October 02, 2017
Cognitive Procurement: Machine Learning Deepens Visibility for Smart Business Decisions
By: Ely X. Colón and Connor Sanders

Most procurement organizations leave value on the table by not leveraging vast amounts of data. The inability to make strategic and tactical decisions using comprehensive data sets can erode trust between internal stakeholders and customers, thereby reducing procurement’s value. Machine learning, a subfield of artificial intelligence, can transform decision-making based on data that is more expansive and accurate than ever.

Machine learning gives computers the ability to learn without being explicitly programmed. Algorithms can explore vast oceans of data in procurement organizations, including public, private and company-owned information (e.g. spend data, contracts and market intelligence).

The burden of complexity
Many oil and gas organizations are challenged to quantify how much is spent on particular goods and services because they do not categorize these thoroughly. Companies may have several hundred million dollars misclassified or grouped in a general bucket (“other” or “unclassified”), which prevents a great deal of spend from being managed effectively.

The usual methods are to classify spend manually, or to use a rules-based software on supplier names and keywords in the free text. These processes are labor-intensive and typically result in accurate classification of up to 85 percent of spend. This approach also limits sub-categorization due to reliance on supplier names vs. material descriptions.

It is not uncommon to find from 200,000 to 400,000 unclassified line items in any given typical set of annual spend data (from a total of 800,000 to 1.2 million line items). Because of the labor-intensive nature of categorization processes, companies typically complete this exercise only every two to four years, at best, limiting visibility into spend and decision-making capability.

For global organizations, the problem is compounded as people from various regions and business units input using different nomenclature (e.g. vendor names, detailed line-item descriptions). This added complexity exacerbates the problem, as the primary challenge with misclassified or unclassified spend data is the lack of rich “text” data, or consistency of that textual, unstructured information.

Algorithms to the rescue
Machine learning helps enable automatic categorization of text-based data through algorithms from specific fields such as the material, supplier or line-item description. The more data processed through the algorithm, the more accurate the classification predictions become.

As a result, companies can improve visibility into 96 percent (or more) of spend data in real time, which would represent an improvement of 11 percentage points, or $110 million of new classified spend for every $1 billion of spend.

Potential Benefits of machine learning

  • Reduces tedious processes of manually examining data, which frees up time for strategic work

  • Improves decision making by augmenting classified spend rate

  • Increases spend leverage, as more categories are classified and can be aggregated to enhance buying channels

  • Creates efficiencies through improved catalog coverage, spend transparency and insight

  • Decreases maverick spend, leading to higher contract coverage and supplier rationalization

  • Accelerates the material-master creation cycle and eliminates duplication of stock-keeping units (SKUs)

Machine learning can be adapted and scaled to suit the needs of each oil and gas organization, and what once took weeks to categorize now takes hours.

Popular Tags

    More blogs on this topic