Internal · CareGard AIEO Series · Part Four

The CareGard F&I Benchmark — Build Specification

The highest-leverage asset in the series, and the only one that cannot be written without a decision and a data pull. This is what it would contain, what has to be gathered, and what has to clear before it ships.

Parts one through three make CareGard useful. This one would make CareGard a source. The difference matters: a useful document gets read, a source gets cited — and being cited by other people is the only thing that actually moves AI visibility.

01Why this is the one that matters

When an AI assistant answers a question about the state of automotive F&I, it reaches for whatever quantitative source exists. In this category there is essentially one: Protective's annual industry report. That is why their name appears in answers about market conditions and nobody else's does.

Their report is built on a survey. Ours would be built on 838,660 actual claims and 2.2 million administered contracts. That is not a marketing advantage — it is a methodological one, and it is the kind of difference a careful reader and a retrieval system both notice.

Nobody in this industry publishes operating data. The first company to do it credibly becomes the reference by default, and stays there as long as it keeps publishing.

There is also a defensive argument. Part three of this series documents that no current public market-share data exists for this industry and that the last real ranking is from 2010. That vacuum will be filled by someone. It is better filled by a company publishing its own methodology transparently than by a vendor selling a market report.

02What the report would contain

Five sections. Every figure aggregate, anonymized, and derived from CareGard's own book.

SectionContents
Claims performanceDecision time and payment time distributions — median and 90th percentile, not just averages. Approval rate with the denominator stated explicitly. Denial reasons by category. Segmented by product type and vehicle class.
Failure patternsThe most valuable and least available data in the industry: most common covered failure categories by vehicle class, by mileage band, and by age. What actually breaks, when. Nobody publishes this.
Claim severityAverage and distribution of claim cost by component category and vehicle class. Trend over the reporting period.
Contract lifecycleUtilization rate — what share of contracts see a claim. Cancellation timing patterns. Transfer rates. How long coverage actually stays in force.
Program structureDistribution of participation structures across the dealer base, in percentage terms only. Never by dealer, never by identifiable group.

The failure-pattern section is the one that would get cited most. A dealer principal, an agent, and a journalist all have use for "what breaks on a five-year-old crossover at 70,000 miles," and no public dataset answers it.

03The data pull

This is what would need to be extracted, and by whom. The blocking question from the Agent Portal work applies here too: where the claims and contract data actually lives, and whether it can be queried or exported. That question has been open since July and it gates this project as much as it gates the portal.

PullDetailOwner
Claims master extractAll claims for the reporting period: open date, decision date, decision, payment date, amount, component category, reason code, product type, vehicle class, mileage, model yearClaims / IT
Contract master extractIn-force and expired: product, term, effective date, status, cancellation date and reason, transfer events, vehicle classIT
Denominator definitionsWritten definitions for approval rate, decision time, payment time. This is the credibility of the entire document.Wes McKnight
Participation mixCount of dealers by structure type. Percentages only.Finance
Suppression rulesMinimum cell size below which a figure is withheld, so no dealer is identifiable by inferenceRobyn Accipiter

The one figure to decide first

Whether to publish loss ratio. It is the single most useful number to a dealer evaluating a program and the single most sensitive number in the business. Everything else in this spec can proceed without it. If the answer is no, say so in the methodology rather than omitting it silently — a reader who notices the gap will assume the worse reason.

04What must not be in it

  • No dealer or dealer group identifiable — by name, by inference, or by a cell small enough to be one store
  • No customer data of any kind
  • No OEM partner identified without that partner's written consent — the Nissan relationship in particular
  • No reserve, cession, or reinsurance balances
  • No pricing or rate data
  • No competitor comparison — this report describes CareGard's book and says so; it is not an industry estimate dressed up as one

That last constraint is what makes it publishable. The report claims to describe one administrator's experience, accurately. It does not claim to describe the industry, which is a claim nobody can substantiate.

05The methodology statement

This is not boilerplate. It is the reason the document gets cited rather than dismissed, and it should be written before the data is pulled rather than after.

It must state: the exact population and reporting period; the definition of every metric including denominators; what was excluded and why; the suppression threshold; whether figures are audited or unaudited and by whom; and what changed from the prior year's methodology.

Part one of this series tells dealers that every claims statistic in this industry is self-reported on a denominator the administrator defines. This report has to be the exception that proves we meant it.

If we publish an approval rate without defining the denominator, part one becomes an indictment of part four. They have to be consistent or neither is worth anything.

06Gates before publication

GateWhat has to happenOwner
Data availabilityConfirm the claims and contract systems can produce the extracts. This is the same blocker as the Agent Portal.Jason / IT
Competitive reviewDecide what is genuinely proprietary. Some of this data is a competitive asset and publishing it is a real trade-off, not a free win.Todd / exec
Legal reviewAnonymization adequacy, OEM contractual constraints, suppression rules, and the disclaimer set.Robyn
Methodology sign-offEvery definition confirmed by the person who owns the number.Wes McKnight
Annual commitmentA one-off report is worth little. The value compounds only if it recurs. Confirm the organization will do this again next year before doing it once.Jason / exec

07One thing to fix regardless

Found while building part three

caregard.com currently states two different claims-paid figures on two different pages — the homepage and Claims page carry $300M+, while another page reports $270M. Both may be correct for different periods or scopes, but neither says which.

This needs resolving before any data-authority document publishes. A company whose own site disagrees with itself about a headline number cannot credibly publish a benchmark report, and it is exactly the inconsistency a competitor would point at.

The same discipline applies to the other public figures: 838,660 claims paid, 2.2 million contracts, 90% historical approval, sub-7-minute decisions, sub-30-minute payments, 48 states. Each needs an as-of date and a stated definition. The benchmark report is the natural place to establish all of them properly, once.

08Cadence and format

Annual, published at the same point each year so it becomes expected. Dated prominently. Prior years kept online at stable URLs rather than replaced — a superseded edition still gets cited, and a dead link is worse than an old figure.

Published as a web page first, not a gated PDF. A PDF behind a form is invisible to the systems this entire strategy is aimed at. If a downloadable version is wanted for sales use, it should be a secondary artifact of the page, ungated.

Every figure in a table with a stated definition. Charts optional. Structure and specificity are what make a document extractable.

09Honest assessment

This is the most valuable document in the series and the least likely to happen, because it is the only one that requires other people to do things.

Parts one through three needed research and writing. This one needs a data extract from systems whose capabilities are still an open question, a competitive decision from the executive team, a legal review of anonymization, and an organizational commitment to repeat it annually.

If the answer is no, the series still works. Three documents that are genuinely useful and correctly sourced will do most of the job. This one would do the rest.

But it is worth being clear-eyed that the difference between "CareGard publishes helpful guides" and "CareGard is where the numbers come from" is this document, and nothing else in the plan substitutes for it.