Are Business Leaders Jaded With Their AI Transformation Investments?8 min read
Table of Contents
In the Allegory of the Cave, philosopher Plato describes the lives of three men who were chained inside a cave since their childhood, with shadows being projected onto a wall in front of them by the world outside. The shadows they see on the wall are not an accurate representation of the real world, but it is their version of reality, for that is the only thing they have known and experienced.
The latter part of the allegory speaks about how one of the men gets unchained and ventures out of the cave to see that the world is vast and that they’ve been tricked and conditioned into believing that what they saw was real. The man returns to tell his companions about his discovery – but alas, the companions refuse to believe him and refuse to acknowledge anything that isn’t the reality that they know.
Plato likened the prisoner who was freed to a Philosopher. We’re taking a step ahead and likening the prisoners to those who see the benefits of AI for Enterprises against those who don’t.
AI, despite having been an active part of enterprise solutions for the latter half of the last decade, still continues to face resistance. Business leaders are still jaded in investing in AI Transformation. Some of them have had their hands burnt by narrow AI or point solutions in the past. Some of them find it hard to understand the ROI due to the complexities involved in implementing AI at scale. No matter what the reason, the bottom line is that akin to the prisoners inside the cave, leaders, today refuse to acknowledge the realities that exist outside of the world they are used to.
But AI itself has changed so much over the last decade. AI has gone from narrow, point solutions that solve a specific problem to an org-wide stack that delivers impact to every function, team and individual in the organization. While a lot of businesses have tried implementing a single AI solution to solve all their issues, they have burned their hands with long implementation cycles that take years of time and ultimately show negligible or no ROI!
When and How does AI fail?
1. Data drifts with time
Implementing AI can be challenging due to data drift. This is the ever-evolving of data in real-time. AI systems tend to overfit the specific data types they were initially trained on, leading to a degradation of the model’s efficiency over time. This data drift significantly affects the overall performance and accuracy of the models over time as the models don’t drift with the data.
Not just this, new data gets continuously fed into the system, creating a never-ending loop of iterating it. Leading to a long deployment process requiring an army of ML engineers and data scientists to maintain and update the models constantly. The deployment process takes a long time, and when it is finally completed, the old models may not work with the real-time data constantly being fed into the system.
2. Connected workflows vs. isolated systems
Business leaders should understand the role of continuous workflows compared to isolated point solutions because they can impact the efficiency and effectiveness of their operations.
Continuous workflows refer to the integration of multiple tasks, allowing data and information to flow seamlessly between them. This approach can lead to increased automation, reduced errors, and improved decision-making as it allows for more accurate and timely data analysis.
On the other hand, isolated point solutions refer to individual tools or processes that are not integrated with other systems. These solutions can lead to siloed data and information, resulting in inefficiencies and a lack of organizational visibility. This lack of context can make it difficult for businesses to realize the ROI they expected from their AI investments.
Additionally, if the model is not properly optimized based on the specific context in which it will be used, it may not be able to make accurate predictions, as multiple vendors are working on the same data. Working with multiple vendors also means multiple people just to manage them.
The “who,” “where,” “when,” and “why” provides the information that shapes human decisions and actions. Without context, AI models are also equally likely to fall short and not meet expectations.
3. Sunk Cost of implementing AI
Businesses spend millions of dollars implementing AI over the course of 2 to 3 years. This is done with large teams of ML engineers and data scientists constantly iterating and updating the models to change with real-time data. Thus there is a large sunk cost over the course of years as the business invests in people, money, and time to make AI work for them. Justifying this investment becomes extremely hard as the money invested is just too much for any tangible ROI to be achieved.
4. Quality & ROI
It is no secret that most enterprises have seen multiple AI investments fail over the last decade. To put it bluntly, the quality and outcomes have consistently fallen short of the initial tall promises. Enterprises that intended to use AI as a magic wand have invariably ended up getting their hands burned by the same wand. AI is not something magical or mysterious. AI delivers only when implemented right, fed the right data, and when the right expectations are set. AI can deliver an order of magnitude ROI over the years as the model learns and becomes more refined. When a realistic action plan is set for an AI Transformation journey, business leaders can depend on AI. They just have to remember AI is not J.A.R.V.I.S. to their Tony Stark, it’s more like Robin to their Batman or Chewbacca to their Han Solo.
A Real-world example of AI delivering exponential ROI
At Vue.ai, we observe a common pattern across most companies that resist AI Transformation. Making progress on automation, efficiency, and improving your ROI comes with the cost of neophobia. Companies are happy to incorporate changes into their existing systems but not pick a better alternative for the systems themselves. The general mindset has largely been, “Make our lives easier, not better”, without realizing that both of these things can happen at the same time.
A European retail company had deployed three different AI solutions – one for search, one for personalization, and another for product tagging. While all three solutions functioned well independently and fulfilled the tasks that they were meant to do, they wouldn’t integrate with each other as seamlessly. The interaction between the solutions caused problems.
An example scenario: Imagine a customer that’s looking for a dress in an orange shade. One system tags the dress as ‘Orange’, another ‘Tangerine’ and the third, ‘Carrot orange’. When said customer searches for a ‘Tangerine Sunset’ dress, orange or carrot orange dresses don’t show up due to the mismatch in tagging. Identifying and fixing these problems proved to be costly for the retailer – as it required significant time, effort and money. Implementation became a recurring problem, which meant a lack of ROI.
The European retailer implemented Vue.ai, a unified AI platform that brought together these solutions under one roof, which made implementation a one-time event and not a recurring effort/cost. Vue.ai delivered solutions for multiple use cases – all through a single platform, dealing with the data variance and complexities involved. With the data being contextualized to each solution and all brought together in a swift, automated workflow, the retailer was able to reap the benefits of significantly lesser maintenance costs and a high ROI.
Enterprises have been using AI for close to a decade but a lot of business leaders are yet to see significant ROI. Implementing narrow, point solutions to solve individual use cases leads to data being siloed. These solutions can lead to siloed data and information, resulting in inefficiencies and a lack of visibility across the organization. This lack of context can make it difficult for businesses to realize the ROI they expected from their AI investments.
On the other hand, implementing AI across the organization was also challenging with the process taking 2 to 3 years. Additionally, the ever-changing real-world data required large teams of ML engineers and data scientists to constantly iterate and improve the algorithm to improve its efficiency.
Today enterprises use context-aware, end-to-end systems like Vue.ai that automate entire workflows to bring orders of magnitude savings and efficiency. These continuously learning systems compound in their efficiency with time. These systems build new models and choose the right model from a federated set of models without needing large teams of ML engineers and data scientists. The data-centric architecture of Vue.ai allows the models to drift with data thus improving their efficiency with more real-time data flowing in.
Vue.ai can be applied to a wide range of applications across industries like retail, finance, insurance, staffing, healthcare, and more.