5 Ideas to Maintain Senior Executive Involvement for Machine Learning
How to keep executive teams engaged throughout the machine learning journey in a highly regulated business environment
Imagine that your CFO just reported that the latest advance in machine learning had saved 30% operating expense at your competitor. However, your executive team is not pulling along with internal ML improvements?
Do not despair. It happens everywhere. Here is a collection of tales about failed and successful attempts to keep senior executives engaged and supportive of ML initiatives.
There are many, many, many, like really a lot of guides on the project and technical aspects; organization culture and design.
Now, this article is neither a guide nor about “how” to execute the project. This is about keeping the executive team on track.
Machine learning or analytics?
First, establish that machine learning is required. In plain language, this boils down to:
- There are many decisions to be made
- Using previously unobserved data
- Make decisions over a sustained period
- Generalizable patterns can be established
- The organization would find it acceptable for machines to decide
The last point is where many organizations struggle most. Learning a great model for credit approvals and then presenting the list of suggested credit approvals to a human committee is not a great use case.
Sure, there needs to be some form of oversight. Model monitoring metrics presented to a human committee may well serve that objective. However, that requires a bit of lateral thinking — and lots and lots of explaining of these metrics! Adopting new technology has never been easy, and it is a safe bet that the same is true for machine learning!
Break the ice
Getting a team to lose pre-existing thinking patterns is particularly difficult. Humans have a hard time unlearning some of their experiences — No matter if the experience is relevant not.
Especially experiences incompatible with machine learning, such as a desire to control individual outcomes, are hard to lose.
It cannot be over-emphasized how tedious the unlearning is. To quote Kaaren Hanson, presently JPMorgan and former Facebook design leader:
“A lot of what we do is habit, and it’s hard to change those habits, but having very clear guardrails can help us.”
Establishing trust that the ML initiative’s team can deliver is a powerful way to win over the executive team. But alas, it can take time!
For one model, it took several months. The team spent countless hours defining and re-defining a sensible decision metric. The effort involved engineers and domain experts, complete with frustrations, misunderstandings, and calculation errors. Several months of iterations with executive leaders followed the basic model design. Many nuances of the metric were discussed in individual and group meetings as well as steering committees — all of that against a regulatory deadline. Even though an acceptable metric eventually was agreed upon, the process wasn’t easy and not a great example of breaking the ice.
A far better way to break the ice involves a design thinking approach.
An entirely unscientific but time tested approach looks roughly like this:
- Get the executives out of daily work for two days and focus just on the AI/ML initiative, e.g., drop phones before entering the room, to be fetched only in case of actual calls.
- Run the idea and vision creation process so that everyone on the executive leadership team gets to contribute.
- Test and reality check assumptions in real-time with engineers in the room (but don’t let that restrict the idea generation!).
- Ensure candid but data-driven discussions by leaving rank, title, and experience out of the room through rituals — e.g., write them all on a paper slip to be picked up after the meeting.
- Quickly produce tangible workshop output with agreeable objectives (read: measurable and understood metrics?) to hang on the office wall — No more than two weeks after the event.
- Celebrate the workshop outcome in internal media, memos and congratulate people after the workshop conclusion.
Once off to a good start, the next challenge is keeping the momentum. After an exciting off-site and reflecting on what this new initiative means for every participant, some may feel a little uneasy.
Remove the fear factor
Like every other employee, executives might be scared to lose their jobs too. After all, making decisions is very closely associated with the very definition of management according to Oxford dictionaries:
“the process of dealing with or controlling things or people”
It would be rather unexpected if people who have grown accustomed to this would not be a little disoriented when suddenly a machine should do part of the work.
This vacuum of “purpose” must be addressed. Similar to a change towards an agile working model, executives will have new and changing responsibilities when some decisions are left to machines. To help executives get their bearing in this changing environment, they may consider:
- Precise definitions of the new roles, e.g., executives are accountable for their initiative, but actual decisions and actions happen within the squad… or an ML model.
- Hands-on training on servant leader culture, e.g., celebrating executives who have unblocked issues raised by squads… or algorithms to the executive board.
- Tools to coach their teams, e.g., draw up a straightforward role, process, and activity description and resist tailoring more than necessary.
- Indication of consequences if executives fail to adapt. Leaders that cannot manage in this environment may be great in a different setup, don’t force people to stay for the wrong reasons; help the transition.
If this sounds much like agile, that is precisely the point. Agile methods are a great training ground for ML/AI adoption readiness.
However, be aware: With the ML/AI initiative substantially rooted and steaming ahead, often competition between executives gives rise to unintended consequences.
Make it fun
Chances are, that executive team members had to wage war with countless adversaries inside and outside the firm to gain their present position. Healthy competition is somewhat encouraged at many firms. However, with regards to ML initiatives, cooperation wins the prize. Setting incentives for collaboration, rituals, and small celebrations can help to lower problematic competitive behavior by:
- Celebrating learning from failure to avoid being told how great the AI initiatives are when it’s not doing anything useful
- Rewarding sharing and cross-team achievements more than initiatives by a single area
- Establishing regular light-hearted/gamified rituals on the heavy technical aspects, e.g., replace “high accuracy” with “King of the hill” and “low accuracy” with “Dirt digger.”
Finally, there is one more thing to try.
Be a champion
The power of executive engagement is not new. However, it is more important for AI and ML initiatives than for other projects. While many initiatives need cross-functional cooperation, ML initiatives need cross-functional acceptance, and tolerance is not enough. Individually speaking to ML teams, learning about blockers and concerns from users firsthand is vital. Even if the firm maintains an open culture, several concerns may only be shared in private for fear of organizational retaliation.
In addition, being personally involved carries a powerful message to the executive team: “This is important. Supporting this is OK!” Even defining specific rewards for collaboration between executives for ML initiatives may be an option.
Closing remarks
Intentionally brief and opinionated, far from being a guide, this may inspire how to focus beyond the initial hype cycle of a project. While some example action points have been indicated, different measures may be more suitable in other circumstances. Any comments are highly appreciated.
That said, I firmly believe applying these techniques will help increase the odds of becoming that company using ML effectively to, e.g., reduce operating expenses. Nothing can guarantee that some ML projects will still fail, but I hope to have initiated some practical thinking to improve the success rate and make the journey a little more enjoyable.
