The State of AI in the Physical Economy

The physical economy is the largest employment sector in the world. Construction, manufacturing, oil and gas, utilities, field service, agriculture, mining, and logistics collectively employ more than 80% of all working people globally. These workers run job sites, operate equipment, dispatch crews, manage field assets, and keep infrastructure functioning.

Enterprise software was not designed for how they work. ERP systems assume office access. CRM platforms assume desktop users. Workflow tools assume keyboard input. The result: most physical economy businesses run on a combination of paper, phone calls, spreadsheets, and the accumulated knowledge of their most experienced people. When those people leave, the knowledge leaves with them.

Less than 1% of enterprise software investment over the last two decades has been directed at deskless workers. That gap is the opportunity. The intelligent systems that fit how field operations actually run — mobile-first, voice-capable, offline-tolerant, designed for crew leads not office managers — are only beginning to exist.

Field billing depends on someone remembering what happened and converting it into an invoice. That chain fails repeatedly: add-on work gets done and not documented, parts get installed without paperwork, hours run long with no record. It is not intentional. It is structural — the billing process was not built to capture what actually happens in the field.

When billing is triggered automatically by job completion data captured on-site, the gap closes. Most operations that deploy billing automation recover 5-15% of revenue they were previously leaving behind. That recovery typically more than covers the cost of the system within the first 60-90 days.

At most field service companies under 50 employees, dispatch is a person with a phone. That person holds the schedule, the crew availability, the skill certifications, the job priorities, and the customer relationships in their head. It works until it doesn't — when that person is out, when the schedule is full, when two emergency calls come in at the same time.

Autonomous dispatch changes this. The right crew to the right job based on skill set, location, certification, current load, and customer history — determined in seconds, not phone calls. A dispatcher managing 15-25 crews spends 3-6 hours per day on coordination tasks that an autonomous agent handles without a human in the loop. That time goes back to the business.

Job completes Monday. Paper ticket gets to the office Thursday. Someone inputs it Friday. Invoice goes out Wednesday. The customer pays the following week. That is 14-21 days from completed work to received payment — and it is the standard operating cycle at most trades businesses under 200 employees.

Cash flow in field service is a timing problem more than a revenue problem. The work is done. The customer is ready to pay. The bottleneck is everything between job completion and invoice delivery. Automating that sequence — triggered by field completion data — compresses the cycle to same-day. The revenue was always there. The process was just slow to collect it.

Job costing in construction is almost universally done after the fact. The project ends, the invoices come in, the hours get tallied, and someone discovers the margin was 4% instead of 12%. At that point, the project is closed and the only output is a lesson learned that often doesn't change the next estimate.

Real-time job costing changes what is possible. When labor hours, materials costs, subcontractor invoices, and equipment usage are captured as the job runs, overruns surface while there is still time to respond. A project manager who knows on day 14 that the job is trending 8% over budget can make decisions. One who finds out on day 90 can only invoice and move on.

Change orders are the single largest source of margin erosion and client conflict in construction. The work gets done. The paperwork doesn't follow. The conversation happens weeks later when the customer is looking at an invoice for work they don't remember authorizing. At that point, both sides are working from memory.

When change orders are issued in the field, reviewed by the office, and signed before work continues, disputes disappear. Not because clients become easier to deal with — but because there is a record. The documentation failure is the conflict. Fix the documentation and the conflict largely resolves itself.

Oil and gas, mining, and heavy industrial operations generate enormous volumes of operational data — sensor feeds, equipment telemetry, maintenance logs, SCADA outputs, GPS, environmental readings. That data sits in separate systems, reviewed separately, by separate teams, days or weeks after the events that generated it.

When that data is connected into a single operational layer and an AI system runs continuously against it, the picture changes. Anomalies surface before they become incidents. Maintenance gets scheduled based on actual equipment condition, not the calendar. Crews get dispatched based on real-time need, not the morning plan. The operations that have done this are seeing cost reductions in the range of 15-17% — not from doing less, but from doing the same work with less waste, less unplanned downtime, and fewer emergency responses.

Reactive maintenance is expensive in ways that do not show up on the maintenance line item. A pump fails on a Tuesday. The crew is pulled from another job. The part is expedited at 3x cost. The downstream equipment sits idle. The customer job misses its window. The total cost of that failure is 4-6x the cost of the repair itself — and none of that shows up in the maintenance budget.

Predictive maintenance does not eliminate failures. It eliminates surprises. When sensor data, usage patterns, and failure histories are analyzed continuously, equipment issues surface days or weeks before they become failures — ranked by consequence, not just probability. The pump that is most likely to fail next week gets maintenance this week. The one running within normal parameters waits. The shift from calendar-based to condition-based maintenance typically reduces unplanned downtime by 30-50% within the first 12 months of deployment.