Replacing your enterprise resource planning (ERP) system can be traumatic—personally and for the company. Upgrading an existing ERP system is almost as risky. Many ERP systems are so heavily customized that no one wants to take the chance of moving versions. Still, companies and people feel the need to enhance the functionality of their enterprise software. Envisioned productivity gains eventually outweigh the risk.
Add-ons
Indeed, many services and additional applications are purchased over the years to work with ERP data in ways the ERP software does not. This is somewhat akin to upgrading the ERP system. The company gets additional functionality and productivity gains using the ERP’s database with much less risk compared to upgrading the core ERP system.
Most of these add-ons come from other software and service vendors. Often, these become mission critical, and the same trepidation of upgrade is felt.
Machine Learning Can Be Different
The reality of today’s ERP software market is that your ERP vendor is likely working on some machine learning (ML) functions to build into upcoming upgrades. Every software vendor needs some artificial intelligence (AI) story, and ML is really a subset of AI that is good at working with enterprise transactional data.
ML seeks to find patterns, trends, and correlations inside of existing data sets. For businesses, that includes ERP databases.
In the past, several high-priced software systems allowed skilled analysts to apply statistical algorithms. These were complicated and expensive. Most companies chose to do that type of analysis on a smaller scale inside of spreadsheets. Like all spreadsheets, sharing and communicating any insight gained was a manual process.
Modern ML works well with other applications. Integrating ML models into business processes is not easy, but it is not technically hard. The tools and functions needed are readily available as open-source software. Finding people skilled in using it is not easy, but the supply of college graduates with classwork in basic analytics is increasing every year.
Roadblocks
The biggest roadblock to deploying ML in most organizations is cultural, not technical. Business processes and people are accustomed to doing things a certain way with a certain set of tools. ML attempts to take part of the analysis process out of the human brain (or spreadsheet).
Getting people and management to trust the insights or predictions coming from the ML system, instead of their trusty spreadsheet, is often the hardest part of making ML operational.
The most common tools for ML are open-source languages and platforms (e.g., R, Python, Jupyter notebooks). Many college students get classroom experience working in these environments, and new college graduates often have a little background in them—even ones from nontechnical majors (think marketing majors and their need to do data analysis).
ERP and ML
Most organizations have people with some basic ML skills. Having those people work with ERP data is not as difficult as it used to be.
I view the limited technical experience of people in the organization as a plus. A small project team will not try to do too much. They should be aware of their limits and seek to keep their project deliverables small and achievable.
Back to my main point—creating a small project using basic ML skills and a focused portion of the ERP data is a quick win for an organization. No additional vendor is required. No outside service needs to be contracted. The data in the ERP system are not being overwritten or erased.
The ML models copy the data, run some basic algorithms, and make some predictions or generate some insight. That is to the side of the ERP system. That insight is just as valid as if the ML had been built into the ERP system. The organization still needs to get people to accept the validity and usefulness of the ML output, but that is no different than if the ERP system did the same ML.
Do a few of these targeted ML projects using ERP data, and it could be the equivalent of upgrading the ERP without the risk, expense, or hassle.