
French Florist: A Stable Platform and AI Support With Human Approval
Stabilized a failing e-commerce platform on AWS, then added an AI support agent that diagnoses tickets and recommends the action a person approves.
Stable commerce and supervised AI
A person approves every AI support action
AWS + Claude + Model Context Protocol
- Ticketing
- Inventory and orders
- Cloud
Watch logs
AT A GLANCE
The Project at a Glance
The Challenge
French Florist's e-commerce platform needed several reboots a day, bugs hit nearly every feature, and an outsourced team diagnosed dozens of support tickets a day by hand.
What We Built
JBS Dev moved the platform to AWS, rebuilt the frontend in React, and added an AI support agent on Claude and MCP that recommends an action a person approves.
The Result
The daily reboots stopped, page load fell from 4.2 to 0.8 seconds, and the agent recommends the correct action first in 70-80% of tickets.
BACKGROUND
Executive Summary
French Florist is a national luxury florist and e-commerce franchise. Its storefronts act as local fulfillment hubs, and about 95% of orders come from online. The business depends on a stable platform to process orders, manage inventory across franchise locations and coordinate fulfillment.
The platform had become fragile. The codebase held hundreds of thousands of lines with no systematic architecture. Bugs affected nearly every feature, and manual fixes were required in almost every business process. The virtual machines needed several reboots a day to keep the site running. A large outsourced support team investigated each ticket by hand across several systems, slowly and at high cost.
JBS Dev stabilized the platform, moved it to AWS, and added an AI support agent that connects to the same systems, diagnoses each ticket and recommends an action. A person approves every action.
THE PROBLEM
Challenges
The client faced two connected problems.
- Bugs affected almost every feature. Critical workflows were broken or inconsistent.
- The team applied manual fixes to nearly every business process.
- The virtual machines needed several reboots a day.
- There was no test coverage, so deployments were unpredictable.
- Checkout errors stopped customers from completing orders.
- Dozens of tickets arrived each day: inventory mismatches, payment errors, code bugs and order issues.
- Each ticket needed a manual search across the ticketing system, the database, inventory, fulfillment and the log systems.
- A large outsourced team had slow resolution times and limited visibility.
- No pattern recognition existed, so each repeated issue went through manual diagnosis again.
- Support cost grew in line with franchise growth.
THE SOLUTION
How JBS Dev helped
JBS Dev delivered two connected initiatives: platform stabilization on AWS, and an agentic AI support system that a person supervises.
- JBS Dev moved the platform from the old virtual machines to AWS cloud architecture. The daily reboots stopped.
- JBS Dev rebuilt the frontend in React and kept the admin panel, the database schema and the proven API endpoints.
- JBS Dev added test coverage and systematic error handling in place of manual fixes.
- JBS Dev optimized database queries, threading and indexing for performance.
- The move had zero downtime.
- JBS Dev built an agentic AI system on Claude from Anthropic, connected through Model Context Protocol (MCP).
- The agent watches the ticket queue and gathers context from the connected systems. It diagnoses the root cause and recommends an action.
- A person reviews each recommendation and approves it. When the first recommendation is correct, the person approves it with no override.
- The agent reads the ticketing system, the code repository, the database, inventory, fulfillment and AWS CloudWatch logs.
- The agent executes the approved action. Complex issues go to an engineer with the full diagnostic context attached.
THE OUTCOME
Results
The site stopped failing. The daily reboots are gone. Systematic error detection and structured error handling replaced the manual fixes. Test coverage catches regressions before they reach production, so deployments ship predictably. The checkout errors that lost customers are gone.
Page load fell from 4.2 seconds to 0.8 seconds, an 81% reduction. Mobile revenue rose 156% in the first quarter after modernization, and conversion rose 34%.
The agent recommends the correct action first in 70-80% of tickets, and the person approves it with no override. A fractional development team replaced the outsourced support team, at lower cost. Tickets that took hours of manual investigation across several systems now take minutes. The agent runs all day and scales with franchise growth without a proportional cost increase.
0.8 s
Page load, down from 4.2 s
+156%
Mobile revenue, first quarter
+34%
Conversion
70-80%
First recommendation correct
0
Downtime in the AWS move
KEEP READING
More Case Studies
Industrial Products
Agentic AI Knowledge Platform for an Industrial Products Provider
A RAG and agentic AI platform on AWS replaced it. Employees use it hundreds of times a week, and customers get source-backed answers about products.
Read the industrial products case study
Retail
Custom Scheduling Platform for a National Specialty Pet Retailer
A new platform on AWS Lambda, API Gateway, and Aurora cut page load from 20 seconds to 1.5 and adds $800K a week in completed grooming appointments.
Read the pet retailer case study
Retail
Core Web Vitals Optimization for a Major Footwear Retailer
Targeted code fixes raised the score to about 25 in four months, a 257% increase.
Read the footwear retailer case study
