The Market Software Left Behind — and Why That's Changing

The terrain

The physical economy — manufacturing, freight and logistics, energy and utilities, distribution — makes, moves, powers, and delivers everything else. It's the largest single share of the US private economy: 31% of GDP and roughly 50 million jobs, about 30% of US employment (BEA/FRED; BLS). The scale is clearest against tech. The entire US Information sector — software, internet, media, telecom — is just 5.6% of GDP and 2.77M jobs. The physical economy is about 5× its output and 17× its workforce. Yet it's the least-digitized part of the economy, running at an estimated 18% of its digital potential (MGI). Two decades of software barely touched it.

It runs on people — and the constraint is judgment

These sectors are extraordinarily fragmented. In trucking, the spine of the freight economy, 91.5% of the ~580,000 active carriers run ten trucks or fewer, and 99.3% run 100 or fewer (ATA/FMCSA, 2025). By firm count, 98.3% of the country's 239,265 manufacturers are small businesses under 500 employees (NAM / US Census SUSB). These are thin-margin operators, not giants.

The work is operational: thousands of coordinated decisions a day, under constant exceptions — the truck runs late, the supplier slips, the spec changes. Handling them takes tacit, relationship-based expertise: which carrier to trust on a lane, the person at the port who can clear a container today, built over decades and written down nowhere. "We know more than we can tell," as Polanyi put it. And that expertise is retiring. Roughly 10,000 Americans reach retirement age every day, with 2.1M US manufacturing roles projected unfilled by 2030 (Deloitte / The Manufacturing Institute) and a driver shortage heading past 160,000 (ATA). The physical floor is automating — 4.66M industrial robots now run globally (IFR, 2025) — but a robot moves the box; a person still decides which box, when, and what to do when it goes wrong. The bottleneck is the office, not the floor.

Thirty years of software — and why it never did the work

Technology never really penetrated these industries — too fragmented, too domain-specific, every operator different. So it arrived in three lighter waves: record (ERP, 1990s), visibility (TMS, telematics, CRM, 2000s–10s), and assistance (RPA, then AI copilots). Each was worth adopting; none did the work. The pattern held — record → visibility → assistance — and execution never left human hands. The reasons are structural. Tacit work breaks software built on explicit rules. Generic AI has no domain knowledge: 95% of enterprise GenAI pilots show no P&L impact (MIT, 2025). Data sits fragmented across ERP, spreadsheets, email, and phone. And enterprise software was priced for enterprises — 1.4–3.2% of revenue on IT against 7–11% in financial services — so the small operators who are this economy were never really the customer.

Why it worked everywhere else — and not here

The same thirty years remade finance, media, and advertising — because there, the work is information. A payment is already a database entry, so software is the medium of the work itself. In the physical economy, data only describes the work: marking a load "delivered" doesn't deliver it. Capital followed the digitizable work — fintech alone raised $51.8B globally in 2025 (Crunchbase), many times what logistics tech drew. These industries were never "behind." The software on offer simply fit the shape of other industries' work.

Why now — and where it goes

Two walls fall at once. AI can finally capture the domain knowledge — where to move the box, and why — drawing an operator's judgment out through conversation instead of rules. And it can act on it: ERP, TMS, and telematics now expose APIs, so a single layer can sit on top of the systems already in place and execute inside them the way a person does, instead of ripping out thirty tools for one nobody has time to learn.

The pressure has never been higher: the labor cliff above, plus rising demand and volatility. Reshoring is already visible in the data — US manufacturing-construction spending hit a record ~$235B in 2024, more than double 2020 (Census/FRED). More to make and move; fewer people who know how.

The trajectory is the same line, one step further: record → visibility → assistance → autonomous operations. Without it, expertise retires faster than it can be replaced and costs climb. With it, capacity decouples from headcount, and the mid-market finally gets what enterprise software never priced for it. That last step is the one Evos is building.

Sources: BEA/FRED, BLS, MGI/McKinsey, ATA/FMCSA (American Trucking Trends 2025), NAM / US Census SUSB (2022), Deloitte & The Manufacturing Institute, IFR World Robotics 2025, MIT Project NANDA 2025, Crunchbase, US Census/FRED. Every figure verified against a primary source.