Back to Hub

How to Hack Your ERP and Create Competition for P2P Suite Providers

11/02/2018 By

Spend Matters welcomes this guest post from Doug Hudgeon, a business automation expert.

P2P software is painful to buy and painful to implement.

In order to get the biggest benefit, you need to rip the purchasing spinal cord out of your company and replace it with an end-to-end P2P system.

This is a big project, requires a big budget and, if you are going to deliver on the promised benefits, you must have an unwavering commitment to change. Sales cycles are long but, fortunately for P2P vendors, the margins on the deals that do get across the line can be pretty good.

But the competitive landscape is about to shift.

The combination of robotic process automation (RPA) software, augmented APIs and cloud-delivered machine learning (ML) solutions enable you to cobble together just the pieces you need to get most of the promised benefits of a full P2P system at a fraction of the cost.

In a recent bid I conducted, a consortium comprised of an integrator, an RPA vendor and an augmented API (invoice data extraction service) came in at half the cost of the next lowest bidder. They were able to hit this price point because they made use of existing functionality in the customer’s ERP to manage things like the approval hierarchy and they proposed targeted solutions for the missing functionality.

The promise of these solutions is that instead of ripping functions out of your ERP and putting them into other systems, such as putting purchasing into a P2P system, you can bolt on just the bits your ERP system doesn’t do well:

  1. RPA systems such as UIPath, Blue Prism and Automation Anywhere make it easy to get data into and out of your ERP system.
  2. Augmented APIs (task-specific services that come bundled with a user interface) such as Xtracta allow you to bolt-on all of the steps required for single discreet task, such as invoice data extraction.
  3. Cloud-based machine learning systems such as Amazon Sagemaker, Azure Machine Learning Studio and Google Cloud Machine Learning Engine enable you to quickly automate much of the decision-making, such as approval routing. (See links below.)

So what does this look like in practice?

The figure below shows a typical tail-end of the P2P process (invoice receipt to ERP) implemented using RPA, augmented APIs and cloud-based machine learning.

The workflow kicks off in step 1 with an invoice arriving by email. An RPA process picks up the invoice and sends it to the data extraction service (step 2). I describe the data extraction service as an augmented API because it combines a cloud-based server (API) using machine learning to extract data from invoices and a single-screen user interface that allows you to train the machine learning algorithm on invoices it doesn’t recognize (step 3). In step 4, a cloud-based machine learning system picks up the data and runs it through a couple of machine learning models. The first model determines who should approve the invoice, and the second classifies products for spend analysis. Step 5 is the spend analytics solution. In step 6, the RPA process routes the invoice for approval, and finally in step 7, the RPA enters the invoice automatically in your ERP system.

Note that this process is for illustration purposes and that there are lots of elements that haven’t been called out. For example, PO data and vendor information needs to be loaded into the cloud-based machine learning model in step 4 but is not shown in the diagram.

For want of an existing name, I call RPA, API and ML solutions “PRIMAL” because it’s nearly an anagram. In effect, RPA, augmented APIs and ML allow a company or an integrator to put together a pretty good approximation of a P2P system fairly quickly. The effect of this will be two-fold:

  1. Buyers bidding for a P2P solution will start including PRIMAL solution integrators in their bids. Even if they ultimately select a P2P software vendor, they will be able to extract a better deal from their P2P vendor because they will be a better-informed buyer.
  2. Buyers will request modular delivery of P2P solutions that will reduce deal size and increase deal complexity for P2P vendors.

Neither of these effects will be good for P2P vendor margins.

So what’s a P2P vendor to do? I see two possible courses of action: They will start making their application more modular and allow their customers to mix and match — or they will resist modularization and push the benefits of a fully integrated suite.

The smaller P2P vendors will have to follow the first course of action. But larger P2P vendors like Ariba and Coupa can pursue the second course of action.

Key to their ability to succeed in the second way will be the size and effectiveness of their supplier network because it’s a key advantage over PRIMAL systems. I’m looking forward to seeing how all this plays out. I suspect it will be good for those in the market for a new P2P system.

Doug Hudgeon is the author of a new book from Manning Publications “Machine Learning for Business,” a hands-on guide to build and deploy machine learning applications using Amazon SageMaker. On, you can use the code spendml4b40 to get a 40% discount at checkout.

Links to PRIMAL applications:


  • UIPath
  • Blue Prism
  • Automation Anywhere:

Augmented APIs

  • Xtracta

Machine Learning

  • Amazon Sagemaker
  • Microsoft Machine Learning Studio
  • Google Cloud ML Engine