Disrupt or Die: The Imperative for Courageous Change in Today's Economy
You can't have the reward of upside with taking the risk and doing the work
Many individuals desire the rewards that come with taking risks but without the potential for failure. They crave the entrepreneurial success without its inherent downside. However, that's not how the world operates. The majority of AI projects in the market fail, north of 70%, and I believe that majority of AI initiatives will continue to fail for the same reasons as before, I’ll explain in this short article. A funny part of this failure is understanding the biases that exist in the market (build-it-yourself) and how to escape those for realized value.
Entrepreneurs, the sea of failure before success
We often tend to glorify the upside, narrating the stories of founders as though they are flawless journeys to success. We overlook the hiccups in the narrative, the hidden drama behind the scenes - board members resigning, payroll narrowly met, existential lawsuits, and so forth. I know a founder who nearly missed payroll three times, once when he had over 100 employees. That founder went on to sell their company for a very good outcome, they took a risk that few were willing to bear.
Having interacted with numerous founders, including those who have faced failure, I've come to realize that the risk of downfall is substantially greater than the potential upside. I've met 10 times more individuals who have failed than those who have succeeded in realizing returns beyond their basic opportunity costs.
Enterprise entrepreneurs
This rings true for organizations as well. Many entities crave innovative thinking and transformational change—yet they balk at risk, substantial change management, significant cultural shifts, and investing in external partners or consultants.
Organizations that recognize their limitations—that, despite their best efforts, they might not progress swiftly enough, or that, despite their best talent, they might still fall short—are the ones set to thrive. These are the organizations willing to drop anchor and pivot, poised to flourish in this new economy. In particular, organizations where an executive's job hinges on a successful #AI transformation stand a fair chance.
Old failure modes remain
Many of the original failure modes remain, they are simple. There are common problems in this space where data scientists tend to be disconnected from the business.
External validation through built-it-yourself cause-you’re-smart
Some data scientists shoulder the responsibility of replacing software vendors, using this endeavor to determine their own value to the organization. They proclaim, "Look! I've managed to port our existing models into a system that takes longer to build models, lacks the same level of support, and fails to keep up with recent innovation shifts. I will now spend several months building what used to take weeks or days. But hey, at least I saved us the cost of that software..."
I understand this temptation, having been that person for many years. It wasn't until I began interacting more closely with the business side that I understood: projects need to align with the CEO/CFO's vision. If they don't, you're essentially a cost center. Many data science teams have seen significant cuts during this recession because they failed to map to value. Verizon remains one of my favorite examples, having mapped all projects to a global attribution number. Every business should mandate that every project has a value estimate for AI.
You don’t know what good looks like yet
Is the model functioning? 85% AUC… alright… but prepare to embrace a world of failure. If you're measuring your models' success based on statistical data science metrics, then you've fallen short of mapping them to appropriate value. You need domain experts, process owners, and business users to assess the value of augmentation/automation. This is primarily because the metric of value will most likely be a number they were already tracking or using.
Tuition, the actual monster under the bed.
Confession time: I've had a rough experience with this one. About seven years ago, I managed a team of data scientists, each member armed with a PhD in Physics. Such academic credentials brought with them a certain level of arrogance, leading to the belief that we could construct anything being offered in the AI/ML realm within 2-3 months, provided it was useful. However, I failed to acknowledge the lurking 'tuition monster' behind me.
Let's delve into the realms of intelligence and knowledge. How can my team and I possibly understand all the different failure modes of AI? Sure, we could read about them. But many aren't documented thoroughly. Or, we could learn from our own painful experiences. Indeed, I've seen my share of failure modes in action, often negatively impacting a customer at my employer's expense. Through the lens of the business, such actions were reckless and completely avoidable.
It's crucial to leverage the expertise of market leaders—their experience, their solutions, their hard-won lessons. These leaders have navigated more failure modes across their customer base than you're likely to encounter in your entire career. So why not reap the benefits? Why pay for discounted tuition?
Despite this, many of us, with our deeply ingrained academic backgrounds, will default to creating solutions from scratch. We embark on these ventures on timelines that extend too long, armed with experience that is far too limited.
"A thousand things to do" is a surefire path to failure.
Enter prioritization. When #AI has advanced to the point where it can be embedded in every process, every facet of your business, that's fantastic, isn't it? Not quite. That's a potential disaster in the making. It's akin to attempting an ultra-marathon without any training. Prioritization is not just helpful, it's essential. Learn to crawl before you try to sprint.
Brainstorm dozens of projects with your business teams. Ask insightful questions such as, "Can you identify a number in your business where a minor shift would elicit excitement?" or "Where is your growth hindered by human capital constraints? If I offered you 1,000 people for free, where would you deploy them to maximize your business growth?"
Collect all the ideas and then rank them by feasibility. Is the data readily accessible? Do you have a domain expert available for the project? Next, rate them on value. Is the impact minimal [0], moderate [$], substantial [$$], or game-changing [$$$]? Don't over-complicate it; game-changing means it's significant enough to warrant boardroom discussions.
Finally, categorize the business leader involved in the project. Are they a green vector (an AI advocate), an orange vector (AI neutral), or a red vector (AI antagonist who may be concerned about job security or stuck in a "this is how we've always done it" mindset)?
Choose projects with green vector leaders, high feasibility, and high value. Begin with a maximum of three projects. Make sure at least one, if not all three, are successful. You don't want failure right out of the starting gate.
Good luck on your quest, I hope you collect the maps from those who have venture further than you and use guides for your trip into the future. It also helps the sooner your data science team can appreciate the actual needs of the business and the reality of the market the better.