From Solution to Problem: Rethinking Machine Learning in Banking
Imagine trying to navigate a bustling city using only landmarks instead of a map. You might end up taking longer routes, retracing your steps, or even getting lost.
This approach mirrors the way some banks approach Machine Learning (ML) — starting with the solution and working backward to find a problem it can fix. While technologically impressive, this approach often leads to misalignment, wasted resources, and frustrated teams.
Now, flip the scenario. What if you start with a clear destination — your operational challenge — and chart your path backward to find the best route?
This is the essence of backward planning: beginning with the problem and designing an ML solution tailored to address it.
Banking, steeped in tradition and governed by strict workflows, stands to gain immensely from a shift to this problem-first approach. By rethinking how ML is deployed, banks can avoid the pitfall of using technology for technology’s sake and instead focus on creating meaningful, impactful solutions.
The Contrast: Solution-to-Problem vs. Problem-to-Solution
Let’s consider two distinct scenarios to illustrate this contrast.
Scenario 1: Solution-to-Problem
A bank decides to implement a state-of-the-art fraud detection algorithm that uses advanced machine learning techniques. It’s deployed with high expectations but quickly meets resistance from frontline teams. Why? The model generates too many false positives, creating unnecessary manual reviews and disrupting workflows. Instead of solving a problem, it creates new ones.
This is akin to installing a powerful new GPS without understanding the terrain. The GPS may suggest the fastest route, but if the roads are flooded or under construction, it fails to serve its purpose.
Scenario 2: Problem-to-Solution
Now imagine the same bank starts by identifying its key pain point: the rising cost of manual fraud reviews due to inaccurate detections. Teams analyze the workflow and uncover specific patterns in false positives. With this clarity, they design an ML model optimized to reduce unnecessary flags while maintaining accuracy. The model integrates seamlessly into operations, saving time and money.
This approach is like starting with a clear map of the destination and its challenges. With the end goal defined, the route can be planned efficiently, accounting for detours and obstacles.
Why Problem-to-Solution Works
Clarity in Purpose — “Technology is best when it brings people together,” said Matt Mullenweg, the creator of WordPress. Starting with the problem ensures ML aligns with business objectives and operational workflows. It forces banks to focus on the purpose of the technology rather than its novelty.
Alignment with Operations- Backward planning naturally integrates ML into existing systems. It prioritizes operational goals — like reducing customer churn or improving loan processing times — over abstract technological benchmarks.
Better Outcomes- By identifying the problem first, the solution is crafted with precision, avoiding the common pitfalls of over-engineering or misalignment. As a result, the ROI of ML initiatives improves significantly.
Making the Shift to Problem-First Thinking
So, how can banks move away from the solution-first mindset? Here’s a streamlined way to think about it:
1. Define the Destination- Banks must identify their most pressing challenges. Is it reducing compliance costs? Improving credit risk assessment? Once the challenge is clear, the path to solving it becomes more apparent.
2. Understand the Terrain- Break down the operational workflows involved in the problem. Engage with teams on the ground to uncover bottlenecks, inefficiencies, or gaps in decision-making processes.
3. Choose the Right Route- Design ML solutions that fit seamlessly into existing operations. Iterate with input from cross-functional teams to ensure the solution is both effective and practical.
This iterative, problem-first approach is what Donella Meadows, in her work on systems thinking, called “dancing with the system.” By observing and aligning with the system’s natural flow, solutions emerge organically rather than being forced.
Rewriting the Narrative of ML in Banking
Banks have long struggled with the dichotomy of innovation versus tradition. ML doesn’t need to add to this tension. By focusing on operational problems first, banks can ensure that ML becomes an enabler of efficiency and growth, not an abstract experiment.
Consider this: Is your organization approaching ML as a hammer looking for nails? Or are you designing the nails and crafting a hammer that fits perfectly? This shift in mindset could be the key to unlocking the full potential of ML.
Simon Sinek famously said, “People don’t buy what you do; they buy why you do it.” The same applies to ML. Its power lies not in its algorithms but in its ability to solve meaningful problems.
By rethinking how we approach ML — starting with the problem and working backward to the solution — banks can harness its true potential. It’s time to embrace this shift and let operations, not technology, drive the narrative.
Here are some reflection questions to ponder upon:
- Are your ML initiatives solving a real problem, or are they solutions in search of a problem?
- How can you better involve operational teams in the design of ML solutions?
- What steps can your organization take to move toward a problem-first mindset?
This shift isn’t just a change in process — it’s a change in philosophy, one that places purpose at the heart of innovation.
