If I were to select one image to describe the relationships of business teams with AI, I would choose the opening scene from 2001 Space Odyssey. Do you remember that scene, and the use of the black monolith, often interpreted as the allegory of change? It seems so well adjusted to the depiction of how business teams relate to AI predictions: as a big black box. If you think about it, AI is here to help them move faster and make data-driven decisions, and yet, a lot of the time, they find themselves unable to understand and trust the predictions, as they may clash with their own subject matter expertise. Besides, and while they are often consulted during the research phase when the requirements are being gathered, once the latter is set, it appears that the models move “to the dark side of the moon”. The business teams only meet them at the other end of the process once predictions are issued.
I am especially thinking about cases where AI is used for core business use cases: fraud detection, customer management, or loan approval.
The teams on the ground, who live the business, often find themselves, looking for more answers. Yet, these answers cannot be found in python notebooks and cannot be expressed in terms of “accuracy”, “log loss” or “ROC”. What they need is more transparency, more independence, and a better understanding of how the predictions come to be.
These silos between expectations and the use of AI may be the culprit of some failed AI programs or the long delays often incurred in the implementation of robust AI solutions. So how do you go about aligning business and ML to make sure your AI program is not at risk?
Independence is the key
If every time your business users need to turn to the data science team to explain changes in the predictions or to get a sense of the behavior of a sub-segment, then you are not hitting your targets. The last thing any of these stakeholders want is to be pulled into meetings and see their teams slow down – as data science teams don’t want to be focusing on troubleshooting, and the business wants to be able to move as fast as possible.
To achieve that, it is important to leverage a view of the models that enables them to access the insights that matter to them: Are there changes in the input? Are there issues that they need to be aware of before making decisions based on the model’s output?
This is also where visibility becomes a crucial part.
Not everything should be in a Python notebook, and business users require a view of the models that speak human. Whether it is by leveraging APIs to be able to connect insights from production to business applications or by using dashboards with views that are easily digestible.
Yet, views are only as valuable as the content they display. As VentureBeat notes, data – even a lot of it – isn’t enough to “provide a company with actionable insights and positive business outcomes”. Business users demand AI-driven insights that lead to the best results, not beautiful graphs or more information. According to Gartner, through 2022, only 20% of analytic insights will deliver business outcomes.
Business users want AI that works for them providing actionable insights and adding value without generating extra work. Through a production-driven approach, you can achieve a robust view of your data and predictions behavior that feeds insights to your business teams – amongst others: how are subpopulations behaving? Are there drifts that reveal opportunities or threats?
In order to have a real-time view of your business, you cannot govern your AI without automation. Whether it is to detect drifts, sudden changes, or changes in seasonality, or else, you need to leverage solutions that give you the right insights at the right time. Business needs to move fast and cannot afford to discover issues when it’s already too late. And with so much today happening in real-time, this is the speed at which business users need their AI to run.
Maximizing your AI investments
This is where Superwise comes into the picture: to help you design a production-first approach by leveraging robust model monitoring that creates a single truth across the organization. Our AI assurance solution bridges the gap between the requirements of data scientists for more efficiency and control over their models in production and those of the business teams for more independence and visibility. At the right time. With one single platform touching upon all aspects of model monitoring: analyze, measure, troubleshoot, and alert, we already support business teams globally to gain value from their AI. To learn more about how we support the different stakeholders, check out the highlights from our latest webinar with monday.com, where their marketing and data science teams share how they leveraged Superwise for better results.