If you’ve yet to experience internal organizational skepticism or resistance to a machine learning (ML) initiative, you will at one point or another.
Like all technologies merging onto the mainstream highway, ML requires some intentional level-setting to be successfully utilized within an organization. By setting expectations and starting off with a few key best practices, you can raise your chances of success.
[ What's coming next? Read AI in 2019: 8 trends to watch. ]
How to level-set with machine learning doubters
A high-level overview of how ML works is the first form of level-setting and a critical step in dealing with the skeptics in your organization. ML uses different algorithms to learn from data – usually lots of data. The result of the learning or training process is an ML model, which has a specific desired outcome or job. Once the model is initially trained, it can process data and determine the desired outcome with some level of certainty or accuracy, usually measured by a percentage. Ideally, over time, the level of certainty or accuracy improves as the model is tweaked and more data is processed.
When you introduce an ML project into your organization, this process must be understood by the ML model’s consumer. I will call them the "dilemma owner."
Otherwise, if you tell them the ML model can solve their dilemma 97.3 percent of the time, they may not be interested in adopting the technology. Keep in mind, the dilemma owner might have a manual process that is 100 percent accurate, so 97.3 percent accuracy would be an inferior approach.
When sharing the ML model to the dilemma owner, focus on conveying the value and scale of this new approach. You have a solution that can automate 97.3 percent of the process, so only 2.7 percent needs to be done manually. Plus, it is likely that the accuracy of the model will improve over time and will scale to meet additional load.
This brings me to my second level-setting point. As a technologist, I like details, but they may have little benefit in convincing the larger organization to adopt a new technology like ML.
Stay focused on the dilemma the technology solves and its business benefit, rather than the greatness of the technology. If you lead with too much technical detail, you may never get the organization on board.
3 best practices to start smart
Like any other new technology or project that’s intended to address a market segment, product gap, or operational efficiency issue, a few best practices can get you started off on the right foot.
1. Be thorough
First, gather your business requirements for ML. This includes gaining a full understanding of the steps and thought process for the manual approach. For instance, if you need to perform an action on every image file, you need to know what you’re looking for and what to do once you find it. This gathering process is critical to building your ML model.
Without proper due diligence, you are likely to present a solution and hear in response that “the model didn’t catch [fill in the bank].” The logical reply, “you never told us about that,” won’t cut it.
Keep in mind that the dilemma owner is new to ML and may not understand what you need to train the ML model. It’s your responsibility to help decode what you need so that they understand what to gather. Requirements might morph a bit over time, but gathering as much detail up front as you can increase your chances of demonstrating value.
2. Gain support
Next, a new ML project requires sponsorship from senior leaders who are acutely focused on the dilemma. To get this, remind the dilemma owners why this ML model can help. It’s also important that sponsors stay engaged throughout the process to prevent the organization from backsliding to “the way we have always done it.” Give periodic email updates to a senior leadership sponsor to accomplish this.
3. Celebrate every milestone
Finally, define what a win looks like and recognize every success, no matter how small. ML projects are a bit like climbing a staircase. Each step gets you closer to where you want to be. Those steps might be cleaning up data, replacing manual steps with automation, or improving the accuracy of the output. These are all examples of tangible accomplishments that should be noted as you ascend the staircase. Doing so will go a long way toward maintaining the enthusiasm, focus, and momentum behind an ML project.
ML solutions are iterative in nature, and unmanaged or mismanaged expectations can propagate the mindset of “this ML stuff does not work.”
Focusing on low-hanging fruit and managing expectations can aid the acceptance of your ML project. By level-setting with dilemma owners, defining business requirements, recruiting executive-level champions of the technology, and finding quick wins, you can take advantage of all the benefits of ML with less resistance.
[ Want lessons learned from CIOs applying AI? Get the new HBR Analytic Services report, An Executive's Guide to Real-World AI. ]
