Two years ago, most conversations about AI centered on whether the investment was worth the risk. That question has been settled. The harder one now is where the money should go, and how much of it.
I see this play out constantly with mid-market leaders and private equity operators. Most of them have already run the pilot phase, trying something in nearly every function. The differentiator now comes down to which of those pilots became part of how the business actually runs, and which ones stayed side projects that never crossed into daily operations.
A recent piece in MIT Sloan Management Review captures the same shift from a different angle: despite decades of investment in technology, many organizations still aren’t seeing meaningful returns, and the research points to something less tangible than the tools themselves: a workforce that’s genuinely capable of using them and a culture built around learning. I’d go a step further than that: buying the tool is no longer the hard part. Building the people, skills, and organizational capability that let the tool actually create value, that’s where decisions and challenges live now.
Why the Companies Pulling Ahead Spend Differently, Not Just More
There’s a growing misconception that technology investment means buying the newest platform, but it rarely does. Organizations succeed when they know exactly what business problem they’re solving before they open a budget line. It’s a point we’ve made before: even the most sophisticated tools fail to deliver meaningful impact without the data foundation underneath them.
That matches what we’ve seen directly with Esquire Depositions, a legal services company backed by Gridiron Capital, which saw its enterprise value increase by 10%, not from adding a new platform on top of what they had, but from centralizing data and standardizing processes that had been built ad hoc across teams for years. The tool matters less than what it’s built on.
The Four Fundamentals for Deciding What to Scale
Every technology conversation I have with a CEO or an operating partner eventually comes down to the same decision: scale it, fix it, or stop it. In my experience, four things separate those three outcomes.
- Someone owns it. Every initiative needs a person accountable for its outcome, not a committee or a rotating sponsor.
- The data underneath it is trustworthy. AI and automation are only as good as the data feeding them. Inconsistent, siloed data turns a promising tool into another thing people route around.
- People actually use it day to day. Adoption doesn’t happen because a platform launches. It comes from the tool fitting how someone already works, or from someone taking the time to redesign the work around it.
- The outcome is measurable, and tied to something specific. Revenue, cost, cycle time, customer experience. Pick one before you build, not after.
Run anything through those four and the decision tends to sort itself out. I’d scale the initiatives that already have an owner, reliable data, and real adoption. I’d keep reviewing the ones that solve a genuine problem but are still clunky in daily use. And I’d stop the ones running mainly on enthusiasm or activity metrics, the dashboards nobody built a decision around.
What to Cut, and What to Protect
That same discipline applies to the budget itself, in both directions.
On the cutting side: ownerless pilots, overlapping tools bought by three different teams for the same job, licenses nobody logs into, data projects that never connected to anything downstream. These don’t just waste money, they add complexity that slows down the next real initiative. Every one of them is easier to keep funding than to shut down, which is exactly why they pile up.
On the protecting side: cybersecurity, data quality, integration work, the customer-facing experiences that actually differentiate the business, and the AI use cases already earning their keep in daily operations. Training belongs on this list too. It’s one of the most underrated technology investments a company can make, and one of the easiest to skip when budgets tighten.
These are quieter investments than launching a new AI initiative. A press release doesn’t write itself around “we fixed our data pipeline.” But quiet doesn’t mean small. Process automation is where a lot of that quiet money is going right now, because it’s specific enough to carry a clear ROI. We’ve designed and built the workflow that took an internal compliance team at a lending company from manually auditing 30% of client calls to reviewing all of them automatically, closing a gap where the other 70% carried real exposure, since regulations treat each call as its own violation, worth $500 to $1,500 apiece.
Another automation project we recently delivered cut the manual data entry eating into an accounting firm’s lowest-margin, highest-volume service, freeing staff to take on higher-value client work without adding headcount to do it. Neither involved a flashy AI product launch. Each one solved a single, specific process and multiplied across a business, that kind of targeted fix adds up to some of the biggest ROI a company sees.
Investing with Intention
Uncertainty makes executives cautious, and that’s a reasonable instinct. The companies pulling ahead right now are still spending aggressively, but concentrating resources deliberately on the capabilities that strengthen operations, sharpen decision-making, and position AI to deliver something you can actually point to at the end of the quarter.
I don’t think the companies that outperform their peers over the next few years will be the ones with the biggest technology budgets, and that’s not even true today. In my experience guiding companies through digital transformation, real ROI comes down to two things: knowing exactly where the investment creates the most value (whether that’s in how the business runs, in what sets it apart, or in the product and service customers actually experience), and defining measurable KPIs upfront so that impact stays visible. That’s the discipline that separates real returns from expensive noise.
Credit: Source link
