Using ERP data Makes a great Case for AI Invoice Analytics

“Any sufficiently advanced technology is indistinguishable from magic. ”
– Arthur C. Clarke

We can credit the phrase “Garbage in, garbage out” (GIGO) to George Fuechsel, an IBM programmer and instructor.

I thought it was appropriate to start this blog with both one of the all time great science fiction writers and also a quote attributed to an IBM programmer that I learned when I was taking Information Systems as a degree course at Carnegie Mellon many moons ago.

Most of us don’t have a clear idea of what machine intelligence really is. Perhaps we think its some vague notion of a learning computer system that is able to take data and develop some output decision.

AI Does Not Make Bad Data Good

But specifically, supervised learning is where you have input variables (x) and an output variable (Y) and you use an algorithm to learn the mapping function from the input to the output.  Eg, Y = f(X). And here we get back to data and the most important lesson of all: a machine-learning algorithm using incomplete, inaccurate or poor quality data is not going to produce quality results, no matter how good the modeling and programming is.

There are a number of companies that are racing to use data to do a better job of financing invoices and predicting dilution. This is not a trivial exercise. But some are trying it:

  • Previse can take three year historical ERP data from large corporates and determine from 20 or 30 features (e.g., country, amount, payment terms, category of invoice (consultant, airline), etc.) to score the invoice for payment.
  • Flowcast CEO Ken So believes that in 10 years, every financing decision will be driven by AI/machine learning.  Ken believes the diversity of data sources is thriving (from digital supplier onboarding data to 3rd party data API), providing a multi-dimensional view of performance risk.
  • Trustbills is focused on smaller companies with their Receivable finance solution and works with Deutsche Bank who offers access to their customers.

What makes ERP data so compelling for this exercise is that it is the basis of a company’s books, and you would hope the vendor master data is reasonably clean. But even here, large companies have grown both organically and through acquisition, and typically sit on multiple ERP systems and several different instances of SAP, Oracle, etc.

Supplier master data can be managed in a decentralized manner across a company’s global operating regions. This makes keeping supplier data up to date a challenging proposition for sure. At least the good news is that PO and payment data is likely to be more accurate than other vendor file information -- as its more important in this case.

Supplier Management Alphabet Soup

A vendor's goods and services or otherwise depends on the outcome of what they provide, regardless of whether these data points cover banking details, purchase orders, goods & services contents, performance management, or other compliance areas.

The supplier information management (SIM) field has more than its fair share of monikers which include SRM (Supplier Relationship Management), SPM (Supplier Performance Management), SLM (Supplier Lifecycle Management), SRPM, SBM and more. For a good primer on the subject, see Spend Matters Supplier Management Landscape Report and Supply Risk Management Landscape Report.

Having the vendor master and the data it contains is a great place to start, but there are many more endogenous variables a learning system can take into account.  You can have economic cycles, tax liens, industry data, and its important to have this data in addition to ERP data.

At the end of the early payment day, it’s all about dilution prediction – see Predicting Dilution is key for invoice finance   

And that comes back to the data:

  1. How much quality data do you have?
  2. Is the data structured the right way?
  3. How do you access external data?
  4. Are you buying data? Why would a company give it to you if it was a strategic advantage to keep it proprietary?

Here’s the issue – because of the amount of data used, just using ERP data itself may not be good enough. But it’s the right place to start.

Don't forget to sign up for TFMs weekly digest delivered to your inbox every Monday here.

Share on Procurious

First Voice

  1. Michel Kilzi:

    ERP data is tremendously important especially when it comes to Export receivable finance, as it gives us a basic granular view and visibility on trade history, delinquencies, concentrations and many other critical input. AI and predicting dilution and other predictions will be easier once the data becomes better structured and clean; blockchain / Hashgraph will play a major role here.

Discuss this:

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.