During my three and a half years at Sievo, I have seen a vast number of spend cubes being created. While industries, clients and IT environments vary, the key success factors for a spend analysis project remain the same. Below, I have shared my 6 key learnings from the 100+ spend cubes that I have seen on the way.
1. The most critical part is getting the right data
No matter how sophisticated algorithms and fancy visualizations we do, none of it is relevant unless we have the right data in place. Data collection is the most critical phase of a spend analysis project. Large corporations typically have multiple ERP systems that have the spend data stored in different formats from system to system.
While it may be easy to technically extract data from multiple source systems, it is definitely not so from a business perspective. Each ERP system has its own logic of storing relevant purchasing data – invoices, POs, GL bookings and related master data. It takes procurement-specific knowledge to extract the data in the right way from each of these systems.
You may also wish to enrich the extracted data sets with data from your suppliers, purchasing cards or purchasing systems.
2. ‘Never send a human do a machine job’ and ‘Never send a machine do a human job’
Once you have the right data in place, it is time to cleanse and harmonize it – classifying spend to relevant categories, consolidating supplier names and translating the data.
As the title suggests – optimally, this is a collaboration between Human and AI.
Machine jobs: There are many things that are best handled by a machine: e.g., utilizing past classification decisions through a neural-network technology to identify the right categories for new spend or consolidating supplier names by comparing string-matches through algorithms designed for the purpose.
Human jobs: There are also human jobs that should not be outsourced to a machine. Especially in the case of direct spend, there are some technical purchases specific to an individual client. You should find a process for client experts to train the machine when needed to ensure classification success in the context of an individual client with their individual needs.
3. Use client/industry specific category taxonomies
I strongly recommend using client or industry specific category taxonomies instead of generic taxonomies, such as, UNSPSC. While generic taxonomies may be great in certain contexts, such as governmental purchases or public sector in general, most of the corporations have very specific needs in terms of how to structure their spend.
A good rule of thumb is to make each lowest-level subcategory sourceable in a single RFP and make top-level categories reflect client category management organization.
4. Improve data quality through your spend analysis
Some of the organizations do not do spend analysis in a systematic way, since they believe that their data quality is not sufficient for reliable results.
Well, that is the root cause for the whole problem. Only when you use the data, there is an incentive to develop processes so that they create more consistent data. You can also create a feedback loop to start fixing incorrect GL account or material group assignments as part of the purchasing process.
5. Utilize your results
So you have created the spend cube – remember to also utilize the results! Leveraging the results goes way beyond getting the basic visibility by categories and suppliers. You can get ideas for identifying opportunities, for example, here in one of our blog posts:
6. Business benefits are more important than utilized technologies
Of course, in the end it is the business benefits that matter. This sometimes gets forgotten when there is so much talk about robotic process automation, machine learning, big data, you know the buzz words. As a data scientist, I love all of that and could talk for the rest of the week only on how those can help you. But in the end, sometimes it is the simplest analysis that will help you the most.