This content is locked. Please login or become a member.
Business vs. Technology
Predictive AI or enterprise machine learning is sort of the last remaining point of differentiation for these large-scale, corporate enterprise decision-making. The problem is and the reason that we have so much untapped potential in that these projects actually fail to deploy comes down to both sides, both the technical and biz, both the data scientists and their stakeholder, the person in charge of whatever large-scale operations are meant to be improved with a model. Both sides are considering these technology projects, but they’re not, their business endeavors. They have to be seen as operations-improvement projects that use machine learning as a core technology.
Now both sides kind of say, “No, that’s not my job.” They point to the other side. So the data scientists say, hey. Look, I’m the subject matter expert. I know how to do the number crunching. I learned from the data with this machine learning software. It generates a model that predicts. I know how to do that stuff. Of course, it’s going to be deployed. The other stuff’s just managerial details and issues that are outside the scope of my work. It’s not my responsibility. It’s a no-brainer value. The organization, of course, it’s going to act on these predictions.
And yet, at the same time, the business stakeholders are saying, woah. Woah. Woah. Hey. Probability? I delegate all that stuff to data scientists. That’s why I have data scientists. You know? Don’t don’t bother me with that level of detail. As a result, the faucet and the hose fail to connect. So we need to reframe and get everyone on the same page. But to do that, there needs to be this sort of, we have to bridge this gap, and run the project from a semi-technical understood vantage.
Advice for Business Stakeholders
Business stakeholders put their hands up and say, look. I don’t need to look under the hood in order to drive a car, which is totally true. And I personally also don’t know much about engines. I know the general idea of internal combustion. However, I am an expert. I know momentum, friction, rules of the road, how the car operates, and the mutual expectations of drivers. The same type of semi-technical expertise is needed in order to drive enterprise machine learning projects successfully through to deployment. And it comes down to this. What’s predicted, how well, and what’s done about it. So, for example, predict who’s going to buy if contacted for targeting marketing. What’s done about it is contact those relatively more likely to buy. And then how well is just arithmetic, but you have to put the right numbers, the right metrics in terms of understanding how well it predicts and how much value acting on those predictions will deliver to the enterprise.
Advice for Data Scientists
Data scientists also need to make some changes to bridge that gap. They need to take more ownership of the value of their work. It’s not just number crunching. You have to make darn sure that the stakeholders and the decision-makers understand that it’s going to take change to operations. They need to work with the stakeholders to make sure they understand the gist of the project, what’s predicted, how well, and what’s done about it.
So, for example, in terms of how well it’s predicted, you can’t just have these arcane technical metrics. The most popular one is called AUC, area under the receiver operating characteristic curve. I’m not joking. This is the main one that all the data scientists use, and it tells you nothing about the potential business value of operationalizing, of changing operations to actually act on those probabilities, those predictive scores that the whole thing is meant to deliver. Let’s speak the language of business. There’s these really straightforward business metrics, KPIs, key performance indicators, like profit, savings, top- and bottom-line revenue, number of customers saved. Whatever the goals of that particular project are, data scientists need to work to also make sound estimates of what those wins will be in those concrete business terms so that stakeholders can make an informed decision, so they have visibility to decide whether, how, and when to actually go to deployment.