A point I’ve raised time and again is that companies can develop tunnel vision around ML/AI. They rush headlong into hiring a team of data scientists without first developing a data strategy or checking whether their data is usable. Or, they set out to build an AI-enabled product but invest most of their time and energy on the model itself.
Lately I’ve been road-testing the phrase “360-degree ML/AI” as a way to remind people of the bigger picture.
I don’t deny that “360” (and its cousin, “365”) has become somewhat cliche in business jargon. Still, I like this phrase because it puts the business goal – “embed ML/AI into our product” – in the center, and encourages us to think about the wider space around that center point.
(Companies often assume that the data scientists and predictive models form the centerpiece. I don’t agree with that perspective, but the nice thing about the “360-degree” concept is that it lets you make pretty much anything the centerpiece and build around it. So if the phrase incidentally aligns with their existing understanding of ML/AI, and gets them to think with a bigger-picture mindset, so be it.)
To data practitioners, the idea of thinking beyond the model seems obvious. But this doesn’t always hold for people outside of those teams. How many companies get into ML/AI with a very limited understanding of what it truly is or what it does? Between vendor pitch materials and media reports of FAANG-level success stories, it’s no wonder they have an incomplete picture of what a successful ML/AI operation requires.
Thinking in terms of 360-degree ML means keeping the entire model lifecycle in mind:
- Understand the business purpose (including: determine whether a model is even a good fit what you’re trying to do)
- Perform a risk assessment around the project
- Gather and evaluate training data
- Train the model
- Deploy and monitor the model
- Use other data technologies around the model
It also means understanding when to support (or even replace) the predictive model in your plans with human oversight and static business rules (built into software).
In turn, all of this requires that we bring everyone to the table: business stakeholders, product management, software engineering, and ops … in addition to the data scientists.
When your business stakeholders are stuck on a very narrow vision of how to use advanced data analysis, I invite you to use the phrase “360-degree ML/AI.” Hopefully it gets them to see the bigger picture, making your job easier and leading to a successful outcome.
Let me know how it goes!