The Hardest Part of AI Is Knowing What’s Worth Building

Key Takeaways

• Cheap experimentation raises the risk of solving inconsequential problems. AI makes building fast, so the harder question is whether the problem is painful enough to deserve a solution.
• Start with friction instead of a catalogue of use cases. Fixing pains that employees already feel drives adoption.
• Prioritize opportunities and test the strongest ones quickly to find a working solution or a reason to stop investing.

For most of my career in technology, one of the biggest constraints on innovation was the ability to build. Organizations had plenty of ideas, but turning those ideas into working technology required significant investment, specialized skills, and often months or years of implementation. Artificial intelligence is rapidly changing that equation. Today, we can prototype an idea in days, create an intelligent agent in weeks, and automate processes that once required significant custom development.

That raises a question companies haven’t always had a quick answer to: Is the idea actually worth building?

I learned a version of this lesson much earlier in my career. I was once involved in an initiative where we became convinced that a sales organization had a significant productivity problem around creating customer proposals. Information came from multiple sources, salespeople reused old presentations and documents, and they spent considerable time assembling proposals manually. It seemed like an obvious opportunity for technology to improve efficiency.

The problem became apparent when I spent more time with the salespeople themselves. Proposal creation was frustrating, but it wasn’t really holding them back. Their bigger challenge happened earlier in the sales process. Customer information was fragmented, account intelligence lived in different systems, pipeline information wasn’t always current, and salespeople spent considerable time figuring out which opportunities actually deserved their attention.

We were preparing to make proposal creation significantly faster while a much more consequential problem was sitting a few steps upstream.

We had found a legitimate problem. We simply hadn’t found a problem that mattered enough.

That experience has stayed with me, and it feels especially relevant now. Generative and agentic AI have dramatically reduced the cost and time required to experiment. But inexpensive experimentation can also make it easier to solve problems that don’t materially affect the business. An impressive agent is demonstrated and a pilot is launched, everyone agrees the technology works, but several months later, few people are actually using it.

Often this gets labeled an “adoption problem” when the tool doesn’t solve a painful enough problem for employees to change how they work.

Follow the friction

At PivotX, this shapes our approach to AI transformation. Instead of beginning with a catalogue of potential AI use cases, we start by looking at how the business operates and where friction exists.

Thankfully, it’s usually not difficult to find. It’s everywhere, including repetitive work that consumes thousands of hours, manual assembly of information that takes days, lost revenue caused by broken processes, customers waiting while employees search multiple systems for answers, and operational bottlenecks that force organizations to add people when transaction volumes increase.

These problems may not sound as exciting as autonomous agents or the latest foundation model, but they are often exactly where AI can become most valuable. Somebody is already paying for that friction. The company is paying through higher costs or lost revenue, while an employee is taxed by wasted time and frustration, and customers pay by enduring a poor experience.

Choosing the right problem makes adoption easier, too. We spend considerable time talking about training, change management, and encouraging employees to use AI. Those efforts matter, but employees have a reason to use a tool when it fixes something they already find frustrating.

If an employee spends two hours every morning reconciling information from several systems and AI reduces that activity to fifteen minutes, adoption becomes far easier. If a customer service representative can immediately access the information needed to resolve an issue, or an operations team can handle greater volume without increasing headcount, people experience the value as part of their everyday work.

The organization also learns from the work. It sees how employees respond, where the data foundation improves, and where they should invest next.

Test the strategy while you develop it

This way of thinking has also made me question the traditional model of large technology strategy programs. Historically, it was common to spend several months assessing the current state, defining the future state and developing a multi-year roadmap before significant implementation began.

I don’t believe that model works particularly well for AI.

This doesn’t mean strategy is less important. Organizations absolutely need clarity around business priorities, data, architecture, governance, operating models, and risk. But strategy and execution can no longer live months apart. The technology is changing too quickly, and organizations learn far more once real users begin interacting with real solutions built on their own data.

Businesses have accumulated enough strategy decks. Now they need shorter, faster programs that combine strategic thinking with execution so they can test ideas under real operating conditions.

A good AI strategy should provide direction, but it should also produce something the company can evaluate, like a working solution, evidence that the business case is sound, or a clear reason not to continue investing in the idea.

Why we developed Explore-to-Value (e2V)

This thinking is one of the reasons we developed our Explore-to-Value (e2V) framework at PivotX. We wanted to bring business discovery, prioritization, strategy, and execution much closer together.

Our SPARK framework supports the first stage of e2V. We look at how the business operates, how customers and employees experience its processes, what data is available, and where work slows down or breaks. We then prioritize the opportunities we find based on business impact, feasibility, data readiness, and whether people are likely to use the result.

e2V takes the highest-value opportunities further by bringing strategy and execution together. Instead of spending months determining what might work and then beginning implementation, teams begin by testing the strongest ideas and learning from the result.

The desired outcome isn’t simply an AI roadmap. Something in the business should already be different. Perhaps a workflow now takes hours instead of weeks, or employees no longer have to perform a repetitive task manually. If the proposed solution did not work, the organization should know why before committing more resources to it.

Progress creates momentum

As AI capabilities become increasingly accessible, most organizations will eventually have access to similar models, tools, and infrastructure. Technology alone will become less of a differentiator. What companies do with that technology will matter more than which model they can access. That will depend on how well they understand their own operations and how carefully they choose where to apply AI.

Before choosing a use case, look at where the business is struggling. Repetitive work, slow decisions, revenue leakage, customer frustration, and operational bottlenecks show where an inefficient process is already taking a toll.

AI has made building faster and cheaper than ever before. That puts more weight on the judgment behind each project. Companies still need to determine whether the problem deserves attention, whether AI is the right way to address it, and what a worthwhile result would look like.

The companies that make the most progress with AI may not be the ones with the most pilots or the longest list of use cases. They will be the ones that become exceptionally good at finding the friction that matters, removing it, learning from the result, and moving on to the next one.

That is how adoption creates momentum, and how momentum eventually becomes transformation.

‍