Despite the spectacles of Westworld and Ex Machina, despite the incredible advances we see in auto-ML, and despite what ChatGPT and now GPT-4V are proving capable of, the path to business value from AI remains locked behind clear and complete problem definition and a sharp, strategic understanding of how the solution will incur minimal cost and create maximum impact. The reality is that despite the incredible pace of innovation in AI, the most impactful solutions for your bottom line are almost always the very narrow, targeted ones that address your specific use case in your specific context. Venturing outside these bounds is tempting, but you will almost always end up with a solution that has higher cost, lower performance, and/or considerably different risks.
Let me provide an anecdote that encapsulates an experience we hear all too often as AI consultants. I moved to Ann Arbor right after college and, as you might expect from a young Texan, wildly underestimated how prepared I would need to be for winter. It turned out to be an especially unlucky holiday season with wind chills reaching -45 degrees Fahrenheit. After my first major blunder of failing to buy an ice scraper before I needed one, I decided to become more proactive and purchase a big, warm pair of mittens so I could more comfortably get across the frozen but otherwise walkable city.
Everything was perfect until days later when I realized that I couldn’t use my phone’s touch screen through the mittens. I had always assumed screens worked via pressure, but it turned out that my first smartphone required electrical conduction prevented by mittens. I had already spent money on the mittens, so I bought a cheap pair of thin, conductive gloves to go under the mittens and resigned myself to frigid un-mittening any time I needed to use my phone. Sure, I also periodically dropped the loose mitten only to have its insides soak in snow or rain to freeze within minutes, but I reasoned it was still better on average. On rare occasions, my nose would get cold enough that I could use it for simple messages instead of my fingers – joyously getting one over on mother nature in my mittens while pecking out responses limited to “ok”, “omw”, or a heartfelt “ily”. Knowing what I know now, I would recommend anyone new to the region to invest in a single pair of warm and conductive gloves well in advance.
Similarly, you want an AI solution that fits like a single pair of gloves. Anything less is likely to incur extra cost and/or hamper effectiveness. Plenty of AI systems out there accurately call themselves gloves, but they might be misaligned to your needs. Gloves are needed for cleaning, bowling, and boxing but no glove does any two of those well. Even more frequently, as in my winter experience, you later discover you had specialized needs you did not account for while selecting or designing your new solution. You might need your AI predictions within milliseconds; you might need AI predictions that can be readily explained to employees or customers; or, especially with generative AI, you may need to confidently restrict the AI’s outputs to ensure coherent, safe, and accurate responses. Also, like my dropped mitten, the workarounds that allow a square peg to fit into a round hole often minimally check a box or create new problems. Making -45 degrees feel like -25 degrees by paying for a second set of gloves may not really amount to much practical benefit.
Finding the right gloves to begin with or helping rethink the combination you find yourself with today is where KUNGFU.AI shines. An effective AI solution requires three inputs:
Most articles about AI seek to differentiate AI development from software development, highlighting that while software requires great code, AI also relies on having great data. However, with more than one hundred successful client projects we are confident that the third requirement, the subject matter expertise, is even more critical and most frequently overlooked by less experienced teams. It’s deceptively easy to verify requirements (1) and (2) in a laboratory setting, only to find out that the real-world details lurking in (3) prevent the impact you planned on.
In fact, if you haven’t carefully thought through (3) then you will almost always overestimate the degree to which (1) and (2) are true. It’s too easy for context to get lost between these inputs and for something critical to get overlooked until it reappears at the eleventh hour. People are more than numbers in a spreadsheet and, ultimately, all tech problems are human problems (or, put differently, All Tech is Human). A theoretically sound model that is 20% more accurate than its predecessor will not make an impact if it disempowers its users. People are exceptionally good at finding ways to work around AI if motivated, like how I learned within a day to watch YouTube on incognito if I didn’t want my recommendations to be impacted. Once that happens, it doesn’t matter how good your original dataset and codebase were.
At KUNGFU.AI, this principle guides us to fundamentally transform your business with AI. We do not pre-select algorithms, technologies, or even technology platforms before understanding the use case in its entirety. We start from (3) in order to genuinely understand your precise needs, exactly how a solution must address them, and whether AI is even a necessary component of that solution. Consequently, we work in smaller teams than you might expect and value close collaboration with our clients. Armed with that full understanding, we can quickly and reliably identify what it will take to move the needle. We aren’t in the business of selling gloves, we’re in the business of keeping your hands warm, dry, and conductive.
.png)
.png)
