Original article was published by Kelley Mak on Artificial Intelligence on Medium
The State of AI: 4 Realities of F500 Adoption & Opportunity
We recently launched our “State of Enterprise Tech” Webinar Series, which kicked off with our first webinar on “The State of AI.”
Our panel of corporate and startup leaders, including Adam Wenchel, CEO and Co-Founder of Arthur, Olga Lagunova, Chief Data and Analytics Officer, VP Commerce Cloud Technologies at Pitney Bowes, and Patrick Wagstrom, Director of Emerging Technologies at Verizon, discussed what Fortune 500 adoption of AI/ML tools and technologies looks like today, and where it is heading in the future.
In case you missed it, you can watch the full webinar on YouTube here and read my top takeaways below:
AI leaders need a technical AND business line
While it’s important to either be a domain expert yourself or have domain experts in your corner to allow you to properly evaluate and translate AI technology, it’s equally important to have business experts weigh in. If you only have technology experts on your team, you could easily miss the larger picture of how AI will actually impact the business.
Pitney Bowes sits at this intersection, as Olga explained, “our business is in the intersection of technology, e-commerce, logistics (domestic and international), and engineering of IoT devices related to shipping and mailing. My role is to help to build scalable cloud-based platforms and products and experiences that are driven by data and AI.”
Similarly, according to Partick, at Verizon, “…for large scale evaluations we’re very careful that we have representatives from other parts of the business as part of those evaluations. We check-in with them routinely and make sure they have all the updates.” Setting up this cadence allows the tech experts to know how and where the technology will provide maximum ROI and how it will impact broader company budgets. For startups, this means always be mindful that you’re selling to technical AND business leaders, and adjust your pitch / demo to demonstrate value to both.
ROI is a long-game
Adam shared how he’s seeing many enterprises start to invest in AI’s further potential and “…realize that [AI] is a long game. In many cases, they’ve spent the last few years beginning to establish larger data science teams, starting to think how to recruit and attract talent, and putting in place infrastructure to make the data science you do have more efficient…”
To help organizations see the value and intangible benefits of an AI program, it’s important to provide them with ongoing education about the impact of increasing their levels of investment.
Patrick added that, “Short term it’s about reducing cost, long term it’s about transformative to increase revenue.” So while the real wins will take time, it’s important for startups to have some quick wins to demonstrate the tech’s importance to invest in and show opportunity for ROI.
Talent is as important as technology
When investing in AI, you can’t forget about the people. When it comes to retaining talent, Patrick said that it comes down to “culture, scale, and opportunity for impact.”
- Culture: Not only do you want a positive organizational culture, but you need an individual culture among data scientists that fosters collaboration, the ability for open exchange of ideas and information, and the ability to learn on the job.
- Scale: Providing data at scale is important because it defines how challenging it will be to unlock the value of that data (this should be the fun part for any data scientist!). Generally, enterprises have an advantage here since they operate on large, often global scales.
- Opportunity for Impact: Creating opportunity for individuals with innovative ideas to try new things, explore new methods, and potentially fail will also help you stand out in the job pool.
Olga added that in her experience building a data science team, it’s important to begin with data science fundamentals given, “there is a hierarchy of needs…and while almost every enterprise has data, there is a minimum, sufficient data availability and quality needed.”
AI needs guardrails and governance
Addressing biases in AI models is of utmost importance. The Black Lives Matter movement has shone a light on the unjust treatment of the Black community and how more action is needed to remove historical biases and systemic racism that perpetuate in AI systems. As these systems require greater transparency and control around how decisions are made, model monitoring solutions will play a crucial role in helping ensure enterprise’s ethical use of AI.
Arthur is already tackling this challenge for their customers in industries like financial services and healthcare with their centralized platform for AI model monitoring, which tracks unwanted bias and allows you to benchmark these biases and track improvements as newer AI models are deployed.
In addition to day-to-day processes, concern for biases should be another set of questions you ask yourself: Why did we choose these features? Could they be proxies or misrepresent something else?
Many organizations are now rethinking their AI operations. For example, many organizations used to avoid collecting data they thought could breed biases in their data models (“fairness through unawareness”), but Adam believes, while “this is better than doing nothing, the industry is headed is to actually do the opposite… [Enterprises should] collect that stuff (you don’t need to feed it into your model necessarily), but make sure you’re actively measuring how different groups are being impacted by your model and take action on that.”