In computer science, the term artificial intelligence (AI) refers to any human-like intelligence exhibited by a computer, robot, or other machine. In popular usage, artificial intelligence refers to the ability of a computer or machine to mimic the capabilities of the human mind — learning from examples and experience, recognizing objects, understanding and responding to language, making decisions, solving problems — and combining these and other capabilities to perform functions a human might perform, such as greeting a hotel guest or driving a car.

AI in Procurement

In order to compete effectively in an increasingly complex global market, many companies of all sizes have embraced a variety of technologies as part of their pursuit of digital transformation. But not every area of business has proven quite as eager to accept emerging technologies like artificial intelligence (AI), real-time analytics, or process automation. For many businesses, regardless of industry, one area in particular — the procurement function — has, until recently, persisted with traditional technologies and processes. In doing so, these organizations may be sacrificing essential gains in efficiency, accuracy, and strategic decision-making that could help them build or maintain a competitive advantage.

Let’s look at some examples…

Example 1: Spend Classification

How to classify spend into procurement categories is one of the oldest and most common applications of AI in procurement. Many procurement organisations follow the ‘Category Management’ model of procurement whereby every purchase is mapped into an often complex hierarchy of categories and sub-categories.

A particular purchase is sorted into a category based on data coming from invoices, purchase orders or other data sources. The technical challenge of this is not the AI itself but rather bringing disparate data sources together to provide as much information as possible to whatever classification algorithm is used. This is because with AI your solution is often only as good as your data.

A common type of classification in AI in spam detection.

Is Spend Classification Useful?

Spend classification is an attractive solution as it enriches and brings some order to an organisations data, this can help when analysing the data further. There is however no direct value add from classifying your data in this way and it is not a perfect solution as even though you have classified a category you have not standardised the way in which information is reported within the category itself so in-depth analysis is not possible.

Beyond Spend Classification…

Axiom uses spend classification algorithms to classify the data we gather and then use information extraction algorithms to fill in standardised forms with any information available in internal databases or external data sources (e.g. environmental and social responsibility measurements). The AI will also try to fill in any missing fields by looking at your past purchasing behaviour. By doing this we reduce the amount of manual input required to accurately specify a product. Once a product is accurately specified, a useful and thorough analysis can be carried out.

Example 2: Supplier Performance

An organisations procurement performance relies on the performance of its suppliers. Organisations will try to identify poor performing suppliers using their own metrics along with ad-hoc analysis. The metrics used are often too qualitative and the analysis over generalised or statistically insignificant.

Adhoc, overgeneralised, statistically insignificant

An Example of Poor Analysis

When a procurement professional is comparing two print suppliers (Supplier A and Supplier B) and trying to determine which is better they may start looking at the cost of certain products on an ERP dashboard. They may observe that for 75% of similar orders of leaflets Supplier A has been less expensive than Supplier B. The professional decides that Supplier A is better and the organisation should no longer use Supplier B.

This is an example of over generalised analysis, the possible reasons Supplier B has been more expensive are numerous and quite often due to the specification of the order, date of when the order was placed or poor/no negotiation with the supplier.

Analysis with AI

Axiom uses advanced statistical models to rank suppliers based on price, quality, risk, etc. down to a specification level so that for every order the customer knows which supplier to go to and how much they are likely to pay. These models allow Axiom to continually review and automatically flag suppliers who are performing poorly.

Example 3: Price Prediction

Procurement purchases most things either by strategic, tactical or spot sourcing. For most organisations, strategic sourcing makes up the majority of purchases. These purchases are those that are planned in advance, take into account company requirements and supplier capabilities and usually have long lead times. Availability and price are considered in the decision, but equally important is the impact on the overall organisation. Finally, a strategic sourcing model encourages communication and aims to keep that communication open throughout the contract lifecycle.

For tactical and spot sourcing the primary factors of consideration are price and delivery time. As there is a low average order value there is often a huge amount of budget wastage in these purchases since they are not given a lot of attention and procurement professionals aren’t keen to spend their time trying to find the best price and the suppliers know this!!! The variation in the price of a small order (<$10,000) placed the same week from the same supplier can frequently be more than 300%!

Past Solutions

  1. Asking for quotes from 2–5 suppliers.

By asking for a number of quotes from different suppliers you are effectively sampling the distribution of prices and therefore making it less likely that you will accept an absurdly high price. This method however is cumbersome since to actually get to a fair price you have to contact many suppliers and in truth, you never know when a price is actually fair.

Sampling the distribution of prices

2. Price prediction for a particular repeated specification.

Some procurement software solutions offer a basic price prediction capability that uses historical purchases for a particular item at many volumes and then fits a polynomial curve through the points using some unknown loss function. This method is fraught with error mathematically as the curve is likely to fit the data poorly due to a small data set size (each specification is treated as its own data set), a small signal to noise ratio within the data and may other reasons. Practically it is a poor method as in order to predict the price of something you need to have purchased it many times already.

Axiom’s Solution

Using advanced statistical methods Axioms TruePrice takes into account every order from every supplier when predicting the price a customer should pay for any specification, whether they have purchased it before or not. The customer decides what they think is fair to pay in the distribution of the market prices, whether that be the 25% or the 75% centile. The customer can then go out to suppliers with knowledge of what a fair price should be and in our experience be able to tell the supplier what they are willing to pay and not ask what the supplier wants them to pay.