Classifying Spend Data – Do Algorithms Really Work?
People who are new to spend analysis are trained to think that there is a magic button to classify their spend data. If I send you my spend data you run it through your algorithms and magically my spend data will be classified perfectly for my business.
The people that have been through failed spend implementations know that there will always be subjectivity in data and an algorithm that works for one client does not work for another.
I am not saying that algorithms do not work but you have to factor in the complexity of your business. If vendors say they have hundreds of thousands of rules, you have to ask, “How do those rules get applied to my spend data?” On average, we create anywhere from 500-2000 custom built rules per project.
Another area of concern when dealing with a classification engine built around algorithms is that change is not easy.
You cannot change a standardized rule because it is just that – a standardized rule that needs to work across multiple industries classifying the data directionally correct.
Directionally correct is not good enough when it comes to running initiatives.
Accuracy is critical to gaining a clear view into your spend.
Sourcing Force was recently working with a client where we ran into a dataset with high level Accounts Payable data and supplier spend that needed business specific categorizations.
To give a little perspective on the client we were working with, they are a holding company with 20+ operating companies.
In this client’s case maybe on a portfolio by portfolio company basis algorithms would work, but doesn’t that defeat the purpose of implementing an entire holding company?
A prime example of classification subjectivity within classifications would be a rule for Cintas. Cintas has a primary Industry listing of Uniform Rental & Laundry Services.
So for all purposes an algorithm would classify this spend with UNSPSC to the code best suited to uniforms. This classification may be correct 95% of the time if you are implementing a manufacturing company. But let’s go back to our current implementation of our holding company.
Within one of the operating companies in this holding company Cintas was classified to uniforms, floor or area mats, document destruction, and one that no algorithm would ever catch, this company was buying back the paper that they were shredding to recycle into their products they are producing.
Again for this project we only have AP data.
So how is it expected that a system built on algorithms is going to classify data based off a supplier name and spend amount?
An algorithm may give you a start to where the classification is directionally correct but until you understand the client you are working with and how they buy you will never feel confident that you have accurate classifications.
Classification is an area where Sourcing Force stands out! With access into the Rule Builder which allows for any data field to be classified it takes the guessing out of classification and puts the confidence in it.
With our functionality you have complete control of your spend!
Spend Analysis the heart of spend management initiatives, shouldn’t you have an accurate view of your spends?