How To Maximize AI’s ROI — Without Getting Sucked Into the Money Pit

By Jeb Horton, Senior Vice President, Global Services, Hitachi Vantara 

Amid the big opportunity and expanding innovation presented by AI is the concern about return on investment (ROI). When will the massive spending produce a profit? Is the breakneck rush to AI causing companies to fall into a money pit of vague promises? While some say the AI boom isn’t generating the return and productivity envisioned, others are moving forward full-steam-ahead on a mission to automate most human jobs — counting on savings and profit that may or may not materialize.

Neither extreme captures our AI reality and how businesses can best use it.

Many enterprise customers spend heavily on text-based or multi-modal LLM usage without seeing a short-term ROI. They also spend enormous amounts developing AI solutions and buying out-of-the-box AI products and tools – and haven’t yet seen returns.

Forrester’s 2024 AI Pulse Survey found that “nearly half of AI decision-makers say their organizations expect ROI on AI investments within one to three years, while another 44% expect a longer timeframe.” The stakeholders antsy for AI use can also push for faster, quarterly returns.

The path to ROI with AI projects of every stripe is to pragmatically ensure they’re solving real-world problems in specific use cases. Investments in copilots and automation must enhance human capabilities in particular business operations and in a measurable, scaled way that yields demonstrable savings.

Related:AI in Healthcare Demands Vigilant Security Measures

Reaching AI’s lucrative sweet spot of innovation tied to financial return involves combining several strategies.

Let’s take a closer look.

Workshop, Prototype, and Start Small — Where There’s Demand

To create AI value, sales, and ROI, begin with customer AI workshops. This helps define use cases, determine models and infrastructure best suited to them, and organize teams to spin up a working prototype quickly.

Best use cases begin with customer concerns — they know their business’ pain points. A fintech customer might ask: How do I better protect financial institutions and identify bad actors more quickly with AI? A healthcare provider might want to know how to generate a better patient experience at first contact or speed up and scale the ability to see a doctor. The AI project should immediately improve a key process, using resources more efficiently while satisfying end-user outcomes.

These kinds of projects are not massive investment ideas that reinvent our world. While some venture capitalists want the unicorn and place wild bets for the chance at one big success, most CEOs can’t afford to play that game. A facilitator in a workshop can keep everyone’s eye on the ball; the goal generally isn’t to go after something big and an unknown ROI but, instead, to pursue a very pragmatic project with predictable returns.

Related:AI-Assisted Development Tools vs. Low-Code/No-Code: What to Use When

Starting small and adding to the portfolio over time is a way to build momentum gradually with reduced financial risk. Maybe a company tries out a sophisticated chatbot, then incrementally gets more sophisticated — building and deploying task automation. If you don't need much infrastructure upfront, you can better align purchases to budget, investment, and return.

Workshopping effective prototypes and starting with smaller projects that may scale is a natural fit for a flexible infrastructure consumption model, where, as you add product improvements over time and see a return, you can add more computing muscle and data power.

Be Ready for Infrastructure Costs With a Flexible Consumption Model

In the AI space, demand for infrastructure will only increase. A flexible infrastructure consumption model means customers can consume only the infrastructure they need at any given time and align it with their business goals and budgets. Being able to tie in procurement on an as-needed basis rather than buying in chunks improves ROI. A program is well positioned if it successfully integrates AI development, advanced infrastructure requirements, and flexible consumption of infrastructure resources. It can mitigate the high upfront cost of AI, measure the implications for ROI, level-set investor expectations, and establish a strong value proposition.

Related:Will AI Ever Pay Off? Those Footing the Bill Are Worrying Already

Importantly, the flexible consumption model must be easy to work with—for example, easier than big public cloud providers from both a term and contractual agreements perspective. Flexible consumption must be tailored to a customer's needs, not force them to do a mountain of homework that may or may not pan out.

Since everybody's in the AI procurement stage right now, being able to plan and scale in a more consumption-oriented way is key. You want to tailor your efforts as tightly as you can to the explosion of AI requirements, known and unknown, that you'll need to deal with. 

Know What Leads To Overspending — And Counter It

It’s a tricky time for “reading the room” with investors. Business leaders buying and building AI often try to satisfy investor expectations to be at the forefront of advanced tech and AI integration. At the same time, stakeholders are pressuring leaders to show quarterly ROI and control costs. If leaders make huge upfront expenditures on AI projects that don't ultimately work, the consequences can be dire. Rein in investor expectations early and often on how AI projects will unfold.

The pressure is on even when you start small. AI is no different than any other IT program over the last 50 or so years in this respect: What begins as a good idea for a project of limited size and scope must scale for value. It's easy to create things — even excellent technology — that may not yield a financial benefit. It’s imperative to ask and answer this question: Do I have ROI with this initiative, or am I just giving somebody a cool AI feature with no real business value? The feature may provide a different experience but an equal or lesser one in terms of value.

For example, consider an AI tool that creates software documentation. Many systems running across businesses are poorly documented, difficult to understand, and missing integration details. It’s a constant aggravation that developers adapt to in many ways. If we introduce an AI tool that enables software developers to document work better or use better documentation as they build, have we changed the economics of the situation? Maybe, maybe not. The software and systems are still running as they were. We've provided more and better documentation, but have we changed developer productivity?

Copilots that consume documentation and then have the actual ability to automate software and correctly fix problems may have ROI, too. But right now, is error detection with those reliable? Complicated security considerations factor into the answer.

Additionally, sometimes organizations try too many different AI experiments at once without planning well for resource and infrastructure expenditures. Savvy IT leaders must be able to recognize this and determine when biases influence their people’s choices in terms of projects. Some choices might not be optimal for ROI. The best IT leaders have exceptional interpersonal aptitude and technical skills and build cohesive teams. They bring the right people together for a project best suited to the organization’s interests.

The market is conspicuously toggling between pragmatism and big ambitions with AI. We all see it. The potential is enormous, but if efforts don't generate benefits early on, we’ll end up with the same unrealized value and wasted time and money that careless digital transformation and cloud initiatives created in the not-so-distant past — and still do for some enterprises today.

About the Author

Jeb Horton is Senior Vice President, Global Services, at Hitachi Vantara.