A system and method for forecasting demand for a retail organization.

01

What the public record shows.

The Indian Patent Office published application 202111003053 on January 29, 2021. The application was filed on January 22, 2021. The publication names BluePi Consulting Private Limited as the applicant and Anjna Bhati and Parth Gadoya as the inventors. It lists international classifications G06N 20/00, G06Q 30/02, and G06Q 30/06. The published specification contains 41 pages and 12 claims.

The Indian Patent Office application-status record identifies the application as granted and assigns patent number 388811. It lists February 8, 2022, as the certificate issue date and February 11, 2022, as the post-grant journal date.

02

The engineering problem behind the patent.

At an ordering cutoff, a planner must decide how much stock each location needs before the next replenishment window. Recent sales can point in one direction while promotions, missing records, lead times, and local demand point in another. A useful forecast must arrive at the right grain and in time for that decision.

Retail demand signals arrive across products, locations, channels, and source systems. Those inputs can have different structures, update times, and quality conditions. A forecasting system must prepare them consistently, generate usable features, train and validate suitable models, and produce a forecast for a defined planning period.

The patent describes this as a connected system rather than an isolated model. That system view remains relevant because a forecast can only support an operating decision when its data, evaluation, output grain, and planning path are controlled.

03

The six subsystems in the published abstract.

1. Gather sales data. Obtain real-time sales input datasets for one or more products from one or more data sources.
2. Prepare the inputs. Preprocess the incoming sales datasets for the forecasting path.
3. Generate sales features. Derive the features used by the machine-learning path.
4. Train with validation data. Train a machine-learning model for validation datasets using the generated features.
5. Train with test data. Generate a machine-learning model for sales test datasets.
6. Predict demand. Produce a sales forecast value for a defined period.

These six subsystems form one connected path: a sales signal enters as data and returns as a forecast a planner can act on before the next ordering decision.

04

From the patent method to a running forecasting system.

BluePi built the forecasting method in this patent, and we apply the same approach when we design forecasting systems for customers today. A production system adds the data and operating controls around the model, and the exact architecture depends on your planning decision and source systems.

• Source integration, data contracts, and cutoff-safe features
• Baselines, candidate models, and rolling backtests
• Product and location hierarchy with reconciliation
• Planner review and structured override reasons
• Planning or replenishment-system integration
• Bias, drift, service-level, and outcome monitoring
• Fallback behavior, runbooks, and operating ownership

05

The named inventors.

Anjna Bhati and Parth Gadoya are the named inventors on the grant. Anjna Bhati is also part of BluePi’s leadership team, which keeps the forecasting work described here close to how we design and deliver today.

06

Common questions.

07

Talk to the team behind the patent.

The method described here is the forecasting work BluePi builds for retail customers today. If you are weighing a forecasting or replenishment program, we can walk through how the same approach would map to your data, planning cycle, and replenishment systems.