Practical Application of Machine Learning in Contracting

I’m confident machine learning is going to play a larger role in all knowledge operations in the future.

I found a book I think is at the right level for us to start thinking about the practical application of machine learning in Contracting.

Here is the book’s definition of machine learning:

For the purposes of this book, think of machine learning as a way to arrive at a decision based on patterns in a dataset. We’ll call this pattern-based decision making . This is in contrast to most software development these days, which is rules-based --where programmers write code that employs a series of rules to perform a task.

Two things make this book awesome:

  1. The first practical example in the book is from a procurement organization. It uses machine learning to help automate a decision about whether a purchase request needs IT department approval.

Let’s take Karen, for example. Her job is to review purchase orders, send them to an approver, and then email the approved purchase orders to the supplier. Karen’s job is both boring and tricky. Every day, Karen makes dozens of decisions about who should approve which orders. Karen has been doing this job for several years, so she knows the simple rules like IT products must be approved by the IT department. But she also knows the exceptions. For example, she knows that when Jim orders toner from the stationery catalog, she needs to send the order to IT for approval, but when Jim orders a new mouse from the IT catalog, she does not.

The reason Karen’s role hasn’t been automated is because programming all of these rules is hard. But harder still is maintaining these rules. Karen doesn’t often apply her “fax machine” rule anymore, but she is increasingly applying her “tablet stylus” rule, which she has developed over the past several years. She considers a tablet stylus to be more like a mouse than a laptop computer, so she doesn’t send stylus orders to IT for approval. If Karen really doesn’t know how to classify a particular product, she’ll call IT to discuss it; but for most things, she makes up her own mind.

Using our concepts of rules-based decision making versus pattern-based decision making, you can see that Karen incorporates a bit of both. Karen applies rules most of time but occasionally makes decisions based on patterns. It’s the pattern-based part of Karen’s work that makes it hard to automate using a rules-based system. That’s why, in the past, it has been easier to have Karen perform these tasks than to program a computer with the rules to perform the same tasks.

  1. It uses new tools that don’t require you to be a data scientist and programmer to use machine learning in your organization.

This book shows you how to implement machine learning decision-making systems in your company to speed up your business processes. “But how can I do that?” you say. “I’m technically-minded and I’m pretty comfortable using Excel, and I’ve never done any programming.” Fortunately for you, we are at a point in time where any technically-minded person can learn how to help their company become dramatically more productive. This book takes you on that journey. On that journey, you’ll learn

  • How to identify where machine learning will create the greatest benefits within your company in areas such as

    • Back-office financials (accounts payable and billing)

    • Customer support and retention

    • Sales and marketing

    • Payroll and human resources

  • How to build machine learning applications that you can implement in your company.

Right now, you can read the 1st chapter for free here:

1 Like

I think with a little creativity, all of the examples could be applied to a contracting/acquisition decision or situation.

The book consists of seven chapters that take you through seven scenarios that will equip you with tackling many of the scenarios you might face in your own company including the following:

  • Should you send a purchase order to a technical approver?

  • Should you call a customer because they are at risk of churning?

  • Should a customer support ticket be handled by a senior support person?

  • Should you question an invoice sent to you by a supplier?

  • How much power (funding) will your company use next month based on historical trends?

  • Adding additional data such as planned holidays and weather forecasts to your power (funding) consumption prediction.

  • How to serve your models over the web using serverless technology.

After reading a few chapters, I think we need to start small to get good at recognizing opportunities and implementing the more simple solution rather than trying to automate our entire process at once.

If you’re interested, at least read the first chapter which covers a lot of foundational machine learning concepts in plain language.

1 Like

How about a suggestion service for strategic sourcing vehicles based on NAICS and other factors?

  1. Create a database of Strategic Sourcing vehicles and the associated contract numbers
  2. Automate a query of FPDS-NG data using the contract numbers
  3. Train the machine learning model to suggest / rank contract vehicles based on NAICS and other factors people can submit through a form or API.
  4. Long-term, connect the API to KT File Share and add a tab/section for suggestions

Theoretically, the suggestions would get better as more people used the various contracts and report them through FPDS-NG. I wonder if maintaining the associated contract numbers would be harder than trying to maintain the associated NAICS.

Does anyone (maybe @Mark_Wagner?) know if someone is already maintaining a database of strategic sourcing vehicles or working on a similar project?

1 Like