The OPEX leak that is designed to look normal

Every operations head has a version of this problem. A cluster of sites where diesel consumption sits above the norm and generators occasionally run dry, taking the site with them. The variance is small enough each month to be absorbed into an average and large enough over a year to matter. It gets discussed, attributed to load or to ageing equipment, and carried forward.

I want to make one argument about that pattern, and it is narrower than the usual advice about tightening fuel management. It is this: the most expensive kinds of OPEX leakage are not visible to your instruments, because they are constructed to produce a normal reading. Not concealed from the report — represented correctly in it.

A site I dealt with in western Uttar Pradesh hosted three ground-based towers for major operators. Consumption was high, run-dry outages were repeated, and every entry in the system was correct. Fuel filled was recorded. Run-hours were recorded. Outages were logged and tickets closed within the process.

The installation itself had been altered. Two generators had been disconnected and their load rerouted onto a single unit which was also carrying premises that had nothing to do with the site. From the outside, and from the data, this looked like three towers on shared infrastructure operating with slightly poor efficiency. In reality one machine was doing the work of three and paying for a fourth.

Restoring each site to independent operation moved uptime up around five per cent and energy cost down around ten across the cluster. Those numbers are the least interesting part of the story. The interesting part is the class of problem: an operation where the reporting was working perfectly and the reality it described did not exist.

You do not need a special audit to look for this. You need three questions asked consistently, and the willingness to go and stand in the place when the answers are unsatisfying.

The third question is the one that does the work. In most operations the answer is nobody — not because people are unwilling, but because variance explanation was never assigned to a role. It sits between the field partner who reports it, the engineer who accepts it, and the finance team who books it, and a thing that sits between three people is owned by none of them.

That is why this is a governance problem rather than a fuel problem. The instrument that finds it is not a better dashboard. It is a named owner, a reconciliation they have to sign, and an expectation that unexplained variance results in someone going to look.