Data Science for operations
We leverage Data, technology, and AI to increase sales, service level and optimize inventory and costs. Our team of data scientists, engineers, business consultants and domain experts provide end-to-end support.
Using AI for sales forecast brings higher accuracy, lower bias and more automation. We typically help clients achieve +10pts to +20pts of forecast accuracy
Machine learning models leverage multiple sources of data, like sales, inventory, stock-outs, promotions, product attributes, pricing. Our models also learn complex patterns and correlations from past situations
- Sell-in & Sell-out forecast
- New products forecast
- Early sales reforecast
- Revenue forecast
- Influencer / Opinion leader sales impact modeling
INVENTORY MANAGEMENT and Obsolescence reduction
Inventory reduction is a strong lever of performance. Pressure on cash and growing environmental regulations are keeping these topics on top of Supply Chain agenda
Using AI to address complexity. AI allows to optimize inventory level of thousands of SKUs and anticipate future obsolete inventory. Machine Learning models typically deliver strong performance on slow-runners, where demand planner have usually less time to work-on
- Store-level inventory optimization
- End-of-life detections
MerchandisING and assortment planning
Assortment planning is a complex process at the crossroad of several functions.
We use historical data, structured data, but also images, product descriptions and customers profile data to build recommendation engines. We help planning teams in their planning processes, using AI as a support for human-based decisions
One challenge is to translate merchandising guidelines into data. Another element is to find similarities between products
- Open-to-buy recommendation
- Store-level merchandising and personalization
- Store clustering
- Dynamic assortment planning
Supply Chain complexity goes beyond what humans can manually monitor every day. Erratic lead-times, quality issues, unpredicted events, sales variations… are many reasons for a product not being at the right place at the right time
We use machine learning to detect anomalies. This goes beyond what typical DRP models and APS software would do, by learning from past situations which are the potential risks of stock-outs
- Product shortage detection
- Lead-time prediction
Logistics resources planning
Next-day delivery being the norm, the pressure on service level is very high. This context creates highly fluctuating demand in logistics, with a rippling effect into warehouse operations, transportation and admin functions.
Using Machine learning helps creating an accurate forecast of logistics activity using historical data, but also context information and exogenous data, like weather, traffic or vacations
- Last-mile activity forecast
- Warehouse resources forecast
New Product cost modeling
In many industries, it can be surprisingly hard to anticipate the COGS of new products. From raw material price variations, to manufacturing productivity, a huge number of variables will influence the actual cost of a product.
For new products, we have developed a methodology to identify key cost drivers from past similar products. We also use machine learning to model manufacturing ramp-up, based on product complexity, materials, production site, etc.
- New product cost modeling
Maintenance operations are on the cusp of a paradigm shift with the growth and accessibility of breakthrough technology, such as the Internet of Things and machine learning.
These developments will bring about significant improvement in our ability to predict changes in machine operating performance.
Using machine learning model helps to predict failures and optimize maintenance operations. These models typically use machinery data, IoT data, but also exogenous data such as temperature, calendar, humidity, etc.
Key topics are
- Predictive maintenance
- Spare parts inventory optimization
Quality can generate a lot of uncertainty in a production process. We use IA models to identify key factors influencing quality, to help manufacturing and quality teams to anticipate issues and take proactive actions.
Machine learning predictive quality models are typically over-performing statistical models as they better interpret complex and multicriteria situations. Machine learning is also better at identifying outliers, ie data points which are extreme and sometimes not relevant.
Typical projects we have worked on are :
- Predictive quality models
- Quality resources planning
predictive SPEND MODELING
Predicting spend with a good accuracy helps operational and procurement team to better buy, plan and negotiate.
We combine our knowledge of procurement processes and our data modeling capabilities to develop predictive spend models. Such models are able to quantify the impact of key drivers on spend. We typically use invoices, accounting data, supplier information, category, product, and price data to power-up these initiatives. We also collect external data like index, product reviews, demographic data, etc., to feed our AI model.
We help clients in developing and industrializing data-driven models for procurement:
- Spend forecast by category