A recommendation engine that predicts e-commerce promotion effectiveness a week ahead so store owners can pick the right discount.
- Prediction accuracy
- 91%
- Accuracy lift
- >50%
- Recommendation
- Discount type and depth
“Working with Datablooz has been an exceptional experience, defined by a high level of professionalism, strong technical standards, and deep expertise. First and foremost, their team succeeded in building a machine learning model that several experienced engineers before them were unable to deliver. This was a critical milestone for our business and a testament to their technical capability and persistence. Beyond that, Datablooz stands out in their ability to build products that are both fast and robust. They go far beyond pure development, offering strong product thinking, project management, and strategic guidance throughout the process. This holistic approach makes a real difference, especially in early-stage environments where speed and clarity are essential. What truly sets them apart, however, is the level of commitment and personal investment from Luka and the entire team. Their dedication, responsiveness, and ownership mindset are difficult to match. For non-technical founders, having a partner like this is invaluable: it creates trust and allows you to focus on building your vision, rather than constantly managing and challenging the development process. In a space where working with agencies can often feel like a struggle, Datablooz is the opposite: a reliable, high-quality partner that you can genuinely depend on.”
What we shipped
PromoLens helps online store owners decide which promotion to run next week. We built the AI behind it: a forecasting and recommendation engine that predicts the effectiveness of percentage discounts, bundles, and other promotional structures one week ahead, then recommends which type and depth of discount to run.
E-commerce store owners constantly run discounts, bundles, and seasonal campaigns to drive sales, but choosing the right strategy is hard. Most teams rely on intuition or simple historical comparisons, which creates uncertainty about short-term sales impact, risks margin erosion through over-discounting, and offers no way to test scenarios before launch. PromoLens needed an AI engine that could learn from real sales data, forecast how different discount types and depths would influence demand, and recommend which promotion to run next week with a level of accuracy that experienced ML engineers had not been able to deliver before.
Blueprint → AI Pilot → Production launch → Scale and operate.
We followed the Datablooz Delivery Model. See our process.
- Blueprint
Inventoried historical sales and promotion data, defined target scenarios, and framed forecasting as next-week decision support for store owners.
- AI Pilot
Trained time-series forecasting and recommendation models on historical promo outcomes and evaluated against baselines and prior attempts.
- Production launch
Deployed the simulation and recommendation engine into the PromoLens product with continuous live-data learning.
- Scale and operate
Extended to dynamic pricing, segment-level promos, inventory-aware discounts, and richer bundling logic.
Business, technical, and governance outcomes.
- 91% forecasting accuracy on discount outcomes one week ahead.
- Over 50% error reduction vs the prior system.
- Recommendations on discount type (percentage vs bundle) and depth.
- Tighter margin and revenue tradeoff control for PromoLens customers.
- Python
- Prophet
- LightGBM
- Airflow
- PostgreSQL
- FastAPI
Model retraining cadence on live data with tracked prediction accuracy and scenario-level audit records.
Working on something similar?
Schedule a call. We will tell you honestly whether AI is the right move.
Reference calls available under NDA after the second working session.