Ask private equity partners how artificial intelligence will affect their investments, and most answers are centered around cutting costs and higher margins.

Bob Root , an operating partner at Southfield Capital , gives a different response. Within five years, he argues, corporate functions such as finance, purchasing and marketing will no longer exist as departments at all. I recently spoke to Root about what he is building in their place, and his belief that small companies will take the lead on AI transformations.

Why the lower middle market might move first

Southfield buys founder-run businesses in the lower middle market, looking for an entry point of roughly $5 million to $10 million in Ebitda. Its portfolio companies sit in services that rarely attract attention and include industrial refrigeration, commercial landscaping, homeowners association property management, and physical guard security. The model is to acquire two or three platform companies and then incorporate other smaller companies on top of these.

Root's hypothesis, formed in part during his years at Microsoft, is that these businesses were historically underserved by technology. This was because it cost too much and demanded more change management than they could absorb. As these barriers fall, he argues, founder-led firms hold an advantage that large corporates cannot buy: few approval layers which minimizes cumbersome decision making.

The early evidence in his portfolio is modest but concrete. In a field service business, Root's team found that bookkeeping and procurement was too time-intensive for field technicians. So, with another of their portfolio companies, Contextual.io , they built an accounting and procurement tool which handles pricing, purchase orders and approvals in the background. The tool was built ignoring traditional departments.

Whilst this is a successful AI implementation, the wider data on AI transformation is far less encouraging. S&P Global Market Intelligence's 451 Research survey of more than 1,000 organizations found that the proportion of organizations abandoning most of their AI initiatives rose from 17% to 42% in a year, with the average organization scrapping 46% of proofs of concept before production. A preliminary report from MIT's Project NANDA , which has not been peer reviewed, put the proportion of enterprise generative AI pilots with no measurable profit and loss impact at about 95%.

The fourth leg, and the moat underneath it

Business planning can be built on three pillars: people, process and technology. Root treats AI as a fourth leg that cuts across the other three, closer to an operating system than a tool.

He is skeptical of buying an operating system off the shelf, on the basis that anything available to everyone commoditizes the function rather than differentiating it. Southfield instead applies what he calls a "data moat thesis" during due diligence, identifying what a business actually knows that competitors do not. This according to Root “stretches their lead."

Ironically, the MIT report points the other way. In its interview sample, externally partnered tools reached deployment roughly twice as often as internally built ones. Root is betting that the competitive advantage created by owning the data matters more than the software. It is a reasonable bet but not yet a settled one.

However, Southfield’s thesis is not new. In 1990, Michael Hammer wrote in the Harvard Business Review about the need not to automate but to obliterate, and that reengineering promised to replace functional silos with cross-functional processes. Some of that happened but most organizational charts survived. So why should this attempt end differently?

The need for AI Ownership

As I noted in my June Forbes article , AI projects rarely fail because the technology underperforms but rather because nobody owns the outcome. Root enforces the point bluntly: "If it's not owned by the CEO, we don't do it."

He describes clients as living with "some sort of level of AI paranoia", the sense of knowing that something must be done without knowing what. To overcome this, Southfield initiates peer conversations between portfolio CEOs, and a one-day workshop in which teams are asked to reimagine an operation from scratch.

However, even CEOs need to curb their over-enthusiasm. Root recalls the chief executive of an industrial refrigeration company, initially a skeptic, who tried to expand a focused two-project initiative into eight simultaneous ones and diluted the return in the process. In my experience this pattern is common, and the discipline of holding a leadership team to solve a single expensive bottleneck is harder than it sounds.

AI can and will force organizational charts to change. The question is whether human inertia will create minor adjustments or will AI be the driving force to remove departments?