Breaking the Hype Cycle

That’s where GenAI should be today. But so far, GenAI hasn’t done anything it was supposed to do. It didn’t snowball interest. It arrived with fanfare and fireworks. It was loud and bright and immediately stole our attention.

More than two years later and with few proven use cases for how it will generate returns, enthusiasm should be waning, but instead it’s continuing to rise — as though a snowball could, upon reaching the top of a mountain, simply take flight.

But 2025 does signal a shift of a different kind. Although interest in the technology hasn’t subdued, it is maturing. It’s focusing on a target.

The target: demonstrable ROI.

The organizations that can use AI to decrease costs or increase revenues will become GenAI’s early winners.

We already know three essential strategies to aid that effort. And some proven use cases have emerged, pointing the way toward returns.

The ROI Trifecta

Regardless of your industry, every GenAI model requires certain ingredients to perform its best and maximize its potential for profits. So far, they include:

1.  Clean data: If you are what you eat, the same holds true for GenAI. To deploy an accurate and reliable model, you need to feed it clean, well-governed data.

As Shawn Rogers, CEO of the research firm BARC noted early this year, 45 percent of leaders point to data quality and lineage as the reason for stalled AI innovation. I have to agree: Cleaning your data is arguably the most important thing you can do to prepare your AI for success.

2.  Combined models: Each GenAI model comes with its own set of pros and cons. Large language models are exceptionally fast, for instance, but they’re also prone to inaccuracies. Small language models are efficient and cost-effective, but they have limited capabilities. Deep learning models are highly adaptable and can take on huge amounts of data, but they’re expensive to run.

The solution is to use a multi-model architecture, which integrates models and allows you to route tasks to AI components according to their strengths. This is the sweet spot.

3.  Accuracy: Beyond using clean data to fuel your AI model, you can increase its accuracy for your specific needs through RAG architecture. RAG stands for retrieval augmented generation, and it’s a way to get domain-specific answers from a commercially available LLM.

Here’s how it works: When a user types a question into a chatbot or AI assistant using a prompt, the question is filtered through relevant data (that’s retrieval). That context is augmented with a query, then it’s fed through an LLM to generate a response.

Use vectorization to feed an LLM with your own data, and then with RAG, you can overcome the limitations of an LLM trained on public data.

While it will likely take time before enterprises reap the biggest rewards of GenAI, you can start seeing returns on your investment today … if you know where to look.

Consider what organizations in two sectors, healthcare and financial services, are doing to immediately increase efficiency and lower workloads.

Healthcare: Less Paperwork, More Patient Care

Here’s an easy trick for effectively and safely using GenAI: Make a matrix. If one axis represents an item’s value, and the other axis its risk, plot all your possible AI use cases accordingly. Go after the ones in the “high value, low risk” quadrant first.

In healthcare, plenty of tasks land in that zone, such as ambient note generation, suggested actions and revenue cycle management.

By focusing AI efforts on these areas rather than flashier but riskier efforts around clinical research or chat bots, health systems are already seeing strong results.

Intuitive technology reduces manual data entry, helping clinicians spend more time with patients. Electronic health records are no longer just a matter of record keeping — they’re at the center of informing and improving care.

Financial Services: Intelligent Insights

By adding GenAI to existing business intelligence systems, data engineers and business analysts can improve enterprise decision making.

AI agents can see across data fabric architectures, help redefine relationships and patterns and validate both the provenance and validity of data. As a result, business users can ask an AI assistant questions in a natural language (for example, “Give me the trades for Goldman Sachs”), and the AI will not only help them clarify relationships between data but improve their questions and decisions over time.

All of this leads to higher returns.

A Snowball’s Chance

What’s next for GenAI and its defiant trajectory? The truth is that no one knows, because GenAI is designed to continually build upon and improve itself, constantly surprising us.

If it were following the Gartner Hype Cycle, GenAI would be beginning to climb out of the “Trough of Disillusionment” via the “Slope of Enlightenment,” that crucial period when proven cases of ROI accumulate. Then, once most ROI-generating use cases had emerged, it would finally stabilize at the “Plateau of Productivity,” where it would roll along in perpetuity.

Perhaps GenAI will succumb to some version of that model. Our flying snowball may ultimately even out and maintain altitude. Or maybe it will continue to gain momentum as it hurtles toward the atmosphere.

One thing we do know: No technology, no matter how exciting, can survive forever without that make-or-break criterion that guides every enterprise — ROI.

And that is precisely what will determine GenAI’s fate.