AI

Not Every AI Opportunity Deserves to Become a Product

Kartik Daware
Kartik DawareTechnical Lead, FEV Group
October 15, 202611 min read

When technology makes almost anything possible, knowing what not to build becomes a competitive advantage.

There is a peculiar problem Product Managers are facing today. For perhaps the first time, the list of things we can build is expanding faster than our ability to decide what is worth building.

An AI model can summarize documents, generate presentations, analyze customer conversations, write code, automate workflows, and answer questions in natural language. Every new capability opens up another set of possibilities. And every possibility can quickly become a product idea.

The temptation is understandable. A team discovers an impressive capability, imagines how it could be applied to its industry, and starts discussing what the product might look like.

Before long, there’s a prototype, a roadmap, and a business case.

But somewhere in that excitement, an important question can get lost.

Just because AI makes something possible, does that make it worth building?

I find this question particularly interesting because the challenge facing Product Leaders is changing. Access to technology is becoming less of a differentiator. The difficult part is increasingly about identifying where that technology creates meaningful customer value and a sustainable business advantage.

And those are two very different things.

The Problem With Starting From What AI Can Do

Imagine a company that develops software for customer support teams.

Its Product Manager discovers that a new AI model can generate remarkably natural responses to customer queries. The team sees an opportunity to introduce an AI-powered reply assistant.

The concept sounds promising. Support agents spend considerable time responding to customers, and AI can generate responses in seconds. Surely that must translate into productivity gains.

But what if writing responses isn’t actually the team’s biggest problem?

Perhaps agents spend most of their time searching across disconnected systems for accurate information. Perhaps the difficult part is interpreting complex refund policies or coordinating with other departments. Perhaps responses are already generated from templates, and writing them takes relatively little time.

In that situation, an AI reply generator might be technically impressive without addressing the most expensive part of the workflow.

The company has identified a capability, but not necessarily an opportunity.

This distinction matters because AI makes it easier to confuse a compelling demonstration with a compelling product.

A demonstration shows what technology can do under favorable conditions. A product has to deliver value repeatedly, in the messy environment where customers actually work.

And that requires a different kind of thinking.

The Three Questions Behind Every AI Product Opportunity

When evaluating AI opportunities, I think Product Leaders need to connect three questions that are often considered separately.

What does the customer need? What can AI improve? And why would that improvement matter to the business?

The first question establishes whether there is a meaningful problem. The second examines whether AI provides an advantage over existing approaches. The third determines whether solving the problem creates enough value to justify the investment.

Consider an AI feature that reduces the time required to complete a task from ten minutes to eight.

That is a measurable improvement. But is it meaningful?

If the task happens once a month, the benefit may be negligible. If it happens thousands of times every day across an enterprise, the same improvement could be significant.

Now consider a different feature that reduces a task from ten minutes to two, but requires users to verify every output because errors could have serious consequences.

The apparent time saving might disappear once the cost of verification is included.

This is why AI opportunities cannot be evaluated through model performance alone. Product Leaders need to understand the complete economic and behavioural context in which the technology operates.

A useful opportunity sits at the intersection of customer importance, technological advantage, and business viability.

Miss one of these, and the product may struggle regardless of how impressive the AI appears.

Product Management Course Mentor Engaging Banner

Sometimes the Best AI Decision Is Not to Use AI

There is an assumption quietly entering product conversations that if AI can perform a task, AI should perform it.

I don’t think that assumption holds up particularly well.

Consider a system that needs to extract a fixed set of fields from standardized invoices. If the documents follow predictable formats, a conventional rules-based solution may be cheaper, faster, and easier to maintain.

Introducing a generative AI model could increase flexibility, but it might also introduce variable costs, inconsistent outputs, additional testing requirements, and new failure modes.

AI becomes valuable when its capabilities address complexity that conventional approaches struggle to handle.

For instance, extracting information from highly varied, unstructured documents may justify an AI-based approach. So might interpreting ambiguous customer requests or identifying patterns across large volumes of qualitative feedback.

The point isn’t that AI is inherently better or worse than traditional automation.

It’s that the technology should earn its place in the product.

For Product Managers, that means comparing AI against the customer’s current experience and against simpler alternatives, rather than assuming that an AI-powered solution is automatically more valuable.

Sometimes a better search function, a clearer interface, or a well-designed automation solves the problem more effectively.

And choosing the simpler solution isn’t a failure of innovation. It is evidence of product judgment.

A Framework for Deciding What Is Worth Building

With so many AI opportunities competing for attention, Product Leaders need a way to distinguish genuine opportunities from interesting possibilities.

I would evaluate potential investments across five dimensions.

1. Customer problem: Is there a meaningful need?

Start by understanding the problem independently of the proposed technology. How frequently does it occur? What does it cost customers in time, money, effort, or missed opportunities? How are they addressing it today?

A strong AI capability applied to a weak customer problem rarely creates a compelling product.

2. AI advantage: Does AI make the solution meaningfully better?

Evaluate what AI contributes that existing approaches cannot deliver as effectively. That advantage might involve handling unstructured information, adapting to varied contexts, improving predictions, or making complex tasks more accessible.

The improvement should be significant enough for customers to notice and value.

3. Business value: Can the improvement support a viable business?

A product may deliver customer value without creating an attractive commercial opportunity.

Product Leaders need to consider willingness to pay, potential cost savings, revenue impact, and the economics of delivering the AI capability. Inference costs, monitoring, support, and human review can materially affect profitability.

4. Feasibility: Can the experience work reliably in practice?

An impressive prototype doesn’t guarantee a dependable product.

Data quality, integration requirements, latency, accuracy, privacy, and operational complexity all influence feasibility. The question is not simply whether the model can perform the task, but whether the complete system can perform it consistently under real conditions.

5. Defensibility: Why would customers choose this product over alternatives?

If competitors can access similar models, the AI capability itself may offer limited differentiation.

A stronger advantage may come from proprietary data, deep workflow integration, distribution, customer relationships, or the ability to deliver a superior end-to-end experience.

These dimensions shouldn’t be treated as a checklist where a product wins simply by scoring highly on most criteria. A critical weakness in one area can undermine the entire opportunity.

A product with strong demand but unacceptable reliability may not be viable. A technically excellent product without a meaningful customer problem may never find sustained adoption.

The objective is not to identify the most impressive AI application. It is to identify the opportunity where AI changes the economics or experience of solving a valuable problem.

What a Product Teardown Can Reveal

Consider two hypothetical approaches to introducing AI into a customer support platform.

The first company launches an AI assistant that drafts responses to incoming queries. Its primary selling point is that agents can generate replies faster.

The second company investigates the support workflow and discovers that agents lose significant time moving between systems to find account details, order histories, and policy information.

Instead of focusing only on response generation, it builds an AI-assisted experience that retrieves relevant context, identifies the applicable policy, recommends a resolution, and prepares a response for the agent to review.

Both products use AI.

But they solve different problems.

The first improves one activity within the workflow. The second attempts to reduce the effort required to complete the workflow itself.

That difference could have important consequences for customer value.

The second approach may reduce resolution time, improve consistency, and lower the effort required to handle complex requests. It also creates opportunities for deeper integration into the customer’s operations.

Of course, those benefits would still need to be validated. The system would need reliable data, appropriate safeguards, and clear human oversight.

But the strategic starting point is stronger because the product is designed around an outcome rather than an isolated technological capability.

This is what makes product teardowns particularly useful when evaluating AI opportunities. Looking beyond the feature list reveals where the product creates value, how it fits into customer behaviour, and whether its business model can support sustainable growth.

The Hidden Cost of Building the Wrong AI Product

One reason AI opportunity selection matters so much is that the cost of a poor decision extends beyond development.

An AI product can be relatively quick to prototype while remaining expensive to operate, maintain, and improve.

Teams may need to manage changing model performance, monitor output quality, maintain integrations, handle data governance requirements, and support customers when the system behaves unexpectedly.

If the underlying problem isn’t important enough, these costs can overwhelm the value the product creates.

There is also an opportunity cost.

Every team working on an AI feature that customers don’t particularly need is a team not working on a problem that might genuinely improve the customer experience or business performance.

This is especially important in organisations pursuing AI transformation across multiple business functions. A long list of pilots can create the appearance of progress, but experimentation alone doesn’t establish strategic impact.

The more meaningful question is how many of those experiments become dependable products or workflows that deliver measurable value.

I think this is where Product Leadership becomes particularly important. The role isn’t simply to encourage AI experimentation. It is to create the discipline that determines which experiments deserve further investment.

What Product Leaders Need to Stop Doing

As AI capabilities continue evolving, I believe Product Leaders will need to become more selective, not less ambitious.

Three shifts are particularly important.

  • Stop treating AI capability as a product strategy. Knowing that a model can perform a task is only the beginning. A strategy explains whose problem is being solved, why the solution matters, how it will be adopted, and how the business will capture value.
  • Stop measuring progress through the number of AI features launched. Feature releases and pilots are indicators of activity. Stronger measures include improvements in customer outcomes, workflow efficiency, retention, revenue, or operating costs.
  • Stop assuming technological sophistication creates differentiation. Customers rarely care how complicated the underlying technology is. They care whether the product helps them accomplish something important, reliably and with less effort.

None of this means organisations should become overly cautious about AI. Experimentation remains essential, particularly when the technology is evolving quickly.

But experimentation becomes far more useful when teams know which assumptions they are testing and what evidence would justify further investment.

The Competitive Advantage of Knowing What Not to Build

AI is changing the economics of product development. Capabilities that once required substantial investment are becoming more accessible, and the distance between an idea and a working prototype continues to shrink.

That creates extraordinary possibilities.

It also creates an uncomfortable reality: when more companies can build similar capabilities, the ability to build them becomes less distinctive.

The advantage increasingly lies in understanding customers deeply, recognizing opportunities others overlook, and connecting technological possibilities to business outcomes.

For me, that is the most interesting part of AI Product Strategy. The challenge isn’t simply keeping up with what the technology can do. It’s developing the judgment to decide where those capabilities genuinely belong.

Some opportunities will deserve investment. Others will be better addressed through conventional technology. And some ideas, however impressive in a demonstration, won’t deserve to become products at all.

Being able to make those distinctions is not a limitation on innovation. It is what makes innovation commercially meaningful.

So perhaps the most important question for Product Leaders isn’t, “What can we build with AI?”

It’s this: If we removed the AI from our proposed product, would the customer problem still be important enough to solve?

If the answer is no, we may be more excited about the technology than the opportunity.

Kartik Daware
AuthorKartik DawareTechnical Lead, FEV Group