This case is about great cooperation between people from the sales industry and software development engineers. The idea was to create the software that can fetch historical deals from various CRM systems and based on standard financial metrics (deal start date, end date, value, sales rep, etc.) predict new deals lifecycle. It was an interesting journey of financial analytic transformation and machine learning algorithms.
Task
Provide full-cycle software development services to build a financial platform based on PoC made in Google Sheet. The system should be able to fetch historical deals from the top 3 CRMs like SalesForce, MS Dynamics, and HubSpot, also, support file import. Having historical deals build sales pipelines and determine the current sales cycle to predict existing deals behavior through time including ML algorithms to explain this behavior. The scope of this project, named TwelveZeros, included end-to-end development of the back-end, front-end, and cloud infrastructure, UI/UX design, unit & integration tests, machine learning algorithms, and whole architecture documentation. Here you can see the TwelveZeros in live: https://www.twelvezeros.co/
Goals
- Build a financial prediction system based on historical information
- Implement graceful integration with CRM systems
- Build flexible but unified data structure
- Implement complex UI with charts that allow users play with data in realtime
- Implement subscription management functionality and integrate with payment system
- Implement cloud infrastructure and configure continuous health monitoring approach
- Collect usage and performance metrics
Challenges
- Translate business requirements from basic PoC calculation tables to the technical specification
- Build interactive UI and understandable UX
- Determine the best ML algorithm to classify deal's features
- Implement custom subscription management functionality
- Deal with terabyte of the fetched data (fetch, store, ETL)
Actions
- Built a dedicated team to cover front & back end development
- Implemented agile development process
- Used google cloud as a sound infrastructure provider
- Conducted about 32 different tests to determine the best ML algorithm
- Designed & implemented custom UI/UX
- Implemented robust database cluster and production infrastructure
- Integrated with Stripe payment system and implemented personal subscription management functionality
- Used reactive-based protocol of the communication between system components
Results
- Production release in 9 months after the start
- First paid clients in 2nd months
- Gained 8 paid clients in the first 4 months after production release
- Manage 3 terabytes of data in the first 6 months
- ML prediction accuracy ~83%
