Guide
Making the repair-or-replace call, and writing it so it holds
This page is about the decision and the paperwork, not the data. The measured table lives on the measured page and the product page explains what the estimator does with it. What follows is how an estimator should reason about the call, and what it costs when the call goes the wrong way.
Last reviewed . Every figure below is read from the measured corpus at build time, so it moves when the corpus does.
The asymmetry, priced
$640
median carrier-paid bumper cover, 80 settled lines
Adjusters replaced the bumper cover in 72% of 118 decisions. Writing it as a repair does not discount that part, it deletes it from the document.
The most contested panel in the corpus
47% replaced
15 decisions
The quarter panel is next at 58% of 31. Both are welded, both are genuine judgement calls, and neither has enough settled prices behind it for this site to tell you what the error costs.
The two errors do not cost the same
Choosing replace where the carrier wanted repair produces an argument about one line, and the line gets adjusted. Choosing repair where the carrier expected replacement produces no argument at all, because there is nothing to argue with: a repair line carries labor and no part. The part is simply absent from the estimate, and it stays absent until somebody notices, which on a closed file is nobody. That asymmetry is why the reasoning below leans the way it does.
Bolt-on or structural, before anything else
The shape of this corpus is blunt. Components that bolt on are replaced and welded or structural panels are argued. That single distinction sorts the decision list faster than any rule about damage severity, and it is worth applying before looking at the dent at all.
- Bolt-on and behind the skin: absorbers, impact bars, liners, radiator support. Replacement is the corpus norm and a repair line here is a line you will lose
- Bolt-on and cosmetic: grilles, emblems, mouldings, lamps. Replacement, and the part number is the whole line
- Welded or structural: rockers, quarter panels, roofs. A real judgement call, and the one worth spending documentation on
- Hinged and bolted panels: doors, fenders, hoods. Genuinely split, and decided by how the damage is described
What the wrong call costs, by panel
A repair line that should have been a replacement removes a part, so the cost of the error is what that part settles for. These are the medians actually paid by carriers in this corpus, next to how often adjusters replaced the panel. Both columns carry their own sample size and the samples are not the same size, because a panel is counted every time somebody decided about it and priced only when a part was bought.
| Panel | Replaced | Decisions | Median paid | Settled lines |
|---|---|---|---|---|
| moulding/trim | 93% | 104 | $105 | 88 |
| bumper cover | 72% | 118 | $640 | 80 |
| emblem/nameplate | 100% | 59 | $61 | 50 |
| door shell | 63% | 93 | $226 | 49 |
| fender | 72% | 61 | $193 | 44 |
| radiator/condenser | 100% | 55 | $524 | 42 |
| grille | 96% | 55 | $222 | 40 |
| headlamp | 89% | 37 | $933 | 37 |
| taillamp | 82% | 45 | $627 | 31 |
| wheel | 94% | 49 | $540 | 29 |
| hood | 81% | 37 | $1,010 | 28 |
| windshield/glass | 66% | 38 | $194 | 27 |
| bumper impact bar | 96% | 25 | $376 | 26 |
| bumper absorber | 100% | 20 | $128 | 24 |
| fender liner | 100% | 24 | $99 | 23 |
| mirror | 94% | 17 | $185 | 14 |
| quarter panel | 58% | 31 | $865 | 13 |
Where this corpus will not price the error
The honest gap in the table above is that the panels where the call is hardest are the panels where the price data is thinnest. deck lid/liftgate (10), roof (6), rocker (4) all fall below the 12-line threshold and are excluded. The rocker and the quarter panel are the two most contested decisions in the whole corpus and they are also two of the thinnest price samples in it, so this site cannot tell you what a wrong rocker call costs. It can tell you the call was split 47% to 53% across 15 decisions, and that is where the useful information stops.
Two thresholds, not one
When the estimator flags a decision that disagrees with the corpus, it uses two thresholds rather than one, and the reasoning transfers to how a person should read the table. A single threshold gets this wrong in both directions. Set it high and the bumper cover goes unremarked, which is the exact call that removes several hundred dollars of part from a front-end estimate. Set it low enough to catch that and it starts firing on doors and fenders, which are close enough to even that the warning is noise. A warning people learn to ignore is worse than no warning.
- Near-unanimous in the corpus: treat a contrary call as something to justify on the line
- A clear majority the other way: worth a second look, and nothing more
- Genuinely split: the corpus has no opinion and should not pretend to one
- Fewer than 12 observations: no flag at all, in either direction
Write the call, do not just make it
A decision that is only in the estimator's head is indistinguishable in review from a decision nobody made. The lines that survive are the ones that carry their own reason, and the reason takes one clause.
- Name the failure, not the panel: "crease across the swage line at the rear edge" rather than "fender damaged"
- Name what the damage reaches: the panel next to it, the mounting points, the gap
- Put the part number on the replace line. A description and a price is a claim; a part number is a fact
- Photograph the mounting face and the fastener holes before anything comes off
- Write the line you expect to lose. A line removed in review was never going to be paid; a line never written cannot be
When the corpus and the car disagree
The corpus is aggregate and the estimator is looking at photographs of one specific car. A contradiction between them is a reason to look again, never a reason to overwrite. On the 2017 Porsche 718 Boxster S published on this site, Progressive allowed $9,002.19 against the shop's $13,922.94, and the gap was mostly labor rate rather than panel decisions. Rate and decision are separate arguments and mixing them loses both.
How this was measured
- Replace rates come from 912 lines in carrier estimates whose operation was R&R or repair, matched to a panel by description. Panels seen fewer than 6 times are not published.
- Median prices come from the settled parts index: carrier-paid part lines with the VIN and claim number stripped before the file was written. A panel is priced only when at least 12 settled lines match it.
- The two samples are independent and are printed separately. A panel can be decided about many times and bought rarely, and blending the two counts into one would hide that.
- Prices are medians of what carriers allowed, not catalogue prices, not listings, and not what the shop billed.
What this page cannot tell you
Published for the same reason the sample sizes are. A guide that never names its own limits is asking to be trusted rather than checked.
- 3 panels have too few settled lines to price at all, and they include the rocker and the quarter panel. Those are the decisions this page most wants to price and the ones it cannot.
- Median prices mix model years, makes and trim levels. A median bumper cover across this corpus is not the price of any specific bumper cover, and on a corpus this weighted toward European and exotic vehicles it will read high against a volume domestic lot.
- The corpus records what adjusters decided, not what was correct. Where a shop believes the industry is systematically wrong on a panel, this table is evidence of the practice and not of the merits.
- Nothing here measures whether writing a decision more carefully changes the outcome. That the well-documented line survives review is the working assumption of every estimator asked, and it is not something this corpus has tested.
Before you ask
How is this page different from the repair-or-replace table?
The measured page publishes the data: every panel, every rate, every sample size. The product page explains what the estimator does with it. This page is about the decision itself, how to reason toward it and how to write it so it holds up, and it links to both rather than repeating either.
Should an estimator default to replace?
No. 83% of decisions in this corpus went to replacement, but that average is carried by bolt-on components that were never repairable. On the panels that actually split, defaulting either way is guessing. The point of the split is that the documentation decides, so the effort belongs in the documentation.
What if the carrier writes repair and the shop cannot repair it?
That is a supplement, and it is the honest kind. Document the failure at the point it becomes visible: the panel on the bench, the split in the metal, the mounting face that will not pull. A supplement backed by a photograph of the thing that could not be done is a different document from one that asserts the first call was wrong.
Does the estimator override a repair-or-replace decision?
No. It reports the disagreement against the measured rate and the sample size behind it, and records that on the line. The model is looking at this car and the corpus is not, so a contradiction is information rather than a verdict.
Check the evidence
Nothing on this page is worth more than what it rests on. These are the pages the figures came from.
The data
Repair or replace, measured
The full table: every panel, every rate, every sample size.
A real repair
2017 Porsche 718 Boxster S, Progressive
Rate and panel decisions pulled apart on one front-end repair.
Written for
For collision centers
Making the same call the same way across a floor of estimators.
Related
Repair or replace in the estimator
What the product does with these rates when it writes a line.
Run it on damage you have already written
Accounts are created by hand, usually within a business day. Bring your own photographs and a repair order you have already closed, so you can see how close it lands before you decide. No card, no trial timer.