Why is MLaaS the best business model for Machine Learning?
Well the first question is, what does MLaaS stand for?
Machine Learning as a Service
The acronym comes from Software as a Service (SaaS) which is a fairly popular and standard software subscription model today. We use MLaaS to emphasize the difference between ML and legacy software business models, while keeping in mind that you don’t need to worry about the care and feeding of the ML. Rogue7 takes the responsibility for building a quality product and keeping it running at an optimal level, even through changes in your operation. Instead of just delivering some ML, we are motivated to deliver a quality product that continues to be useful (i.e. making you money) every single month.
For something like a spreadsheet or a word processor, some people would prefer to make a one time purchase and then use that version of software for several years before they consider making a new purchase. A lot of SCADA systems are purchased in the same way. There is so much cost and effort involved with upgrading and commissioning a SCADA system that they do not want updated software every year, much less every month. And that does not even get into the huge costs involved should a software upgrade break one of the critical functions of your industrial system.
There are two reasons that Machine Learning in an industrial automation setting is particularly suited for MLaaS.
The power of Machine Learning is still on the exponential growth curve. There are new techniques and new ML tools becoming available monthly. (A lot faster than your SCADA system replacement/upgrade). Taking advantage of the most recent advancements can give you a substantial advantage. Rogue7 is better able to keep up with this fast growing technology than most oil and gas companies.
Machine Learning benefits (or learns) from the myriad of data that your industrial system generates every single day. Through continuous ML algorithm improvements and technology stack upgrades, the Rogue7 service model delivers a product with cutting edge performance. This evolving and nimble approach allows for the incorporation of new data on a regular basis to make sure that the ML product is improving every single week, while also making sure that the proper lessons are learned from shutdowns and failures.
Some ML companies would make you believe that you need to have one or more (10?) PhD(s) in Machine Learning to build and keep a ML project functioning properly. Rogue7 has focused on developing ML products instead of implementing ML projects. An ML product:
Has a defined business model, and payback period.
Fits a specified industrial niche (not a general category).
Utilizes a general type of data without requiring any specific type of data source.
Needs to run on an industrial grade platform that is both redundant and secure. What good is ML if it is unreliable, if it is not available when you need it?
By using MLaaS, you and your service provider enter a partnership, and every month they will prove to you that the ML products are making/saving you money.
-Douglas Fisher