Sandeep Saha

Not every but significant AI solutions fail. In many cases it’s the rush to prove technologically updated with the industry instead of focusing on the business requirement or need. I would like to emphasize on the banking sector, where I have experienced notable issues such as targeting the problem statement, low quality data input and a resistance in adoption to change. Due to conservative nature of work many AI initiatives hit a roadblock as they prompt regulatory and customer/ stakeholder issues which result in impacting the yield coming out of the AI solution to the business problem statement Most cases there’s a heavy push from the top management without a convincing use case which cause issues for the project team. The project team gets into a loop of approvals, data cleaning, cross selling, user access, etc. Following are few reasons why AI solutions fail wrt banking sector 1. Data quality, access, and conservative approach When a retrospective session is held it traces back to unadministered, poor quality data based upon which the models are developed which result in instability and biasedness. 2. Lack of E2E workflows integration In many cases a successful proof of concept is the extent to which an AI solution project reaches smoothly. However, the risk & control, routes and process engineering around the decision is neglected. The result is a low confident prototype that is overshadowed by manual interventions. 3. Inefficient change management Business stakeholders, analyst, relationship manager are not upbeat with the technology. There is also a resistance due to fear of job loss. Because the concept is not widely shared and lack of training the models remain underutilized. A local champion, bonus/incentive and the confidence to even fail is missing . 4. Regulatory, Compliance and Risk There are unexplainable documentation and no monitoring, in such case audits fail and cause risk. In the banking sector delayed decision making , approvals, frequent roll backs are critical for banking department such as AML, frauds, credit & collection , etc 5. Undervalued operational cost Banks need to be cautious over frequent trainings, new features, data cleansing. There needs to be a standard and discipline that needs to be followed over the time as training get stale, models deteriorate, cost increase resulting in lesser business benefit, so the projects die a slow death Now the question arises, how the banks can avoid these issues. Following are some of the pointers: 1. Attention towards sustainable value generation Develop every use case with specific problem statement , quantifiable metrices and impacts. Operational cost on governance, discipline and training need to be compared with the cost of manual intervention , monitoring, fraud loss and other inefficient metrices. 2. Early investment in governance, MPOps and Data cleansing To keep the model consistent in production, data foundation and high risk use cases need to be robustly build. Frequent monitoring and drift detection keep the model up to date. 3. Design efficient Change strategy Collaborate with the stakeholders, operations, relationship managers and other staff to explain the usage of tools in daily operations. Use AI not to replace humans but to engage AI for low impact decisions , exception highlighting, prioritization . The above, translate directly into value‑stream‑level design, measurable benefit tracking, and strong change management around AI‑enabled processes in banking environments.

AI can be creative in a real, operational sense, but its creativity is fundamentally different from human creativity because it depends on pattern-learning from human-made data and lacks lived experience, emotion, and intrinsic intent. Modern generative models learn statistical patterns from massive datasets and then recombine and transform those patterns to produce new texts, images, music, or designs. Even though the mechanism is recombination, the specific outputs can be things that have never existed before—down to word sequences or visual arrangements that are not found anywhere in the training set. Human creativity is tied to consciousness, emotion, culture, and long-term goals, which shape why and what people choose to create. AI creativity, by contrast, is instrumental and dependent: it emerges from algorithms operating on prior human outputs, and from human choices about prompts, training data, and evaluation. Under a human-centered, experiential definition—creativity as an expression of a conscious, feeling agent—current AI is not truly creative and mainly remixes human ideas. Under a functional definition—creativity as generating novel and valuable artifacts—AI systems clearly can be creative, especially when used as partners that augment and extend human imagination rather than replace it.