Automate you company’s supply chain with AI/ML Presenters: Paul McClure James Triggs Chief Revenue Officer AI Product & Delivery Lead Ultra Commerce Intellify pmcclure@ultracommerce.co james.triggs@intellify.com.au @paul_mcclure
Agenda Ultra Commerce & Intellify Benefits to Expect Journey to Automation AI/ML Data Driven Commerce AI/ML Use Case Examples ML Optimised Order Processing Demand Forecasting Purchasing Recommendations Summary Q&A
Ultra Commerce & Intellify Data Digital Data Driven Commerce Science Commerce
Benefits to Expect Deliver the desired customer/supplier experiences 1 Improve productivity through digitisation 2 3 Deliver cost efficiencies through automation Increased revenues 4
Journey to Automation
Journey to Automation Offline Channels 1 Source: Accenture, Channel Shift: Measuring B2B Efforts to Shift Customers Online
Journey to Automation While 30% of buyers would prefer to buy 90% or all of their products online, only 19% are doing so. 98% of buyers are forced to pursue Offline offline help from a sales Channels representative at some point throughout the purchase journey. 1 1 Source: Accenture, Channel Shift: Measuring B2B Efforts to Shift Customers Online
Journey to Automation Integrations Offline Channels 1 Source: Gartner, 2019 Strategic Roadmap for Digital Commerce
Journey to Automation The three biggest challenges in launching a digital commerce site — including B2B implementations — are integrations to other systems , lack of internal resources Integrations and implementation costs. 1 Offline Channels 1 Source: Gartner, 2019 Strategic Roadmap for Digital Commerce
Journey to Automation Much of the digital business technology platform is in essence an integration engine that adds some new digital capabilities to Integrations those the company already has. 1 Offline Channels 1 Source: Gartner, Survey Analysis: Building a Digital Business Technology Platform Requires New Technologies and Methods
Journey to Automation Integrations Offline Data Channels
Journey to Automation “The world’s most Integrations valuable resource is no longer oil, but Offline data”. 1 Data Channels 1 Source: The Economist, May 6th 2017 edition
Journey to Automation Automation Integrations Offline Data Channels
AI/ML
By 2023, 80% of organizations using AI for digital commerce will achieve at least 25% improvement in customer satisfaction, revenue or cost reduction 1 1 Source: Gartner Inc. - Use Personalization to Enrich Customer Experience and Drive Revenue
Data Driven Commerce Data Sources ETL / Query Pre-defined ML Models Insights/BI Streaming App Fraud Time-series Personalisation Detection Forecasting Historical DB Data Aggregate Data Repository CSV File Upload External Systems / Integrations
AI/ML Use Case Examples Insights: Customer Journey Analytics Sentiment Analysis Voice of Customer Analysis Trend Analysis Process Automation: Search Fraud Detection Customer Segmentation NLP & Visual Search Order Processing AI/ML Decision-Making Augmentation: Product / Inventory Dynamic/Optimised Pricing Demand Forecasting Supply Chain Forecasting (Shipping & Logistics)
AI/ML Use Case Examples Insights: Customer Journey Analytics Sentiment Analysis Voice of Customer Analysis Trend Analysis Process Automation: Search Fraud Detection Customer Segmentation NLP & Visual Search Order Processing AI/ML Decision-Making Augmentation: Product / Inventory Dynamic/Optimised Pricing Demand Forecasting Supply Chain Forecasting (Shipping & Logistics)
Order Processing
64% of B2B organizations claim that their long- term customers’ resistance to change is a barrier to driving more online sales. 1 1 Source: Accenture, Channel Shift: Measuring B2B Efforts to Shift Customers Online
ML Optimised Order Processing Post JSON to Ultra Commerce API’s for Automated Order Scan of a Purchase Order Instantly “read” virtually any Automatically identify key Processing & Fulfilment based on a workflow type of document information By using machine learning to instantly “read” virtually any type of document to accurately extract text and data without the need for any manual effort or custom code.
Order Processing Customer Service Representative @ B2B Manufacturing Company 15% of orders via Web, 85% of orders via Email (PDF) and Phone Ideal: • Efficiently and accurately process thousands of orders per month across all sales channels (phone, email and web/online) including the ability to read order emails and their attachments for automated order processing and fulfillment Reality: • Manual order processing by 7-10 Customer Service Representatives • Limited visibility on stock availability / inventory levels due to disparate business systems and data silos • Limited ability for customers to track orders Consequences: • Poor customer/supplier experience • Lost revenues due to order processing delays during high demand periods • Lost revenues due to human errors related to order processing • Increased costs due to higher staffing/resource requirements
Order Processing 85% of order volume Email Orders Phone Orders Web Cart / Checkout Storefront/Marketplace Supply Chain Management Customer Service Representatives Warehouse Manager Business Systems
ML Optimised Order Processing Email Orders Phone Orders Web Assisted Customer Account/Profile Shopping Payment Preferences Shipping Preferences Cart / Checkout Quotes Approvals Workflow Storefront/Marketplace Supply Chain Management Amazon Textract Customer Service Representatives Content Management Product Management Post Order Management Business Systems
Outcomes Delivered Increased order status visibility for customer 1 Reduced operational headcount for CSR 2 teams Increased revenues through increased order 3 processing capacity and reduction in human errors
Demand Forecasting
Time Series Forecasting Examples/Scenarios Product Demand Call Center Staffing Utilities Usage Transportation Needs (Electricity, Gas and Water) (Geography based) Cash Flow Cash in an ATM Flight Ticket Prices Ad Impressions Time Series Forecasting Inventory Planning PAYG service usage Deliveries per Post Code Patient Volume Real Estate Prices Sensor Network Monitoring Broadband Usage Selling Price of Crops (Telco)
Demand Forecasting Demand Planner @ Trade Distributor Forecasting demand across 200,000 SKU’s Ideal: • Easy mechanism to produce accurate demand forecasts. Reality: • Software based on traditional statistical techniques. • Good intuition on what drives demand, but not enough time to apply this to the forecast for each SKU. Consequences: • Forecasts that vary greatly in quality. • Forecast errors flow through the rest of the supply chain planning. • Demand forecasts are not trusted by the purchasing planners.
The True Costs of Inaccurate Forecasting Product demand Actual Demand vs Forecast Demand Over-forecasting leads to wasted resources Under-forecasting leads to lost opportunity Actual Demand Forecast Demand
ML Optimised Demand Forecasting • Amazon.com achieved 15x forecast accuracy improvement from 2007 to 2017. • Shift from univariate (each forecast is done in isolation just looking at volume) to multivariate (pulling in other data that drives demand) • Categorical data about products (e.g. Colour, size, product category, brand) • Related time series (e.g. How prices, promotions and the weather changed over time) • Seasonality hints (e.g. Holidays, weekends, closedown periods) • Algorithm improvements from ARIMA of the 1970’s to today’s deep learning models.
Outcomes Delivered Improved forecast accuracy. 1 More granular forecasts by SKU, by location, 2 by warehouse. 3 Scenario modelling such as how pricing could impact demand.
The enemy of accuracy-based planning is uncertainty. Uncertainty makes the plan inaccurate. What ML really helps with is its ability to convert unknown uncertainty to known variability. 1 1 Source: Gartner, Mastering Uncertainty: The Rise of Resilient Supply Chain Planning
Purchasing Recommendations
Purchasing Recommendations Purchasing Planner @ Distributor Responsible for ordering 5,000 SKU’s Ideal: • Optimised inventory purchases for each SKU, maintain the right stock in the right locations at the right time. Reality: • Pressure from finance to reduce working capital • Pressure from sales to improve service levels • Notified late of promotional items (too late to source via lower cost transport) Consequences: • Working capital tied up in slow moving stock • Can’t invest in the inventory required to improve service levels on critical products
Purchasing Recommendations How ML can make Finance, Sales and Inventory happy at the same time • Drivers of safety stock: • Unexpected variability in demand ML can forecast most of this variability, reducing • Unexpected variability in lead time the amount of safety stock required. • ML generates recommended inventory purchases to the purchasing planner – “Buy 300 widgets on 1 st July.” • Reduces over and understocks. Sanity check for the purchasing planner. • Run simulations of the recommended purchases to demonstrate the value.
Recommend
More recommend
