Click the button below to start exploring the future of Procurement
Start exploring

How Big data will change P2P beyond recognition

In a previous blog post I spend some time on sharing my thoughts about how new technologies are going to impact business processes. I made a promise to present a concrete example of the application of Big Data, Artificial Intelligence and Machine Learning on P2P.

The world is constantly changing, and so are organizations. Thus it is logical that processes need to adjust to changing circumstances. Usually this doesn’t happen in the right pace, causing processes to pinch and people to invent workarounds to get around bottlenecks. Then, at a certain moment, when there is consensus that things can’t go on like that, an initiative is launched to implement improvements. That is a pity, since the consequence is that the process is sub optimal for a long time. Next to that, these initiatives tend to have a substantial impact on the organization. It would be great if processes – and systems – would adjust automatically to the changes in an organization. We think that this is possible and that this will become a matter of course.

A well-known bottleneck in the operational procurement process is approval of purchase requisitions or invoices. It often happens that it takes (too) much time before an approval is obtained or that the approver approves, due to time pressure, without really assessing the requisition or invoice. To disburden the approver, it is quite common that a threshold is used for approvals. For instance: requisitions (or invoices) with a value below €250 are approved automatically by the system. Basically, there is no check on the expenditure. Is there no way to do this more intelligent?

Imagine that an employee has created ten times a similar requisition: the supplier, cost center, procurement category and value are more or less equal. And imagine that all these requisitions were approved by the same approver. Statistically, it seems more than likely that the eleventh requisition will also be approved. If that’s true, then it is a pity that the approver has to spend time on it.

The sole reason for our desire to check each and every requisition is that we want to be in control. An organization can be asked by the accountant or the fiscal authorities to prove the legitimacy of certain expenditures. To be able to do so, time-consuming checks and balances are put in place. Partly this results in false security. It happens a lot that approvers are routinely approving requisitions or invoices, without really checking them, because they trust the approver (and in many cases rightly so). Officially you’re in control, but in reality this is a sham.

How can we improve the control, while at the same time increasing the efficiency (an apparent contradiction)? By using smart software that guarantees that you are demonstrably – statistically robust – in control, while it takes less time to attend to.

  • If – as in the example presented above – similar requisitions have been approved ten times, the software determines that future requisitions of the same kind are not all presented for approval. Instead, the approver only sees one out of each five requisitions. Until he rejects one, in that case the next three will all be presented to him.
  • This will also happen when the pattern suddenly changes. Instead of once a week, now four similar requisitions are created within two weeks. Again the software (more precisely, the smart algorithm that is embedded in it) will decide that it is wise to offer a few extra requisitions for approval.
  • But if that same pattern happens multiple times, then the program is smart enough to recognize that and to take that into account for future decisions about which requisitions to present for approval.

This is a very simple example. When you start to think of it, you will come up with numerous other variants. Or applications in other steps of the P2P process, for instance goods receipt postings or interpretations of price and quantity differences on invoices.

The point is that this is all possible: because we record so much data, we are able to combine data and search and recognize patterns. Smart software (artificial intelligence) enables us to make this software self-learning, so that it can recognize patterns, put a value to it and – based on that – make a change in the routine that is executed.

By measuring how people use systems and what the consequences are, we are able to offer information in a different manner. And if we log everything that happens in a smart way, it will always be traceable how transactions were executed and on which grounds certain decisions were taken. If we apply statistical formulas, we will be able to present “scientific” evidence that we are really in control. Something no controller currently dares to put his money on.