The people, process, and technology impediment to making ML work in your business
Depending on who you speak with, you will hear lots of opinions on whether ML is ready for broad enterprise adoption, or whether the true promise of ML for most companies is still several years away. That said, the market is exploding with new ML software startups aimed at automating and democratizing ML for the masses. This is driving lots of VC investment and market momentum, but the customers I speak with generally feel unsure about where to start, how to make sense of all these new technologies, and whether or not they have the people and culture to be successful.
Frankly, I don't blame customers from being hesitant (and unclear) about how to operationalize ML in their business. This stuff is complex, and it's not just a question of people/expertise and technology. I would argue that equally critical (and daunting) is the process component, and unfortunately there are not many technologies on the market that present a "magic" solution for this. ML process expertise is born from hard won experience. It is the result of extensive, and often painful, trial and error. What is super challenging is that there is really no one size fits all approach to this. The ML process that is right for you is dependent on your team's experience level, what you are trying to accomplish, what technologies you are working with, and your data landscape.
In our opinion, given you have the right approach, ML is very much ready today and can be leveraged to do incredible things such as prediction, optimization, process automation, fraud detection, signal analysis, and more. However, it is imperative that you nail these 3 foundational components in a tightly coordinated way: people, process & technology.
I know this may sound a little cliche, but we see this proven time and time again, and we have gone as far as to build our ML consulting methodology around these components. Here's how we go about this.
When we are first working with our customers we start with a focus on understanding the business objective. Are you trying to improve manufacturing yield? Optimize distribution of goods? Predict fraud or churn? We always start here, with the "what", and never with the technology or the "how".
With the business objective established, we begin looking at the people, process and technology components. Let's explore these at a high level:
People. What does the team look like? Do you have data engineers on staff? Data Scientists? If so, what is their experience level; junior or senior? Is there good executive support and buy-in for the project? Are they well aligned with your IT team and the line of business?
Process. Have you successfully deployed machine learning models into production before? Do you have a process that works for one model, will it scale to N models? Do you have a way to easily monitor performance, manage and iterate?
Technology. This should really be the last consideration, dependent on the people and process reality. You may have some technologies in place, and you may have needs. For example, if you haven't done much with ML yet, are you doing any data/analytics, or big data projects? Do you have data centralized in a data lake or data warehouse? Have you made any efforts around data prep? Ultimately you will need technology that works for you to help with each step in the ML lifecycle; Data>extraction/pipelining>feature engineering>model development>deployment>monitoring/management.
With the right people, the right process for your business, and the right technology building blocks, effective machine learning becomes incredibly accessible for every business. We even built our own feature store and ML lifecycle management platform called Whisky.AI because most of our customers simply didn't have the depth of experience, nor the time to spend doing trial and error on the process. Whisky was built to give our customers a prescriptive process framework to scale ML in production that works with any underlying technology components, and gives the data scientist a pure working environment by abstracting the underlying data engineering.
Bottom line, when you are thinking about leveraging ML in your business, start with an honest self assessment on your people, process and technology...and put the tech last. This is the best way to demystify ML and get an approach that will be successful for your business. Oh, and if you need any guidance give us a call - we would love to help you develop the right approach to get you the best outcome most rapidly.