When Does the Repair Clock Start: MTTR Inflation and the Real Cost of Downtime

In the downtime table presented at the monthly operations review, last month's line interruption appears as forty-five minutes; the hour at which that same line resumed producing saleable output, recorded in the shift log, falls in the afternoon. The two records do not contradict one another, because they measure different things — one captures the interval a technician spent at the machine, the other the interval during which the plant could not produce. The difference is almost never raised in the meeting, since each figure is correct within its own definition and nobody requests agenda time to interrogate a definition. Senior management, reading the table, sees an incident that maintenance resolved in forty-five minutes and an incident that cost planning half a shift occupying the same row.

The second manifestation of the same pattern is the divergence between the curve on the maintenance dashboard and the reality on the calendar. Mean repair duration improves quarter over quarter while the total hours of unplanned downtime hold flat or climb, and the slippage in customer delivery dates appears on neither curve. The maintenance manager reports progress against his own metric, the production planner quietly enlarges his buffer, and the commercial side adds days to the delivery commitment without announcing it. Each of the three functions behaves rationally by its own measure, none observes the correction being made by the others, and the institution's aggregate loss is concealed in three separate places in fragments small enough to escape notice.

This divergence has a name — MTTR inflation, the regular excess of realized repair duration over assumed duration, together with the failure of that excess to surface in the record system. At the center of the mechanism sits a decision about measurement boundaries: whether the clock starts when the operator notices the fault, when the work order is opened, or when the technician reaches the equipment; and whether it stops when the line produces its first part, or when full-rate operation within specification has been verified. The territory between those two decisions contains diagnosis, waiting on spares, calling an external service provider, obtaining spending approval, safety isolation and requalification. Drawn narrowly, the boundary isolates the domain the technical team actually controls, which is a defensible narrowing for purposes of staffing and training.

The difficulty therefore arises not from the narrowing itself, but from the use of the narrowed figure as an answer to a different question. When an interval designed to measure technician performance is carried over unchanged into the sizing of a work-in-process buffer, the construction of a capacity plan, or the pricing of an availability undertaking, the entire structure rests on a velocity the plant does not possess. That carryover rarely reflects a deliberate decision; it happens because a single number exists and is available. Absent a request for a second measure, the number in hand continues to serve as the answer to questions nobody has framed.

A second dynamic feeding the mechanism is the definition of the event itself. A failure closed with a temporary intervention, its root cause untreated, recurs on the same equipment shortly afterward, receives a fresh work order number and enters the record as a second event. One long outage is thereby split into two short ones: the average falls, the event count rises, and total downtime remains exactly where it was. The same fragmentation occurs across weekends and shift handovers, since most record systems are configured to stop the clock at end of shift and restart it the following morning. Reporting the mean alone conceals the effect entirely, when what matters is the long event sitting in the upper tail of the distribution — the one the plant genuinely has reason to fear.

The balance sheet consequence surfaces first in working capital. Work-in-process buffers and spare parts inventories are sized, whether deliberately or by habit, as a multiple of assumed repair duration; where the assumption sits below reality, the buffer is correspondingly short, and the gap is closed through expedited freight, subcontracted production, overtime and priority order premiums — items individually small and collectively substantial. Those items accumulate in logistics and manufacturing overhead rather than in the maintenance budget, with the result that cause and effect never appear in the same report. Deterioration in inventory turnover is then commonly read as weak purchasing discipline, when the underlying driver is uncertainty on the reliability side.

The second exposed surface is contractual. An availability undertaking in an operations and maintenance agreement, a service level in a data center, a delivery commitment in a manufacturing facility — each is constructed on an assumed repair duration, and where that assumption fails to hold, the liquidated damages mechanism, the warranty scope and the deductible period under business interruption cover are triggered in concert. None of those three instruments will accept, as a defense, the argument that the committed duration was unrealistic because the measurement boundary had been drawn narrowly. At that point the institutional cost ceases to be operational friction and becomes a liability line in its own right.

The third surface emerges in a transaction or a financing process. A party conducting operational diligence, comparing the maintenance record in the data room against production and shipping records, will observe that the two series do not reconcile; that discrepancy typically converts not into a question list item but directly into a condition precedent, a price adjustment, or an expansion of the representations and warranties package. What governs valuation at this stage is not the plant's actual reliability but the demonstrable recording of that reliability, and where the record is internally inconsistent the buyer is reasonably free to price the adverse case. Combined with a repair practice resting on the knowledge of a single technician and never committed to writing, the same inconsistency is priced a second time, as key-person dependency.

The tendency is neutralized by measurement architecture rather than by individual attentiveness, and the intervention has four components. The first is the maintenance of disaggregated sub-clocks in place of a single clock: detection, response, diagnosis, parts procurement, physical repair, requalification and return to full rate are each measured separately, so that the stage at which the extension occurred ceases to be a matter of argument. The second is reporting the upper band of the distribution alongside the mean, and constructing planning decisions against the band rather than the average. The third is an event consolidation rule, under which recurrences on the same asset, in the same failure mode, within a defined window, are reported as a single event. The fourth is tracking approval latency separately as a managerial rather than a technical delay, so that time spent waiting on spending authorization, external service dispatch or a work permit is not charged to maintenance performance.

BEIREK's intervention in this area consists not of setting the plant a new target but of redefining the record itself. In operations we take on, we first commit the boundaries of the event clock to writing, rebuild the work order schema in the maintenance management system around the sub-clocks, and embed the event consolidation rule in the system; we then run a monthly reconciliation rhythm between the maintenance record and the production and shipping records, on the reasoning that the point at which the two series fail to agree is the first point to correct. The same framework is carried through to the contractual side, with availability undertakings, damages thresholds and minimum spare parts levels calibrated against the recorded upper band rather than the mean.

In a transaction or financing context the same discipline is run in reverse. Rather than accepting the stated availability figure as presented, we reconstruct an event-level timeline from raw work order records, compare it against production output and shipment dates, and frame the resulting gap not as a diligence finding but as a structural question for the investment committee: is this gap a recording problem remediable before closing, or a structural reliability shortfall requiring capital on the spares and staffing side. The two answers demand different transaction architectures — the first is addressed through a condition precedent, the second through price or a post-closing investment undertaking.

The most honest statement available about a plant's reliability concerns not how many hours its mean repair duration runs to, but where that clock is started and where it is stopped; the party drawing that boundary is deciding, knowingly or otherwise, what the institution will be permitted to know about its own capacity.