Corca Project
We are currently focusing on two products: ADCIO(Retail Media Platform) and DSP(Demand Side Platform)
ADCIO (Retail Media Platform)
(Specialty) Product exposure placements can be set up with a single line of script tag and personalized suggestion can be introduced in addition to LLM based AI chatbot function.
(Target Customer) Platforms with multiple sellers who want to enhance customer experience and advertising sales, and Sellers who want to increase conversion rate and sales revenue at their online malls.
As 3rd party data is limited due to privacy issues, retail media solutions with their own 1st party data are emerging as an alternative in this changing paradigm.
Corca’s ADCIO is a B2B SaaS solution which is easy to introduce and suggests products that are most likely to be clicked or purchased based on customer data.
1.
Solution (1): ADCIO Placement
•
Existing product grid can be set as a ‘personalized suggestion placement’ by simply inserting a script tag without any additional development costs
•
Status(period, product, target customer, etc.) of campaigns to expose can be set up on the web admin page.
•
Stores, where multiple sellers are entered, can generate advertising revenue with advertising credit paid by sellers
2.
Solution (2): ADCIO Assistant
•
LLM(GPT) based chatbot acts as a personal shopping assistant, interacting with customers giving product suggestion
•
Increase conversion rate while inducing movement toward product details page
•
Link customer login information for more accurate suggestion and saving conversation history
DSP (Demand Side Platform)
(Specialty) Effectively buy ad impressions programmatically in real-time-bidding environment, maximizing user response to display advertisements
(Target Customer) A platform that matches advertisers to purchase digital advertising inventory(ad placements) efficiently
1.
Real-time high-capacity parallel processing ML system
•
Develop a system deducing 300,000 data requests per minute in 10ms
•
24 / 7 Service
2.
Summary of Corca’s technology
•
Dependency Management
•
Scalability
•
Fault Tolerance
•
CI / CD Pipeline
•
AWS Resource Management
•
Sturctured Test Environment
3.
Achievement
•
CPC
◦
23.52% advantage over traditional systems in a 5:5 traffic period
◦
29.65% advantage over traditional systems in a 7:3 traffic period
•
ROAS
◦
Consistently achieved monthly ROAS ranging from 1200% to 2000%, varying across different advertisers.
Project History
Corca has project experience using ML research to drive the development of technology needed in the market.
HDS (Hyper Dimensional Sales Solution)
(Specialty) Support AI Transformation of enterprises by providing product data-driven demand prediction and automative inventory management
(Target Customer) Retail companies that continue to purcahse and sell products
•
Structure of Solution
◦
HDS model operates on two stages: HDS Predictor and HDS Recommender
▪
HDS Predictor : Prediction Model
•
Reflect the continuity of time using the Neural ODE algorithm
•
Learn about sales trends through general pattern identification
▪
HDS Recommender : Recommendation Model
•
Recommend a sales method maximizing sales revenue while maintaining profitability
•
Operation Flow of Solution
•
Deployment Process
◦
Establish a data pipeline through POS terminal interworking
◦
Develop sales/demand prediction model considering purchase history and customer purchase patterns
◦
Learn ML models for continuously updated data
◦
Provide a dashboard making it easy to see model predictions
•
Deployment Effects
◦
Proactively leverage customer purchase data to predict demand based on past purchase patterns
◦
Proactively replenish the right quantity of inventory through demand prediction to maintain the right amount of inventory
◦
Provide meaningful insights by analyzing monthly/quarterly patterens of each company’s purchase/sales history
◦
Predict concentration of logistics considering regional information
•
Deployment Examples
◦
Discover products expected to sell more
◦
Recommend a sales method to maximize sales revenue per product
⇒ Customize the solution to suggest products and sales methods, including a company’s conditions
⇒ Sales were overwhelmingly increasead after the introduction of the solution
Vendre: Deep Learning Video Search Prototype
(specialty) support business innovation through video analysis and in-video search function
(target customer) Companies specializing in video contents that want to extract specific information
•
Structure of Solution
◦
Vendre operates on two stages: Video Analyst and Image Labeler
•
Deployment Process
◦
Build a data pipeline with video upload platform interworking
◦
Develop an analysis model dividing images into images in seconds
◦
Develop a labeling model matching keywords by images
◦
Learn ML models for continuously updated data
◦
Provide a search tool to find matches to keywords in images
•
Deployment Effects
◦
Be able to upload the video to a web page and find the location with the keyword
•
Deployment Examples
◦
Search for key points in CCTV/black box images
◦
Manage training through lecture video navigation and bookmark functions
•
Etc.
◦
Develop the solution with technology of CLIP (Contrastive Language-Image Pre-Training), Pytorch, FastAPI, etc.
◦
Development Period : March 7, 2022 ~March 18, 2022 (2 weeks)
Reference : Twelve Labs attracted $5M investment on March 20 with similar techonology(”Video Understanding AI”)
ORTools: Solver to find the best combination within constraints
(Specialty) Support business innovation through optimized organization and selection of promotions
(Target Customer) Distribution companies looking to increase sales through promotions
•
Structure of Solution
◦
OR Tools are applied and operated as a functional tool in the prediction model
•
Deployment Process
◦
Establish a data pipeline through POS terminal interworking and product details
◦
Develop prediction model considering purchase history, product properties and correlation among products
◦
Derive optimal results after conducting the experiment of configuration and combination of promotions
◦
Learn ML models for continuously updated data
◦
Provide a dashboard making it easy to see model predictions
•
Deployment Effects
◦
Quickly derive the optimal combination of products by conducting various combination experiments at a given time and conditions
•
Deployment Examples
◦
Identify product configurations and combinations expected to increase sales
◦
Recommend a sales combination to maximize sales revenue per product
•
Etc.
◦
Google's processing size is more than 3 times larger than Google's open solver, and it is expected to improve performance more than 10 times after optimization developed in a quadratic manner
▪
Development Period: November 2021 ~ January 2022 (2 months)
Reference : Google OR-Tools