Let’s consider a large car manufacturer such as Toyota for a moment. Building a typical car requires around 30'000 components, from the engine block to the smallest screw. Toyota could never afford to build all of these components itself, and therefore has many suppliers around the world. Reports say it currently has 400'000 different parts suppliers registered on their system, building a vast and complex network and requiring massive coordination efforts. In North America alone, Toyota’s purchasing volume for parts amounts to about $26 billion each year.
Now imagine you’re working for Toyota and find a way to save 1% per year, finding the right combination of suppliers and volume discounts for these car parts. That’s savings of up to $260 million in the first year. And that’s just the beginning.
According to the Chartered Institute of Procurement & Supply (CIPS),
“procurement is the business management function that ensures identification, sourcing, access and management of the external resources that an organization needs or may need to fulfill its strategic objectives.”
This is a mouth full of words. In other words, procurement covers the entire process from identifying a demand in goods or services, over sourcing vendors, to negotiation, ordering, and invoicing payments. Even simpler, it’s everything in a company that has to do with purchasing supplies.
In most organizations, the procurement departments are notorious for being penny-savers. You can’t blame them; in the end, it’s their job to buy the ideal combination of materials or services to generate maximum savings for a company.
Though procurement optimization doesn’t sound exciting like robots and lightsabers in a volcano on Mars, I hope I can shed some light on the potential and interesting use cases that organizations are working on and envisioning.
Artificial Intelligence (AI) is frequently perceived as a one-in-all automation solution to existing problems. However, this perception is often shaped by its future potential rather than business reality and nuanced assessment in the industry.
It is crucial to distinguish between general AI and narrow AI. Most media outlets account for general AI, which is corroborated by science fiction movie scenes, where robot characters perform intelligent tasks at human-like level. The high uncertainty about general AI use cases often brings about an overstatement of the benefits and a fear-mongering agenda, while ignoring that technological change takes time to unfold and often requires entire ecosystems to adapt. Headlines like “AI is highly likely to destroy humans” purported by Elon Musk further push this agenda.
However, most of AI methods and tools, especially in supply chains and procurement, are currently based on narrow AI. While limited in thematic scope, narrow AI tackles specifically framed and defined tasks. Thereby, it possesses the “more immediate potential for improving operational efficiency”.
Thereby, AI should be seen more as a tool for optimizing process efficiency, rather than an all-in-one solution to all existing operational problems. To put it simply, it is a powerful aid for process optimization.
I’ll outline four different use cases, where AI can have a serious impact on procurement in the next few years.
With severe transparency problems along the supply chain, risk quantification is a crucial criterion for procurement departments to make buying decisions. Unexpected price fluctuations and impactful world events, for example, bring in several dimensions of uncertainty.
The Kobe earthquake in 1995 and Hurricane Katrina in 2005 is evidence to the recurring underestimation of business and supply chain disruption. Having killed more than 6’400 people, destroying 100’000 buildings, and forcing the harbor lockdown for two months, the earthquake caused more than $100 billion in damages. Toyota was unable to produce 20’000 cars, as it was forced to scramble for alternative production and transportation.
Deloitte found that many of the greatest market capitalization losses in the world were attributable to events that were considered extremely unlikely and the corresponding companies failed to account for in advance. Several companies noted in the study lost more than 20% of their market value in the month after the event. Often it took more than a year to recover and regain their original share level.
According to a Gartner subsidiary report from 2009, a supply chain disruption can cause operational costs to rise 11%, decrease sales growth by 7% percent, and decrease shareholder returns by 35%.
From a technical perspective, AI can be utilized to identify and monitor risk positions across the supply chain. The Munich-based company riskmethods has developed a supplier risk management tool that collects and screens millions of data points from dozens of public sources to quantify and anticipate purchasing risk.
Furthermore, classical machine learning methods would be able to automatically detect price anomalies or a change in credit ratings of an individual or set of suppliers, and model supplier switching scenarios accordingly. Moreover, a system could notify the procurement team if unexpected price fluctuations occur for a specific commodity.
The detection of these anomalies could be additionally empowered by scraping social media channels and interpret this text with the help of Natural Language Processing (NLP). It could searching for signals that pinpoint to a political, economic, or natural occurrence that could seriously impair a supplier’s product quality and ability to deliver.
Machine-learning methods could also recognize patterns in supplier data such as frequency, material quality, timing to identify a vendor hierarchy and connect local subsidiaries of a freight and logistics company to one international supplier.
In essence, spend analysis is the review of procurement spend data to decrease costs, increase efficiency or improve supplier relationships. This data includes — among others — repositories collected in Enterprise Resource Planning (ERP) tools such as SAP software, general ledger information, purchase orders, (meta-)data shared by suppliers. The quantification and concretization of key performance indicators (KPIs) characteristic for spend behavior sets the foundation for profound analysis. These KPIs include, among others: Spend by commodity or category, number of suppliers by commodity or category, key figures and reports regarding compliance, spending distribution of the key customers, material prices or price changes, payment terms and conditions, etc. etc. etc.
Many corporations face the challenge of classifying unique transactions into certain categories based on data from invoices, purchase orders, and other relevant documentation. For this reason, procurement departments create a complex hierarchical database with categories and sub-categories to organize and maintain this information. Oftentimes, this complexity impairs data quality and effectiveness during maintenance.
This database usually serves as a basis to perform spend analysis once or a couple times per year. In order to meet the high-frequency demand of today’s business, a real-time spend analysis update is required.
Sievo, a Finnish procurement analytics software company, developed a way to get a thorough overview of your spend for different categories. The solution visualizes key KPIs, uses natural language processing to categorize spend, and identify savings opportunities by switching suppliers. Let’s take the purchase of a new Dell computer for example:
It […] may be labelled as IT equipment in the general ledger, while the invoice line description provides additional detail distinguishing it as a laptop computer. The purchase order for this item may even have a different description, referring to vendor or maker specific data-points. While all these data sources refer to the same item, it requires intelligence to make a correct classification.
Being able to organize these products is an extremely time-consuming task that requires knowledge of each category and sub-category.
Procurement departments hire dozens of analysts who sift through Excel files to cherry-pick the optimal sourcing decision mainly based on internal business constraints (e.g. how many suppliers to work with in certain regions, how to diversify risks) and data from the supplier bids (e.g. volume discounts, capacity constraints, etc.). The analysis and evaluation of suppliers and the analysis of bids is time-consuming, and the focus often lies on the data processing rather than on effective negotiations.
Furthermore, written or visual reporting to support negotiation processes come short of the high-speed demands of today’s business climate. It is therefore key to be able to instantly create scenarios by adjusting internal business constraints and thus to find the most desired outcome.
The Swiss start-up Archlet has looked into solving this challenge with advanced statistics and combinatorics, offering deep analysis and creating different scenarios and recommendations for an organization to choose its suppliers. Other sourcing optimization solutions lack the ability to actively guide the buyer in the most intuitive way to the most desired scenario outcome. With more structured data becoming available in procurement (also due to the growing use of e-sourcing solutions), AI algorithms can search for risks and opportunities, in a much faster and more profound manner than human beings ever will.
Apart from being one of the only start-ups active in procurement in Switzerland and winning multiple start-up prizes since their birth, they are gaining heavy traction from the industry for one reason (in the context of my Toyota example from before):
Finding the right combination of suppliers offering the right price — while considering all the internal and external constraints — can save and organization tons of money.
Chief Procurement Officers (yes, that’s a thing) are — as macabre as it sounds — envisioning one control center for procurement that requires no human intervention: automated product request processes, automated quotation analyses, automated purchase orders, automated invoicing, and automated payment releases. In the eye of the CPO, the more decisions shifted toward hard data, the better. In my opinion, we are moving in this direction, however siloed softwares and information flows as well as departments filled with analysts understandably in fear of losing their jobs still pose huge barriers to making this a reality.
In the end, AI in procurement is a matter of process optimization, improving time efficiency and boring, manual work and relocate man hours from administrative, repetitive to work that involves more creative, human-human interaction.