X

Machine Learning with H2O.ai and Oracle ERP

Ben Lackey
Director, Data and AI Partnerships

Packaged line-of-business (LOB) applications are an area where Oracle is a market leader. These applications contain an enormous amount of data that has the potential to give amazing insight into core business functions, enabling gains in areas such as:

  • Operational efficiency
  • Cross sell / up sell
  • Customer experience

Oracle is investing heavily in moving these LOB applications to our cloud, Oracle Cloud Infrastructure (OCI). In parallel, we're investing in the ecosystem, partnering with key ISVs to enable their workloads on our cloud, which allows our customers to leverage the latest innovations in conjunction with the LOB applications that they depend on.

H2O Driverless AI

H2O.ai is a leader in the AI/ML space. Their platform, Driverless AI (DAI), automates much of the machine learning lifecycle, from data ingestion to data engineering and modeling, on to deployment. It enables both data scientists and relatively naive users to generate sophisticated ML models that can have an enormous impact on their business.

Late last year, we began integration work with H2O.ai. Initial efforts focused on creating Terraform templates to automate the deployment of Driverless AI on Oracle Cloud Infrastructure. Building on that, H2O.ai was the first of our Quick Start templates to go live. Driverless AI can be deployed today on Oracle Cloud Infrastructure by using Terraform modules that our team and the H2O.ai team developed jointly. Those modules are available on the Quick Start on GitHub.

We’re currently exploring ways to use this kind of data in Oracle enterprise applications to build tailored ML models. An example architecture might look like this:

On Oracle Cloud Infrastructure, Driverless AI can be deployed on NVIDIA GPU machines. This accelerates the building of models, further reducing the end-to-end lifecycle for machine learning.

Oracle Retail Advanced Inventory Planning

The Oracle Retail Advanced Inventory Planning (AIP) module in Oracle ERP is one potential source of interesting data for ML with H2O.ai. An external merchandising system, forecasting system, and replenishment optimization system are integrated with AIP to provide the inventory/foundation data and the forecasting data to AIP to effectively plan the inventory flow across the retailer's supply chain. 

Because AIP can integrate with any forecasting system, Driverless AI could be used to build a model that accounts for both high frequency (for example, weekend) and lower frequency (for example, holiday) seasonalities. Driverless AI ships with a time-series recipe based on causal splits (moving windows), lag features, interactions thereof, and the automatic detection of time grouping columns (such as Store and Dept for a dataset with weekly sales for each store and department).

Oracle Retail Merchandising System

The Oracle Retail Merchandising System (RMS) module in Oracle ERP is another fascinating touchpoint. This module includes the following information:

  • Expenses: The direct and indirect costs incurred in moving a purchased item from the supplier's warehouse/factory to the purchase order receiving location.
  • Inventory Transfers: An organized framework for monitoring the movement of stock.
  • Return to Vendor (RTV): Transactions that are used to send merchandise back to a vendor.
  • Inventory Adjustments: Increase or decrease inventory to account for events that occur outside the normal course of business (for example, receipts, sales, and stock counts).
  • Purchase Order Receipts (Shipments): Record the increment to on-hand when goods are received from a supplier.
  • Stock Counts: Inventory is counted in the store and compared against the system inventory level for discrepancies.  

RMS contains a rich dataset that could be used to build models in Driverless AI for anomaly detection around RTV, inventory adjustment, and other events.

Oracle Retail Price Management

The Oracle Retail Price Management module in Oracle ERP includes the following information:

  • Item ID: ID that is assigned when the price event is created at the transaction item level.
  • Cost Change Date: The effective date of the past or future cost change.
  • Retail Change Date: The effective date of the past or future retail change.
  • Cost: The cost on the effective date of the cost or retail change.
  • Retail: The regular selling retail on the effective date of the cost or retail change.
  • Markup %: The markup percent on the effective date of the cost or retail change. The markup percent is calculated using the calculation method specified by your system options.  

With Driverless AI, we could use past cost changes to train a regression model. That model could suggest future pricing, automatically incorporating both seasonality and product lifecycle. Also, by combining Retail Price Management data with marketing, clickstream, or other end-customer data, a regression model could be built to predict the benefit of pricing changes while accounting for other variables that affect sales.

Oracle Retail Trade Management

The Oracle Retail Trade Management module in Oracle ERP includes the following information:

  • Landed Cost: The total cost of an item received from a vendor inclusive of the supplier cost and all costs associated with moving the item from the supplier's warehouse or factory to the purchase order receiving location.
  • Expenses: The direct and indirect costs incurred in moving a purchased item from the supplier's warehouse/factory to the purchase order receiving location.
  • Country Level Expenses: The costs of bringing merchandise from the origin country, through the lading port, to the import country's discharge port.
  • Zone Level Expenses: The costs of bringing merchandise from the import country's discharge port to the purchase order receiving location.
  • Assessments: The cost components that represent the total tax, fee, and duty charges for an item.
  • Transportation: The facility to track information from trading partners as merchandise is transported from the manufacturer through customs clearance in the importing country.
  • Actual Landed Costs: The actual landed cost incurred when buying an import item.  

With Retail Trade Management data tracking costs and delays in items being received at their final stocking location, Driverless AI could be used to build a risk model to estimate the impact of changing exact import/transportation routes.

Oracle Retail Invoice Matching

The Oracle Retail Invoice Matching module in Oracle ERP includes the following information:

  • Invoice Matching Results for Shipments: Shipment records are updated with the invoice matching, which attempts to match all invoices in ready-to-match, unresolved, or multi-unresolved status.
  • Receiver Cost Adjustments: Updates the purchase order, shipment, and potentially the item cost in RMS, depending on the reason code action used.
  • Receiver Unit Adjustments: Invoice matching discrepancies are resolved through a receiver unit adjustment.

By joining the information in Retail Invoice Matching with data in other modules, we can build a risk model in Driverless AI for suppliers to predict the probability of invoicing issues for future orders.

Next Steps

This post gives a high-level view of how an open Oracle ecosystem enables our customers to leverage the latest technologies from our partner ecosystem with the LOB applications that they've relied on for decades to run their business. We're actively working with several customers to prove this out in their environments. In addition, my team is working to create a more detailed demo of the integration described here. We look forward to presenting that in more detail, both on this blog and at several upcoming meetups that Oracle and H2O.ai are jointly organizing.

If you have questions, please reach out to Ben.Lackey@Oracle.com or Peter.Solimo@H2O.ai. We'd love to work with you and see what ML can do with your data!

Be the first to comment

Comments ( 0 )
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.