Ask two refurbishment operations for their repair rate and you will get two numbers that cannot be compared. Not because either is dishonest, but because "repair rate" is not a defined term: it depends on what entered the count, what counted as success, and over what period. Circular-economy reporting is full of numbers like this: confident, published, and not comparable to anything. The metrics below are the ones that actually steer an operation, each with the definition it needs to mean something.
The two mistakes that break almost any metric
Before the individual measures, the two errors that make them useless. Both are about context, not arithmetic.
The first is a missing denominator. A percentage without a stated intake is not a result. Repair yield on a flow of three-year-old corporate laptops and repair yield on a mixed consumer flow of unknown age are different problems wearing the same number: the first will always look better, and it says nothing about which operation is run well. Any yield figure needs its intake profile attached: device categories, age, source, and how they arrived.
The second is a moving boundary. Does a device stripped for working parts count as a success or a failure? Does a unit sold as "for parts" count as reuse? Does a repair that comes back under guarantee still count in the original month? Each answer is defensible; what is not defensible is changing it between reporting periods, or leaving it unstated so that readers assume the most flattering one.
Operational metrics: is the workshop working?
Repair yield is the headline productivity measure: the share of devices entering the workshop that leave working and saleable, not scrapped or harvested. It compresses technician skill, spare-parts availability, diagnostic accuracy and intake quality into one number, which is what makes it useful and what makes it easy to misread. Track it per device category, not as a single site-wide figure.
First-time fix rate measures how efficiently those repairs happened: the share that pass final quality control on the first attempt, without rework. A unit that comes back for a second pass consumed two sets of parts and two blocks of technician time, and delayed everything queued behind it. Low first-time fix rarely means careless technicians. It usually points to incomplete diagnosis at intake, a procedure missing a verification step, or spare parts of uncertain quality.
Cycle time, or dwell time, is how long a device spends between arriving and being ready to sell. It matters commercially because used-device prices fall continuously: a unit held for eight weeks is worth measurably less than the same unit held for two. Break it down by stage, because the delay is almost never in the repair itself. It is in waiting for a part, waiting for a decision, or waiting in a queue.
Grade accuracy is the gap between the grade assigned in the workshop and the grade the buyer confirms. It leaks margin in both directions: over-grading produces returns and disputes, under-grading gives value away silently. The silent direction is the one that goes unmeasured for years.
Commercial metrics: is repair paying?
Recovered value per device is the number that decides whether the operation has a reason to exist. It compares what a device returns after repair against what the same device would have returned as materials. The comparison is not close: a device that is repaired and resold whole returns a multiple of its material worth, which is why every point of repair yield moves the business case rather than trimming it.
Cost per device processed is the counterweight, and it has to include everything: labour, parts, consumables, floor space, tooling, logistics and the cost of the units that were processed and never sold. Costing only successful repairs is a common and comfortable error: the failures consumed resources too, and they belong in the denominator.
Parts self-sufficiency, the share of spare parts sourced from harvesting, not purchased, becomes decisive as models age. Manufacturer supply for older devices thins out and then stops; an operation that harvests systematically can keep repairing long after new parts are unavailable, and at a materially different cost base.
Compliance metrics: can you prove it?
Traceability coverage is the share of devices with a complete, retrievable record: intake, diagnosis, erasure result, repair history, grade, disposition. The honest measure is not whether the system is capable of holding that record but whether it actually did, for every unit, last month. Ninety-five percent coverage means one device in twenty cannot be accounted for, which, under audit, is the one that will be asked about.
Reuse rate versus recycling rate is the split that WEEE reporting turns on. Equipment prepared for reuse leaves the waste stream; the same equipment sent to recycling is counted as waste. The distinction is regulatory and commercial at once, and proving which side a device fell on requires documentation per device, not per pallet.
Sanitisation outcomes should be reported as a distribution, not a headline: how many devices were cleared, purged, or physically destroyed, and how many failed verification and were escalated. An operation reporting a 100% erasure success rate is either not verifying or not reporting the failures: some media always fails, and the destruction path is the correct answer when it does.
On carbon figures
Avoided emissions is the metric most often published and least often comparable. The figure depends entirely on the counterfactual chosen, what would have happened to this device otherwise, and on the emissions factor assumed for manufacturing the new device it displaced. Both are assumptions, and small changes in either move the result substantially.
That does not make the number worthless, but it does mean it should never travel without its method. State the counterfactual, state the factors and their source, state the scope. A carbon figure published with its methodology is a credible claim; the same figure published alone is a marketing number, and under tightened rules on environmental claims it is an increasingly expensive one.
Start with fewer
An operation that tracks four metrics well outperforms one that reports twenty badly. If you are starting: repair yield by category, first-time fix rate, cost per device processed, and traceability coverage. Those four cover whether the work succeeds, whether it succeeds efficiently, whether it pays, and whether you can prove any of it.
Write the definitions down before collecting the data, including the boundary decisions: what counts as intake, what counts as success, when a device leaves the count. Then keep those definitions stable. A metric whose definition drifts is worse than no metric, because it produces a trend line that looks like information.