How Proof of Value Helps Midmarket IT Know When To Stop Spending on AI
West Monroe’s CIO explains why proof of value, a clear business sponsor and smaller bets keep AI budgets in check.
(Kevin Rooney, CIO, West Monroe)
The pressure on organizations to prove return on investment with artificial intelligence initiatives has understandably made some IT leaders trepidatious about AI spending.
As a result, many AI efforts across midmarket organizations are stuck in evaluation phases or endless pilots, or deployed only in limited use cases, as MES Computing’s recent 2026 IT Leaders Spending Intent Survey, which examined investment priorities heading into 2027, has shown.
Being cautious, by any measure, is not a bad way to deal with technology that still raises considerable concerns for IT leaders around governance and questions about measurable business value.
In an interview with MES Computing, Kevin Rooney, chief information officer at West Monroe, a Chicago-based IT consulting firm, discussed why it's critical for CIOs and other tech leaders to reach a proof-of-value stage with a vendor before spending on any AI solution.
Rooney touched on what that stage should look like, the right questions to ask before any money is committed, and how midmarket IT leaders can keep their AI spending tied to results they can measure.
Testing AI Beyond the Sandbox
Establishing proof of value requires some groundwork, and Rooney said it should begin before a vendor's product is installed in any test environment.
Every AI tool under consideration should first pass through a proof of concept (PoC) stage, along with a data review that examines the vendor’s compliance, its SOC 2 reporting and how it handles confidential information. The outcomes of the test should also be agreed on in advance.
According to Rooney, that was the playbook West Monroe used when it made early bets on AI startups for its service desk and enterprise search, and those bets paid off.
“You have to go the PoC route before you go into contracting. And you have to name a clear stakeholder and also have clear outcomes of what you’re trying to figure out from the PoC,” he said.
A well-run PoC is therefore a necessary first step, but it can only tell a buyer so much. A PoC can show whether a product works in an isolated environment, but it reveals far less about how the tool will perform once integrated into production. That is the job of the PoV stage, in which a select group of users with a specific persona puts the product to work on a defined business outcome inside the live environment.
“That’s when you resolve any integration challenges,” he said. “If people go straight from PoC to long-term contracting, they’re going to get in trouble. You learn a lot from the value.”
MES Computing has previously examined why measurable outcomes, rather than continued experimentation, are becoming the real test of AI ROI for midmarket IT.
Returns Must Outweigh Integration Costs
Insisting on a PoV stage will, of course, stretch the evaluation and add to its cost, and for midmarket firms on tighter budgets, that extra spend needs careful thought. Rooney’s answer is to make sure every AI project has a senior sponsor, such as the chief financial officer or chief operating officer, who can be “crystal clear about what the size of the prize is.” Whether that prize comes from lower costs or revenue growth, he argued, “you have to go after the big things if you’re going to invest a big amount of money.”
The expected return should also rise with the effort a tool demands, which makes solutions that already fit a company’s architecture and security standards the easier case, since there’s less investment to make. "You can’t pick everything because there are too many shiny toys,” he said, and added that the selection has to begin with the business problem, a step organizations that fail to see a return from AI tend to overlook.
“Grounding it in what the business needs is and what the most important strategic things that this tool can help transform the work you’re doing is the most important conversation before you start talking about the technology itself,” Rooney said.
Know When to Stop Spending
However carefully a product is tested, some AI projects will still fall short of expectations. The harder decision is knowing when to stop funding them.
According to Rooney, a project that keeps swallowing money without results is a sign that the right guardrails and gates were never in place.
“I firmly believe that you can get the ROI out of AI, but that doesn’t mean keep hitting,” he said.
It is therefore important for the business to agree on how much it is willing to invest in change before any AI project begins, a ceiling Rooney believes the CIO, CFO, COO and CEO should set together. Within that limit, a portfolio of smaller bets that can be scaled or dropped early is a far safer way to go than a handful of large ones, and buyers, as he put it, need “to be more agile and more nimble” in how they spread their money.
That challenge is already playing out as midmarket IT leaders build their 2027 budgets and try to determine where AI spending belongs and what other costs it might offset.
Projects that change core business processes deserve extra care too, since they depend on security and data foundations, including identity management, data governance and controls over model drift, that many organizations have yet to get right.
For Rooney, however, the main obstacle is rarely the technology itself.
“For the most part, this is not a technology problem. This is around people, culture, [and] ways of [a] working problem. Companies that get into these problems don’t have the right business leaders.”
For midmarket CIOs, Rooney's advice points to a practical takeaway: Define the stop conditions before the experiment begins. That means going into an AI investment with a named business sponsor, a measurable outcome, an agreed investment ceiling and a decision point for whether it will be scaled or dropped. If the project cannot demonstrate sufficient value within those boundaries, continuing to fund it risks turning an experiment into an open-ended bet.