We are a medium-sized company, specialized in providing digital omni-channel transactional solutions (mainly online and mobile channels) to banks in Latin America, for both Retail and Business segments. Our digital solutions are offered under a SaaS model (Software as a Service).
The Problem
We are currently aiming to evolve its digital banking solutions, mostly focused in transactional operations (ex. balance inquiry, fund transfers, credit card payments) into relational banking ones.
In order to achieve this, In the Innovation team, we are working on a concept to create full contextual applications. These apps are required to analyze end user’s historical data of his interactions with the channel (mobile app or web banking portal) + real time data in order to predict what the user will perform in the app. Thus, the concept supports on a dynamic dashboard that displays the functionalities accordingly to the user’s context.
The recommendation engine should be able to use the internally and externally available data about the users/behavior and leverage sophisticated machine learning models to arrive at relevant recommendations.
The recommendation engine should be able to address capabilities including but not limited to below points:
-Recommend relevant actions to users based on their In App-behaviour.
-Predict the order in which banking product(s) should be presented (deposit accounts, credit cards, investment accounts, loans, etc.), so the front end can display them accordingly.
-Predict what information of each product will be consulted, so the front can display the information that is actually relevant for the user (ex. should display the balance, or should display last transaction movements)
-Provide recommendations based on the time of the day/month the info is consulted. Eg: A user might have a habit consulting certain info or performing certain transactions during a specific day of the month or hour.
-Provide higher weightage to transactions performed more regularly.
-For new users, the model should be able to suggest transactions according to demographically similar users.
-To further refine recommendations, along with relevance to user it could also consider overall popularity of the transactions.
-Even though our initial interest is to develop a POC for this recommendation engine, our production platforms handle millions of users on a monthly/daily basis hence the recommendation engine should be able to scale to process the data and provide real time results. It should be able to learn from the interaction the user takes with the provided recommendations.
Expertise Required
We are in need of creating a tailored recommendation system to Initially support a POC for our digital omni-channel solutions. We are looking for experts with experience about:
-Data architecture consulting
-Tailored recommendation engine expertise
-Management of large databases
-Multiple data sources
-Robust APIs development
-Data Sources and formats
We will provide raw data, the third-party will be in charge of analyzing it, and make necessary transformations to achieve the objective. We expect the receive suggestions regarding which other data we should gather in order to improve the engine’s accuracy.
We record the logs of the Interaction of the end users with the app and logs of the transactions that end users execute In the app (mobile and web).
We initially have 16 transactions that are of interest:
OWN CREDIT CARD PAYMENT
BILL PAYMENT WITH AMOUNT VALIDATION
BILL PAYMENT WITHOUT AMOUNT VALIDATION
MORTGAGE LOAN PAYMENT
LOAN PAYMENT
TRANSFER TO SUBSCRIBED ACCOUNT (SAME BANK)
TRANSFER TO NON-SUSCRIBED ACCOUNTS
TRANSFER TO INVESTMENT ACCOUNT
TRANSFER TO OTHER BANKS
TRANSFER WITH QR CODE
PRODUCT TRANSACTION INQUIRY - CURRENT ACCOUNT
PRODUCT TRANSACTION INQUIRY - SAVING ACCOUNT
PRODUCT TRANSACTION INQUIRY - INVESTMENT ACCOUNT
PRODUCT TRANSACTION INQUIRY - VIRTUAL INVESTMENT ACCOUNT
PRODUCT TRANSACTION INQUIRY - LOAN
PRODUCT TRANSACTION INQUIRY - CREDIT CARD
Technology stack:
The expert should specify which frameworks and data infrastructure is currently using or has used for his/her last projects.
Deliverable for the Proof of concept:
The deliverable should be an algorithm which is able to predict which are the actions that users are most likely to take when they login, on an individual level. This prediction/recommendation engine should be able return the prediction for each of the users based on the context of his/her interaction. The engine should make available this information In order to display It through a dynamic dashboard (via API)
This engine should work in real time (although for the PoC a batch process will be enough, the final product should be live). The algorithm should be robust enough to be scaled and must be able to take thousands of concurrent API calls.
In your proposal:
We have the following Information Request/ questions to gather the required details for considering your organization to develop the aforementioned Recommendation Engine:
1.For which industry have you developed recommendation engines before? Please share broad description of underlying logic. Also we would like to explore one of the recommendation engines that you have deployed for your clients in the past.
2.Profile of experts that will be working on the project along with details of their qualification, past projects and Tools and technologies used earlier?
3.What is the technology stack that you would suggest for developing the recommendation engine?
4.What approach would you use to develop the POC for the recommendation engine?
5.How would you suggest to evaluate the recommendation quality?
Timeline
Since we are working on this project through our Innovation process, we expect to have preliminary results between 1 and 2 months.
Additional videos, prototypes and data related with transactions will be shared with shortlisted candidates.
