Setting up Einstein Vision for product detection in Consumer Goods Cloud involves three primary stages. First, the feature must be technically enabled in Salesforce Setup (B). Second, the AI model must be trained to recognize the products; this is accomplished by creating a dataset with product images and training the Einstein Vision model (E). Finally, to make this functionality available to field reps, a specific "Planogram Check" task must be defined within the retail execution process (D). This task type is specifically designed to trigger the image capture and analysis workflow using the trained model.

A. Setup in store location and associate planogram: This is a foundational step for general retail execution, not a specific configuration step for the Einstein Vision feature itself.

C. Upload the training pictures as attachments to the planogram for Einstein Object Detection to work: Training images are uploaded to a dataset and associated with product records to train the model, not attached to individual planogram records.

1. Salesforce Help, Consumer Goods Cloud Documentation, "Set Up Einstein Object Detection for Consumer Goods Cloud": This guide explicitly outlines the required setup process. It lists the main steps as: "Enable Einstein Object Detection," "Create a Planogram Check Task," and "Train Your Einstein Vision Model." This directly supports options B, D, and E.

2. Salesforce Help, Consumer Goods Cloud Documentation, "Einstein Object Detection": This overview states, "To get started, enable Einstein Object Detection. Then, create a planogram check task. Finally, train your Einstein Vision model." This confirms the sequence and validity of options B, D, and E.

3. Salesforce Help, Consumer Goods Cloud Documentation, "Guidelines for Creating a Training Dataset": This document details the process of preparing images for training the model. It specifies associating images with products to create a dataset, which contradicts the idea of attaching images to a planogram (Option C). It also implicitly supports the necessity of training the model (Option E).