Generative AI raises uncomfortable questions that often no one wants to answer. One is whether the AI bet is worth taking, or is it still la-la land? In this post, I want to touch upon this topic and unpack some perspectives from people in the field. Most of it is subjective to our experiences at Tursio, so it should be taken with a grain of salt.

Newfound World

More of our world problems are unsolved today than we would like to believe. We are already past the first quarter of the twenty-first century, and yet most organizations we come across continue to grapple with staggering inefficiencies in their daily operations. From producing goods and delivering services to running sales, marketing, finance, operations, and customer service, these organizations rely on a considerable amount of human effort. Sadly, many of the tasks involved are insanely repetitive, and yet they consume valuable human resources, something that is getting more precious every passing day.

Scaling with better efficiencies is an age-old problem. During the medieval age, larger empires were considered stronger and more efficient. More landmass meant more resources that could be put to use. This changed to putting more people into the labor force during the industrial age and then to getting more machines on demand in the modern cloud computing age. The goal has always been to parallelize the work with more workers. Modern-day workers are knowledge workers who need to think and ask questions to get things done. Access to information, therefore, becomes the new resource, and Generative AI is showing promising signs of putting information in the hands of these knowledge workers.

The Dark Clouds

Interestingly, the promising new world of generative AI is already muddled in dark clouds. The market is rapidly getting flooded with new AI products and services every day. Especially, all major incumbents have infused “AI” throughout their products, sales, and marketing. The noise is overwhelming not just for the investors, who I see having little to no chance of figuring out how one AI differs from the other in a market that is more willing to go with the incumbents, but also the customers who I see grappling to understand where to extract the value and how to move the needle. And lastly, the builders who I see having a hard time figuring out the killer app that works perfectly and is a no-brainer for ROI.

A big part of the above din stems from the fact that generative AI is still in its nascent stages. Generative AI models are getting perfected every single day with new capabilities to make them more usable and reliable. Current models hardly work out-of-the-box, with a lot of work required from developers to craft them into applications that work.

Horse Looking for a Carriage

AI is seen as the exciting new magic, and so the first thing a customer asks, I hear, is to “Show me something I don’t know?”. In fact, I routinely find the use of traditional statistics and machine learning not impressive anymore. They may even feel let down if the output is something they can understand easily. Apart from being a by-product of the hype, the expectation is to see something magical. As such, I find it very easy to fall into the trap of having an AI horse that is looking for a carriage to drive home. To navigate this better, I see reframing the goals into one of the following to be helpful:

• Simplify (something I want to do better) — Identifying tasks that could be simplified and made more efficient with AI is a wonderful framing. This is my favorite because it focuses on making people more productive, and so naturally, the discussion revolves around the pain points they are seeing. It is also more achievable since it is targeted at a specific user.
• Automate (Something I don’t want to do) — Many people look for specific business processes to be automated. It is more ambitious but still concrete, i.e., clear outcomes and expected ROI, typically in terms of cost, that move the discussion more towards feasibility. However, there is still a significant learning curve to understand the current process.
• Breakthrough (Something I don’t know I should be doing) — Searching for things that people should be doing is very open. You are chasing magical findings, and it requires a much bigger appetite for investment and risk. It takes dedicated focus, deep expertise, and a lot of luck to discover that one big thing — in short, it happens rarely!

It's Too Complex

Unfortunately, many AI solutions are way too complex today. I often see customers struggling to find their way and making the solution work for their scenario. This has several implications:

1. Despite the AI hype, people often don’t have the basic things working in the first place. Increasingly, I see customers seeing value in simpler end-to-end products, rather than drowning them in features and bells and whistles. This is especially true for AI, where a simple search box is the standardized interface.
2. A complex solution is hard to trust. Organizations have dedicated all their time and energy to building what they build, but how does the AI do better than what they do every single day? This is particularly a challenge when trying to automate processes or search for breakthroughs. The domain knowledge is paramount in most businesses, and it's hard to mask it under complexity.
3. Training AI models requires humongous amounts of data, and therefore, the question when applying AI to enterprises is how many data points do we have anyways? In reality, the data points are way smaller and the space is too sparse. Building complex solutions and workflows oversimplifies the business reality and likely make things worse.
4. Human workers spent countless hours and training sessions to learn how to operate, and they still use their judgement to handle tricky scenarios. Machines need to learn far more in order to cover those corner cases and have to incorporate feedback, i.e., constantly growing training data. Without this infrastructure, complex solutions find it tough to hold their ground.
5. Often I hear customers asking “how does it work?” or “what is the AI behind the scenes?” or “what did it learn?”. In reality, they are having trust issues with running their business using this new magic wand. The goal should be to revisit the assumptions and make the solution as simple as possible. More complexity needs to earn the trust first.

Keep It Simple

Making AI work is the single biggest challenge out there today. Customers love products that work every single time, no matter how simple they are, as opposed to something that works only 50–60% of the time. So while it is important to recalibrate customer expectations, it is equally important to make the AI product work. And here, despite all the competition, every product is essentially competing with itself on things like hallucination, accuracy, security, and onboarding speed when it comes to AI. This is precisely our focus at Tursio — to keep it simple and to make it work!