Amazon Supply Chain and Logistics

AWS offerings for visibility and on-time arrival of maintenance spares for mining and energy

The aim of any inbound supply chain for materials is to make sure that spare parts, rotable spares, and consumables arrive to operations in time to execute planned maintenance work. Too often, though, low performance, the lack of trust between teams and limited visibility for materials lead to frequent replanning and excess reliance on expediting to ensure maintenance work is executed.

Leading mining and energy players increasingly use cloud computing technology to drive better business outcomes and mitigate risks as follows:

  • Reduce asset downtime by making sure materials reach operations prior to the date when work needs to start. The opportunity size is an estimated $3–6M per year per operated asset.
  • Increase maintenance execution productivity with maintenance execution teams spending less time mobilizing work and looking for materials and more time conducting maintenance work. The opportunity size is five to eight percent increased productivity.
  • Have less excess inventory when visibility in material status and trust in underlying data is ensured. The opportunity size is a one to five percent reduction of inventory levels and two to five percent fewer expedited and air freight shipments.
  • Reduce indirect procurement spend for materials by enabling direct and guided buying for engineers. This streamlines the procurement process and uplevels the role of procurement from a transactional towards a consultative one. The opportunity at hand is to reduce procurement spend by up to five percent.

This post provides maintenance order execution control tower use cases and related business outcomes specifically for the mining and energy industry.

A maintenance order execution control tower powered by cloud technologies

Our vision is a maintenance order (MO) execution control tower that provides each stakeholder with full visibility on the key actions required to keep materials flowing and to understand the impact of their decisions on the total cost of business. In particular, it provides transparency on what the impact of scheduling or rescheduling a MO on a particular date is on material availability, maintenance execution productivity, transport, and accommodation utilization. To move towards this vision, companies need to build digital and machine learning (ML) capabilities both at the point of maintenance planning and when a MO has been planned.

At the point of maintenance planning

Future demand visibility with data-driven, ML-powered predictions and simulations

At the point of maintenance planning, teams generally lack confidence that required materials will be on site when they’re planning and scheduling out rough-cut work revisions and execution dates. The net result is frequent replanning and the corresponding juggling effort of moving work, requesting that materials be expedited, aligning labor for maintenance work execution, and setting safety stock levels at a higher level to ensure material availability. Suppliers lack visibility on what the future material demand looks like. That visibility would benefit their production planning and improve the relationship with the buyer. To reduce replanning maintenance activities––in other words, increase maintenance order in full on time (MIFOT)––and to achieve balanced inventory holding, leading companies apply cloud-based technology for material demand forecasting, safety stock analysis, and what-if scenarios.

ML using Amazon SageMaker supports the material demand planning process by overlaying the following:

  • Historical material consumption to predict material demand for future campaigns
  • Consumption-based data in the enterprise resource planning (ERP) software system for equipment as well as the equipment-installed base

The latter allows for calculating equipment and material failure rates (in other words, mean time between failure) using analytics. This provides maintenance planners with an automated forecast for maintenance activities, material demand, and visibility in historical material consumption. It also makes it easier for them to identify relevant task lists and bill of materials for future MOs.

AWS with our partner Deloitte supported an energy company that managed to automate the demand forecast for 25 percent of all materials despite limited data quality (such as their material master data and task list BOMs). Further enhancements are likely to improve coverage up to about 50 percent of all materials after addressing data quality issues with data enhancement technology. This allows for semi-automating of maintenance planning activities.

Furthermore, based on the material demand predictions, safety stock levels are calculated in a dynamic rather than static manner. The calculation takes desired service levels, material criticality, stock on hand, delivery times, and material lead times into account. Material managers benefit from this dynamic calculation of safety stock levels because they’re able to actively manage inventory and reduce overall stock levels.

Finally, what-if simulations create full visibility of production availability, cost, safety implications associated with changes to scheduling MOs, and even entire maintenance campaigns through time (for example, changing a start date by 1 month). This allows a maintenance scheduler and planner to quickly see the implications on the onsite availability of required materials when scheduling an MO on a particular date.

Think big. This view shouldn’t be limited to materials only. Consider the total cost of business with relevant information such as labor, transport, and accommodation utilization, and––most importantly–– potential risks for asset uptime. To achieve a complete view of the business, companies require a strong underlying data platform and robust data structure provided through constructs such as Lake House Architecture powered by Amazon S3 to store and quickly retrieve all relevant financial, operational, and material information.

Realistic vendor and transportation lead times for material availability and scheduling confidence

Maintenance planners often lack confidence to schedule a MO’s start date given insufficient information on how long it takes until materials reach on site. For direct purchases, the vendor lead time often consumes more than 60 percent of the entire end-to-end lead time. Also, maintenance planners lack reliable information when no outline agreements (which are similar to frame contracts in other industries) are in place or haven’t been updated in the recent past. ML using SageMaker analyzes historic PO information to predict actual vendor lead time.

Material purchases are often one-offs given the low standardization of assets and equipment, which calls for running supervised ML algorithms, such as XGBoost, on material level or material group and vendor level using similar materials whenever insufficient historic data is available. These models have proven moderate-to-strong predictive power with R-squared above 0.7.

A key challenge is to define the confidence level, such as a 95 percent probability level, at which the model predicts. In practical terms, this means that the prediction algorithm is trained to prevent the predicted arrival date from significantly preceding a required onsite date, which in turn means that high buffers are used that make MO planning unrealistic. In contrast, maintenance planners lose trust when the model predicts too short lead times which puts the timely execution of MOs at risk.

The following chart depicts an example for drivers that affect vendor lead time prediction, such as planned delivery time, vendor name, INCOTERMS, and PO creation month.

The chart providers users with high- or low-level predictors for a particular vendor, i.e. what features are likely to cause delays in material arrivals.

The chart providers users with high- or low-level predictors for a particular vendor, i.e. what features are likely to cause delays in material arrivals. Furthermore, analyzing historic vendor performance vs. contractual obligations helps users to manage vendor performance accordingly.

Whenever INCOTERMS require the buyer to organize transportation (such as Ex Works), a second ML-model needs to be built that predicts the lead time for transportation, particularly for international ocean freight shipments. The use of ML in predicting vessel time of arrival substantially increases the accuracy of landside operations planning and implementation, in comparison to traditional, manual estimation methodologies that are used widely across the industry.

When a maintenance order has been planned

Material visibility once purchased materials have been ordered

A common challenge for mining and energy organizations in operating the supply chain for materials is that maintenance stakeholders have limited visibility of materials in terms of progress from order approval to being available onsite. This applies especially to direct purchases from vendors and to some extent to supplies from inventory.

Procurement and logistics teams have a limited view on which materials are on the critical path or already delayed. They also lack guidance on what tasks to prioritize. This lack of visibility sets maintenance work at risk when critical materials aren’t onsite when required, or when there’s a lack of confidence that they’ll turn up on time to meet a scheduled date.

A material control tower creates visibility on how required materials are progressing along the end-to-end process. A series of milestones is defined for progress tracking purposes ranging from MO approved to delivered onsite. This visibility gives users such as maintenance execution, procurement, and logistics-valuable insights and confidence that their material will arrive prior to the planned execution dates.

Additionally, the visibility helps to prioritize and align actions required to ensure the flow of material and trust in the material supply chain. For example, the procurement team often works on thousands of purchase requisitions per month but lacks understanding of which material requests or related actions are most important or which materials are on a critical path or already delayed. A prioritization engine points users to the few vital tasks and provides contextual information, such as material and order criticality, and actions or escalations that are controllable by various participants in the supply chain.

A material’s progress can become blocked, and poor visibility can leave the owner of the demand unclear on the status of the material and who is responsible for a resolution. For example, a material can become obsolete if the vendor can’t manufacture or deliver the item anymore. This blocker can occur and be detected at multiple stages of the ordering and delivery process––at the time of PR or in development of a Request for Quote (RfQ)––and the best recommendations need to be highly contextual based on this. The material control tower provides cognitive capabilities both to detect these blockers and to provide non-trivial recommendations, such as finding a similar supplier or using clustering techniques to identify a substitutable material.

The material control tower focuses on the flow of materials and is an important building block of our vision of a MO execution control tower, which focuses on the MO execution and spans across both planning and execution processes. The MO execution control tower includes not only materials but also maintenance execution teams and other activities required (such as operating aircrafts and camps) to enable the execution of maintenance tasks.

Some leading AWS industrial customers are exploring the use of AWS Supply Chain, an AWS service, to provide material tracking and recommendations along four core capabilities:

  • Easier connection to data across systems. Relevant maintenance planning, procurement, and logistics data often sits in siloes, and relevant teams and users struggle to create an easy-to-use view about material status along the end-to-end supply chain. AWS Supply Chain helps to access these data siloes and uses a pre-trained natural language processing (NLP) model, an ML algorithm, to transform existing data into a target Data Model.
  • ML-powered insights. AWS Supply Chain then contextualizes the data in a visual map and highlights material progress, current inventory selection and quantity, and the health of inventory at each warehouse and site.
  • Material arrival time prediction using ML. This ML model can be extended by adding lead times for all remaining milestones and taking blocker resolution times into account to estimate a realistic arrival date onsite.
  • Recommendations and collaboration. The user gets automatic insights on recommended actions and urgent inventory issues. Watch lists can be set up to get alerts when there’s a potential supply chain risk. It also provides a collaboration feature, which allows users to communicate with each other and keep track of relevant information within the system.

Cognitive procurement for a fast, data-driven, Amazon-like buying experience for business users

Procurement plays a key role in optimizing the material supply chain for two reasons. Firstly, procurement spend of mining and energy companies tends to be in billions of dollars, and spend reduction translates straight to the company’s bottom line. Secondly, the median time from PR to PO acceptance by a vendor can often take multiple days, if not weeks. This is especially true for items with limited purchasing history variability in processing time, often driven by time-consuming approval processes, which adds to the complexity of confidently knowing how long the procurement process will take.

At the core of procurement transformation sits the aspiration to transform the procurement team from a transactional, reactive role toward a consultative one by enabling business users such as the maintenance engineers to directly purchase items from the vendors. This aspiration underpins the AWS offering called Cognitive Procurement Engine (COPE), which provides the following:

  • Self-service for spare parts that are below a set threshold value.
  • An Amazon-like buying experience for materials, making the end-to-end process to procure the right products quicker, easier, and more efficient.
  • Data for comparative analysis on products, prices, suppliers, historical spend, and personal user experience, which enhances and compliments punchout.
  • Recommendations, buying patterns, alternative products options, and a focused set of options based on the equipment type, role, and purchasing history to better match business users’ needs.
  • Continued control for the procurement team through full spend visibility and, where required, an escalation process exists for exceptions.

An engineer logs into the COPE web-portal and sees a few key options for the search term entered. The buying screen states why items are a good fit and why not. In the example of an O-ring, items 1 and 2 are a great fit for the purpose of use. If the buyer still considers item 3 as a good fit, they can go for the cheapest option. In this case, having the right equipment for the right process improves safety, asset life and reduces maintenance costs.

The illustration shows three examples of O-rings spare parts with their features and costs.

An Amazon-like buying experience with COPE reinvents today’s buying practices, automates large parts of time and resource consuming procurement processes, and provides the opportunity to reduce procurement spend by five to eight percent. Particularly for lower value items such as gaskets, it enables organizations to reduce time from PR to delivery and reduce labor costs to procure a long tail item from $200–300 per transaction to $5–25.

Amazon Business for bulk ordering and procurement process improvement

Several companies such as Chevron and Exxon Mobile have gone a step further with Amazon Business. Amazon Business builds on the user-friendly shopping experience of Amazon and combines it with essential features for modern procurement, such as multi-user business accounts that allow for creating purchasing groups with specific permissions, creation, guided buying, advanced search capabilities, and strong spend visibility and analytics.

Chevron, for example, procured a wide range of materials, often managed by business partners on behalf of Chevron. Managers created shopping lists for monthly shipments. Building the list, getting quotes, approving a quote, ordering, and accepting delivery would take four to six weeks. Internal customers such as maintenance teams had limited control over getting exactly what was needed, leaving many sites disappointed. Chevron moved purchases to Amazon Business to consolidate spending in a dynamic store environment, where everyone can find what they need and still take advantage of large-scale and bulk ordering. Amazon Business was integrated with their SMART GEP system Punchout to create a seamless user experience. Decentralized user orders are automatically consolidated into one purchasing card for master expense reporting and orders are consolidated into a weekly delivery (Amazon Day). Despite fragmented, transaction-driven materials categories, suppliers were consolidated, for example, by opening POs with flexible payment terms.


In this post, we discussed the opportunity for mining and energy companies to optimize their material supply chain by using cloud computing technologies. Opportunities exist both during the maintenance planning process and when material demand has been confirmed. Increased maintenance plan and execution stability helps to increase maintenance execution team’s productivity by five to eight percent, reduces costly expedite and air freight shipments by two to five percent and reduces the risk for asset downtime with a potential of $3–6M per annum per operated asset.

AWS provides several offerings that help create visibility of material status along the end-to-end supply chain, increase confidence that materials will reach site prior to the required dates, reduce indirect procurement spend by up to five percent, and increase trust in supply chain performance and commitments. Leave your opinions in the comments because we’re eager to learn and discuss your approaches around material supply chain management.

If you wish to explore how AWS could support you in cutting through the complexity of your supply chain, please reach out to your AWS account manager to set up a discovery workshop with our industry experts from the mining, energy, and industrial (MEI) practice as well as supply chain and logistics experts.

Manuel Baeuml

Manuel Baeuml

Dr. Manuel Baeuml is the Head of AWS ProServe Supply Chain & Logistics in Asia Pacific and Japan. Manuel and his team are responsible for sharing leading digital supply chain practices and for solving our customers’ most pressing supply chain problems using AWS cloud technology and offerings. Over the last 15 years, he has had the privilege to work with industry leaders in Asia Pacific and Europe in mining, energy, retail/CPG, as well as transportation and logistics. Manuel is based in Singapore.

Ben van Vliet

Ben van Vliet

Ben van Vliet is a Senior Customer Delivery Architect working with mining, energy, and industrial customers in Australia and New Zealand. Ben’s work spans from helping AWS ProServe customers define their digital and operational strategies through to framing and delivering strategic initiatives in Supply Chain, Industrial IoT, and greenfield product development. Ben has a deep background in mining, oil, and gas, and has held operational and leadership roles across outbound supply chain planning, maintenance effectiveness, and digital innovation. Ben is based in Perth, Australia.

Brett Birkbeck

Brett Birkbeck

Brett Birkbeck is the Head of Solutions Architecture for Mining, Energy & Industrials for Australian and New Zealand. Brett leads a team of AWS professionals who are responsible for helping our customers to harness the transformative potential of the AWS cloud. Brett has in-depth experience in mining and energy, advising in Digital Strategy, Technology-led Modernization, Industrial IoT, AI/ML and Digital Twins. Brett is based in Perth, Australia.