The most valuable automation target is often not the work on the organisation chart. It is the work between the boxes.
A driver moves a shipment. A technician repairs a machine. A buyer places an order. Those are the visible activities. Behind each one sits another layer of work: scheduling, dispatching, checking, approving, updating, reconciling, chasing, explaining and handling exceptions. This coordination layer is easy to miss because no single step looks large enough to transform. In aggregate, it can consume a surprising share of the operation.
Most AI products approach this problem through employee assistance. They give someone a new interface and ask that person to remember when to use it, judge the output and copy the result back into the system where the work already lives. The tool may be capable, yet its value still depends on thousands of small acts of adoption.
Employees do not particularly want another place to do the work. They want less work.
That suggests a different design. Put the agent across the systems where coordination already happens. Let it extract the invoice, compare it with the purchase order, prepare the routine approval and route the exception to the person who owns the judgment. The useful interface is not necessarily a chatbot. It may be a short queue containing only the cases that genuinely require attention.
This is why coordination cost is a better place to begin AI use-case discovery than visible task volume. Map the handoffs. Find where information is re-entered, translated between teams, checked against a rule, delayed for approval or reconciled after the fact. Then ask how frequent the path is, how standard the normal case is, how expensive the exceptions are and what evidence would show that the process improved.
There is, however, a seductive mistake hidden in this argument. Low margins make savings look dramatic when measured as a percentage of profit. If a company earns a three percent margin, saving three quarters of one percent of revenue raises profit by twenty-five percent. The arithmetic is correct. It does not prove that the company has found a durable advantage.
Time saved is not cash saved. Capacity released is not margin realised. Avoided hiring, working capital, recovered revenue and reduced risk are different economic outcomes. A business case that adds them together as though they were interchangeable is manufacturing value on a spreadsheet.
Every proposed automation needs a conversion chain. Which activity disappears? What resource becomes available? What will the company do with that resource? Will it reduce spending, avoid future spending, serve more customers, lower losses or improve prices? How long will conversion take? Who owns it? Without those answers, the project may create genuine operational improvement and still produce no visible earnings effect.
Competition creates a second problem. Thin-margin industries are often thin-margin because products are hard to differentiate and customers have bargaining power. Once competitors adopt the same efficiency, some of the gain will flow into lower prices. The early mover may bank a temporary uplift, but the industry can still return towards its old margin structure.
Lasting advantage therefore has to come from more than a cheaper process. The deployment might compound proprietary operational data, reduce service variance, respond faster to disruption, deepen customer integration or make a different business model possible. Efficiency is the opening. The moat, if one exists, is what the organisation learns and becomes able to do.
Calling this infrastructure should not mean making it invisible or unaccountable. A process owner still needs to see what happened, pause the workflow, change a rule, inspect an exception and bring a person back in. Good automation removes routine interaction while preserving meaningful control.
The biggest AI winners will not necessarily be the companies with the lowest margins. They will be the companies that can see the work behind the work, redesign it without losing control, convert operational gains into cash and build an advantage that competitors cannot immediately price away.