VRP worked with Beyond Better Foods to unlock their data using Einstein Analytics in three stages, which was followed by extensive training to enable their team to fully understand and evolve their instance of Einstein Analytics. The project started with the identification of value-creation goals, through interviews with internal stakeholders to untap hidden opportunities and mine existing datasets. The next step involved creating of a proof-of-concept to validate the value of Einstein Analytics from dashboards to predictions in a Salesforce sandbox. The final phase included fine-tuning datasets as Einstein assets were deployed in the production platform.
The starting position was a complex system of Microsoft Excel spreadsheets that collected both point-of-sales data and marketing spend. VRP used Einstein Analytics Data Manager to create a single source of truth. First, adding Salesforce connections to own client’s local and external objects custom related to distributors, categories, and supply chain orders. Second, transforming, grouping and joining the aforementioned sources together into datasets, ready for fast analysis of business metrics, such as distributor sales, sales velocity, using Einstein Dataflows and Recipes.
The final step required VRP to build an interactive multivariate analysis with Einstein Analytics Studio. The Einstein dashboard consisted of SAQL Lenses, obtaining statistical measures from the pricing list and trading terms, and interactive steps that upon selection switched measures and groupings.
Further releases will include a VRP developed AI Story within Einstein Discovery, with the goal of predicting discount levels by lot size and easily explaining the factors behind each prediction. VRP followed a full data science workflow within Einstein Discovery based on the client’s dataset: training, building, validating and deploying an advanced machine learning regression model. VRP deployed the model inside the opportunity object as a prediction field since Beyond Better Food's model showed strong performance in terms of r-squared, residuals and mean square error. Despite the sophistication of the AI workflow, only clicks were needed without code or Jupyter notebooks at all.
The future will also include expanded AI predictive modelling, predictive actions, alerts and triggers as well as geo mapping, location insights, recommendations and remote connections to relational databases for real-time competitive intelligence.