In an era where generative AI (GenAI), LLMs, and automation are dominating headlines, it’s easy to get caught in the excitement of new tools and frameworks. The temptation is real: you see what a model like GPT can do and start thinking, “Where can I apply this?” But this tool-first mindset flips the innovation process on its head.

The most meaningful and sustainable solutions start not with technology, but with the problem. What challenge are we solving? Who are we solving it for? What’s stopping them from moving forward today?

The risks of a technology-first mindset

When we start with the tech, we often fall into the trap of looking for a problem to justify a solution. This results in flashy demos, pilot projects that never scale, or products that don’t stick with users. We end up building for ourselves, not the people we’re trying to help.

Instead, the better approach is to stay grounded in the user’s context. Not just what they say they need, but what their workflow looks like, what their goals are, and what friction points they face daily. Only then should we begin asking: What’s the smallest, simplest thing we can build to move them forward?

Use case: AI in customer support

At one company I was advising, the initial request from the leadership was to “leverage GenAI to improve customer support.” The team jumped straight into building a chatbot prototype powered by an LLM. The responses were impressive, but the project stalled. Why? Because they hadn’t understood what the support agents or customers needed.

When we went back to basics, interviewing service agents, shadowing support calls, and mapping out escalation workflows, we found that the real bottleneck wasn’t the FAQ responses. It was found that agents spent 30% of their time searching through internal documentation during live calls. What they needed wasn’t a full chatbot, but a smarter search layer across their knowledge base.

We built a lightweight semantic search tool using embeddings. It indexed the support docs and plugged into the existing agent dashboard. It was simple, fast, and reduced average handle time by 22%, delivering real value, with far less complexity.

Only later did the team revisit GenAI, layering it on to generate summaries or draft responses based on retrieved content. But it started with the right foundation: a clearly defined problem.

The 6 questions to ask

Before writing a single line of code or choosing a model, teams should ask:

1. What specific problem are we solving?
2. Who are the end users, and what does success look like for them?
3. What’s blocking them today?
4. What’s the simplest way to validate our assumptions?
5. What is the business benefit of solving the problem?
6. How will we measure impact? What is the expected ROI?

These are product questions. And that’s the mindset we need to bring to AI and data projects, not just whether something is possible, but whether it’s useful, usable, and valuable.

Crawl, walk, run: Building in layers

Another benefit of a problem-first approach is that it allows for progressive complexity. You don’t need to launch with a sophisticated AI model. Often, a well-tuned heuristic or a rule-based system is enough to get early traction. From there, as usage grows and needs evolve, you can layer in more advanced techniques with confidence that you’re solving the right thing.

This “crawl, walk, run” approach also makes it easier to course-correct. When you’re tightly aligned with user outcomes, you’ll spot when something isn’t working and have the clarity to fix it quickly.

Conclusion

Tools like GenAI are powerful. But power without purpose can be distracting. When we start with the problem, understand the people behind it, and build from first principles, we make technology matter.

Real innovation is not about what’s possible — it’s about what’s useful. And that starts with listening.

About the Author:

Sai Sharanya Nalla is a Machine Learning leader with over a decade of experience driving innovation at leading global organizations including American Express, AWS, and Nike. She has a proven track record of successfully leading cross-functional teams in the design, development, and deployment of large-scale AI and machine learning solutions.

Nalla specializes in taking products from 0->1, building from the  ground up, scaling teams, and guiding organizations through growth and transformation. She is known for identifying unaddressed gaps, and creating value by focusing on real user needs and opportunities that others often overlook.