Vision Powered Demand Forecasting

Demand Forecasting – Importance & Challenge

Accuracy of demand prediction can have significant impact on any business. It can help both – enhance revenues, and reduce costs.
Statistical forecasting methods merely present historical trends in demand but fail to conclusively predict future demand of a product with an ever changing external environment.
Any product’s ability to sell depends on a variety of factors such as Product Characteristics including its non-visual & visual features, External Factors such as weather, events etc and Historical Demand.
Most demand prediction solutions fail to address Product’s Visual characteristics and External Factors.

Vision Powered Demand Forecasting

View any product as a set of pre-defined visual & non-visual features and characteristics and build Image recognition model to extract these visual features from vast pool of images of the product (including those of different models, product variants etc.).
Incorporate External Factors including weather, events such as sporting events, festivals, competitor events and sentiment analysis through social listening.
Train Demand Prediction models using Artificial Intelligence techniques incorporating all influencing factors & historical sales data.
Predict Demand of products using trained models.

Configure Demand – Product Visual Features

Visual Features are extracted using Image recognition techniques.
For example, in the case of spectacles, shown features were extracted from product images of glasses and shoes.

Predicting Demand - Results

AI models predict demand using uploaded product details (meta-data & images) and future demand is displayed.
Filter and Drill-down feature allows user to filter demand based on location, size and specific products.
External Influencers such as festivals, promotions, competitor play can be added to see their effect on demand.

Predicting Demand – Understanding Results

Predicted demand is a composite effect of Seasonality, Trend, Product characteristics such as price, style & color and effects of external factors such as festivals.
Downward trend from May to August shows sunglass demand reduces as summer recedes.
Upward trend from August to October shows demand increases around Diwali festival.

Solution Framework

Framework Approach:

User Interface Module:

Web-interface to access different modules of the solution.

Product Configurator Module:

Helps define potentially any product in terms of its Visual and non-Visual Features.

External Factors Module:

Helps define external factors including weather, events and sentiment analysis.

Trainer Module:

Trains AI models based on product characteristics, external factors and historical sales data.

Demand Predictor Module:

Predicts demand for defined set of products for which demand is to be ascertained. Provides RESTful APIs for seamless integration with other IT systems.

Emerge.ai - SaaS Portal

Emerge.ai has been developed as a Software as a Service (SaaS) Solution leveraging Microsoft’s Azure cloud platform.

Dashboard

Personalized & configurable Dashboard shows key statistics from the previous demand predictions.

Product Categories Supported

Emerge.ai currently supports Eyewear and Footwear product categories.
More products can be added by building specific image recognition models.

Configure Demand – Location & Time

User must configure Demand Features before the AI models can be trained and demand can be predicted.
Location and Time hierarchy must be configured first.

Configure Demand – Product Features

Product features include non-visual and visual features.
These features have been pre-defined for product categories supported by emerge.ai.
Visual features can be selected from available features. These are extracted from product images.

Configure Demand – External Data

User can select External data that influences demand.
Weather features include temperature, rainfall, snowfall and humidity.
Events related to festivals, sporting events, promotions & competitor events can be specified.
Social media sentiment influencers can also be selected.

Configure Demand – Custom Features

Custom features, which are specific to a customer’s business e.g. Sub-brand, can be added to demand features.

Configure Demand – Sales Data

Finally, demand features related to Sales such as Sales Volume, Price, Discounts etc. can be selected.

Train Model – Demand Data Structure

Once Demand is configured, user can train AI models by clicking on Train button on Configure & Train screen.
Data Structure for the demand can be viewed.

Train Model – Training and Choosing AI model

Historical Demand Data can be uploaded manually using demand template or can be uploaded directly from the ERP system using emerge.ai API
By clicking on Upload & Train button, image features are extracted and AI models are trained.
User can select the best model from the list having the least RMSE score.

Predicting Demand – Uploading Data

Once AI models are trained, they can be used to predict demand.
User must first upload a list of products and their images for which demand has to be predicted.
Future period for which demand has to be predicted must be specified.

Contact Us

Demand Forecasting – Importance & Challenge