March 2, 2021
I have a confession. I spent several years helping Fortune 1000 companies develop AI strategies before I recognized some gaps in my approach. I used to believe that an AI strategy should primarily answer the question, “What can AI do for us?” I even wrote a book about it.
Of course, answering this question is a critical first step. But I’ve learned that it isn’t enough. The hardest part of becoming an AI-driven company is making organizational changes. An AI strategy needs to address this challenge.
CEOs and boards always ask about required organizational changes. In a few instances, I’ve been caught without answers to their questions. In this issue of Feedforward, I’ll share some key concepts that pushed my AI strategies to the next level. My five aha moments can help you make your AI strategy more persuasive to your leadership and more successful in the long run.
Let me first explain why new program oversight processes are necessary. Most companies start traditional technology investments by choosing a single line-of-business (LOB) sponsor. This approach doesn’t scale with AI because many LOBs need similar capabilities, such as reading and classifying documents. So most AI strategies advocate for developing centralized capabilities for multiple LOBs. But this structure creates new oversight and funding questions, such as:
An AI strategy must describe how program oversight will work. Because no one department “owns” AI, I usually recommend creating a centralized committee for governance. The committee should include senior representatives from every department. Specific initiative oversight can happen at the subcommittee level.
CEOs almost always ask questions about people. They correctly recognize that the organization doesn’t have skills to build and leverage this new technology. They worry about internal blowback or bad press about job losses.
CEOs first want to know if the existing workforce can acquire the necessary skills through training. Although a “nobody gets left behind” approach is admirable, it’s also wholly unrealistic. Almost every company that wants to incorporate AI needs to evolve to a smaller, more skilled workforce. The gap between the skills of your existing workforce and your future needs is too big for training alone.
Don’t expect your leadership to confront this uncomfortable reality right now. They’re not ready to hear it. But don’t entirely avoid the topic of workforce transformation. People are going to ask your C-suite about it, and you need to arm them with an answer.
Budget for specific skills training in your AI strategy without committing to training for everyone. For example, you can provide training for basic Python or R analytics.
Like most AI leaders, I tend to invest most of my energy in execution. Communications used to be an afterthought for me, but now it’s a top priority.
As I’ve written previously, getting organizational buy-in for your AI strategy is much harder than getting the CEO and board to approve it. You need a plan for internal and external communications to drive enthusiasm.
The usual tactics of public relations, blogs, and speaking at conferences are all fine for external communications. You probably just need to budget for them.
Internal communications are harder. But they also present a great opportunity for you professionally. At a minimum I suggest you plan to give all-hands presentations and “ask-me-anything” sessions with the company’s AI leaders and C-suite. If you have the budget and time, I suggest also creating a comic about your company’s AI future. Take a look at this example of an AI comic we created.
Comics are a great medium for evangelizing AI because they make the future more conceivable through stories and images.
Everyone has read about AI risks in the media. Although concerns such as bias, model drift, and interpretability are usually valid, they’re not immediate concerns for the AI strategy. (Often the people who are the least informed about AI have the strongest opinions about these issues.)
Regardless, your C-suite will ask about model risks. Your AI strategy needs to provide answers.
Don’t attempt to write out your model governance policy in your strategy. It takes too long. Instead, make model governance one of your strategy’s action items. Set the stage by explaining the risks you plan to mitigate and address. Describe your approach to developing the governance policy.
You can use the three P’s of model governance as a framework to organize your thinking:
You’ve got data silos. Your infrastructure is brittle. You have outdated tools and platforms.
You need to start fixing these infrastructure problems before you can scale your AI program. But your business customers don’t care about infrastructure—they want solutions, and they want them ASAP. How do you resolve this dilemma?
Don’t try to delay delivering business solutions until infrastructure problems are “fixed.” This approach creates unrealistic expectations; you’ll always have infrastructure challenges. Instead, create a roadmap for delivering business value while you harden your infrastructure.
For example, don’t tie your AI projects to the completion of a data lake. Instead, identify the most critical data inputs for your initial models. Invest in the data pipelines that make those models reliably available. This approach allows you to deliver near-term business value while still hardening your infrastructure to support increasing scale.
I hope you learn from my aha moments and address these topics in your AI strategy. You don’t need answers to all possible organizational questions related to your company’s AI future. Simply acknowledging the need for changes to areas like risk, oversight, and training will help galvanize organizational buy-in.
You play a unique role as an AI leader. Most of the people at your company—and perhaps your entire C-suite—are hesitant and confused about AI. By addressing the necessary organizational changes head-on, you can overcome their biggest objections. Best of all, you showcase your unique role when you address these concerns. You’re not just another AI evangelist. You’re a leader who will guide the company into an exciting future.
March 31, 2021
In this video Justin Pounders, Director of Machine Learning and AI Research at Prolego, breaks down natural language generation (NLG) into its most basic components and describes how you can begin building out these components in your business. (And, no, it doesn’t depend on GPT-3!) He describes how NLG depends critically on two questions (WHAT you want to say and HOW you say it), the types of data you can feed into NLG systems, and a development path for being able to summarize multiple sources of data in plain English.
March 30, 2021
Like most engineers, I hate tedious work. That’s why I love the idea of automatic machine learning (AutoML). As much as I want to love AutoML, it’s been incorrectly framed as a substitute for data scientists. This confusion arises from a misunderstanding of what actually happens in machine learning projects.
March 24, 2021
Document analysis and understanding is an active area of research in the applied NLP community. In this talk, we demonstrate an unsupervised method to organize a body of text into a set of topics and outliers. This approach uses a transformer model that has been fine-tuned for semantic similarity (SentenceTransformers hyperlink: sbert.net). It can be used to quickly review a large set of documents to identify areas of interest or concern without requiring a human to exhaustively read through each document one-by-one. We demonstrate this approach applied to the lyrics of an early-2000s hit musical piece.