The F&I Technology Stack: Menu, Reporting and What AI Actually Does

01The stack, layer by layer

"F&I technology" is not one product. It is a chain of systems, usually from several vendors, that a deal passes through from a shopper's first click to a claim paid years later. The companies named below are examples drawn from their own product pages and press releases. The list is not exhaustive, inclusion is not endorsement, and the order means nothing.

The layers have different owners with different interests. A menu may belong to a DMS company, a lender consortium, a data company or a software firm; the rating network to a joint venture of DMS companies. The practical point is that your administrator does not control most of the stack your deals move through, and neither do you.

Ownership changes. Under the acquisition PCMI announced in November 2025, StoneEagle's administration software went to PCMI, while its reporting, analytics, menu and service-drive products stayed with StoneEagle's retail unit. Check who owns a vendor today, not who owned it when you signed.

The administrator's own layers are the ones you can diligence directly — the administration platform, claims system, reports and portals. That is where the rest of this guide concentrates.

02Menu selling: what it is for, and what regulators have said

A menu is a structured way of presenting every available protection product to every customer, with a price and a payment for each option and a record of what the customer chose. It does three jobs at once, and vendors tend to market only the third.

• Disclosure. The customer sees what each product costs and what the payment is with and without it.
• Consistency. Every customer is offered the same products on the same terms, which gives a dealer a record to point to if asked whether some customers were quoted differently from others.
• Penetration. A clear presentation tends to sell more product. This is the benefit most menu marketing leads with.

The vendors say this themselves. Reynolds and Reynolds describes docuPAD as helping "ensure every consumer is presented every product and disclosure with transparency." Dealertrack states that the Darwin menu "helps ensure consistency in menu presentations across all your customers." Those are vendor claims we have not tested, but they show the industry treats the menu as a compliance control, not only a sales tool.

What regulators have said

The add-on enforcement record is covered in Part 02 of this series. In short: the FTC's actions against Napleton (April 2022) and Leader Automotive Group (December 2024) both centered on allegations that add-on products were charged without authorization or represented as required, and the Napleton complaint alleged that Black customers were charged more for comparable add-ons. The FTC's CARS Rule, which would have imposed express-consent requirements for add-ons nationally, was vacated by the Fifth Circuit in January 2025 on procedural grounds.

Those cases are about conduct, not software. A menu on its own settles nothing, but a menu used on every deal, with the record kept, is the record a dealer can point to when someone later asks what a customer was shown.

Two California provisions are worth knowing about for dealers selling there:

• Cal. Civ. Code § 2982.2 requires, before a conditional sale contract (a dealer-financed deal) is signed, a separate, signed disclosure listing the price of each optional item — including service contracts, insurance, GAP or debt cancellation agreements, theft deterrent devices and surface protection — with the installment payment shown both excluding and including them.
• The California CARS Act (SB 766). According to published legal summaries (October 2025), the act was signed 6 October 2025, becomes operative 1 October 2026, and adds new requirements on the sale and disclosure of add-on products and on certain used-vehicle sales. We could not access the enrolled text directly and do not describe its provisions here; the details are in the chaptered statute.

What to check in a menu

03Reporting: what you should be able to see, and how fast

Administrators commonly say they offer reporting. The useful questions are narrower: which numbers, defined how, available how quickly, and whether you can take the data with you.

Loss ratio: know which one you are looking at

A loss ratio is claims divided by premium, but both halves have more than one definition. Paid claims over written premium is the easiest to produce and, early in a program, the most flattering, because premium arrives up front and claims arrive over years. A ratio built on earned premium (the portion of the premium attributable to coverage already provided) and incurred losses (claims paid plus the change in reserves for claims expected but not yet paid) is harder to produce and closer to the economic truth. Neither is wrong, but two reports using different definitions cannot be compared, and a report that does not say which it uses has not yet told you anything.

Latency

Ask for freshness in hours or days, not adjectives. Contract registration and claim status can reasonably be near-current; cession statements legitimately lag, because they depend on month-end accounting and reserves. As a published example of a stated latency, RouteOne, as reported by Auto Remarketing, describes its Insights Premium product for finance sources as offering next-day data availability. Part 01, section 07 sets out the questions that turn "real-time" into something testable; they apply to reporting unchanged.

A dashboard is not data

A dashboard shows you what its designer chose to show. Data you can export, at the contract and claim level, lets you check the dashboard, combine it with your own DMS data, and take it with you if you change administrators.

04AI in F&I and claims: what is actually documented

Many vendors in this category now describe something as AI-powered. We looked for what has been announced as shipped, by whom, doing what, and separated it from pilots, roadmaps and commentary. The table reflects what the companies published. We have not tested these products; a vendor's description is a claim, not a finding.

What we looked for and did not find

Absence of an announcement is not evidence of absence, but as of September 2026 we could not document:

• Automated claim denials by a learned model. We found no published announcement from a vehicle service contract administrator describing an AI system that denies claims without a human decision. The administrator deployment above is described as decision support. A rules engine, which applies logic the administrator configures, is a different thing. Whether a person approves each denial is a question to ask of any system, AI or not (section 06, question 2).
• Repair-order document extraction specific to service contract claims. The technique is widely marketed across insurance generally. We did not find a named VSC administrator publicly describing it as deployed.
• AI voice agents on the claims line. The documented voice deployments we found are on the dealership service drive — appointment booking and call review — not administrator claims centers.
• AI in VSC pricing. An August 2026 white paper published on WarrantyNews argues providers should use AI for rate adequacy and fraud; it is prescriptive and names no deployments. Some administrators publish data-science leadership roles (see the TPA executive profile), which is evidence of intent, not of a product.

Where it plausibly works

The documented AI uses share a pattern. They sit where a person is overloaded with documents or calls, and the machine assembles, compares and flags — the tedious, rules-shaped parts of a claim or deal jacket. They stop short of the final adverse decision, which is where error costs most and where much of the rulemaking described in the next section is focused.

Performance figures are common in this category's marketing — first-pass rates, accuracy percentages, handling-time reductions, hours saved. We did not verify any and reproduce none. Without knowing what was counted, on what sample, over what period, no such figure can be compared with another. That is the question to ask about every such number, from any vendor or administrator, including us.

05The regulatory frame for AI claims and AI decisions

Two different legal questions get blurred together. The first is about marketing: is what a company says about its AI true and substantiated? The second is about decisions: when a model contributes to a decision about a consumer — credit, coverage, a claim — what does the law say? Both come up in F&I technology. This is not legal advice, and several of these rules changed within the last year.

Claims about AI: the FTC

In September 2024 the FTC announced "Operation AI Comply," five actions against companies alleged to have used AI claims to deceive consumers. Then-Chair Lina Khan's statement was direct: "there is no AI exemption from the laws on the books." The Rytr consent order, which came out of one of those five actions, was reopened and set aside by the Commission in December 2025. The Commission concluded that the complaint did not meet the FTC Act's legal requirements and that the order unduly burdened AI innovation, citing the administration's AI Action Plan — while stating that the FTC "will continue to hold accountable actors that use AI to violate the law or deceive consumers about the capabilities of their generative AI."

Two cases about accuracy claims are the most relevant to a buyer of F&I technology:

• IntelliVision (December 2024): the FTC alleged the company claimed its facial recognition software was free of gender and racial bias and trained on millions of faces, without evidence for the first claim and having trained on roughly 100,000 individuals.
• Workado (April 2025; final order August 2025): the FTC alleged the company advertised its AI content detector as 98 percent accurate when independent testing on general-purpose content showed 53 percent. The final order requires "competent and reliable evidence" to support effectiveness claims at the time they are made.

The FTC's long-standing Policy Statement Regarding Advertising Substantiation states that objective advertising claims need a reasonable basis before they are made. For a buyer, the practical question is what basis a vendor has for a stated accuracy rate for a claims model, or a stated increase in penetration from a menu engine — and the same question can be put to an administrator about its own systems, including us.

Decisions about credit: ECOA and Regulation B

When a model contributes to a credit decision at the dealership, the Equal Credit Opportunity Act and Regulation B require that the reasons given for adverse action be specific. The regulation is explicit: the statement "must be specific and indicate the principal reason(s) for the adverse action," and a statement that the applicant failed to achieve a qualifying score is not sufficient (12 C.F.R. § 1002.9(b)(2)).

The CFPB had issued two circulars applying that rule to complex algorithms — Circular 2022-03 and Circular 2023-03. Both were among 67 guidance documents the CFPB withdrew on 12 May 2025. The withdrawal did not amend the text of the statute or of Regulation B, including § 1002.9(b)(2). The same notice said the Bureau would deprioritize enforcement against conduct that does not conform to withdrawn guidance; that is a statement of the Bureau's enforcement priorities, not a change in the law.

F&I sits beside the credit decision, often in the same software. When a vendor describes "AI decisioning," ask which decisions it means — credit, product presentation, or something else — and ask your counsel how the rules above are addressed for each.

Decisions about consumers generally: state AI laws

• Colorado. According to published legal analyses (May 2026), Colorado enacted SB 26-189 on 14 May 2026, replacing its 2024 AI Act (SB 24-205), with an effective date of 1 January 2027. Those analyses describe the new law as addressing automated decision-making technology used in consequential decisions, including in financial or lending services and insurance. We could not read the enacted text and do not describe its provisions here.
• Texas. According to a published legal summary (July 2025), the Texas Responsible Artificial Intelligence Governance Act (HB 149) took effect 1 January 2026 and places certain restrictions on how AI is developed and deployed. We did not read the enacted text and do not describe its provisions here.
• Federal preemption efforts. Executive Order 14365, "Ensuring a National Policy Framework for Artificial Intelligence" (11 December 2025), directed the Attorney General to establish an AI Litigation Task Force to challenge state AI laws inconsistent with the order's stated policy, directed the Commerce Department to identify such laws, and tied certain federal broadband funding to them. The order does not by its own terms repeal any state law; check the current status of any state AI law with counsel.

Decisions about insured claims: the NAIC Model Bulletin

In December 2023 the NAIC adopted a Model Bulletin on the Use of Artificial Intelligence Systems by Insurers. The NAIC's own implementation map, status as of 1 April 2026, lists, by our count, 24 states and the District of Columbia as having adopted it, and California, Colorado, New York and Texas as having separate insurance-specific AI guidance. As adopted in New Jersey in February 2025, it applies across the insurance life cycle, including "claim administration and payment, and fraud detection"; asks insurers to weigh "the extent to which humans are involved in the final decision-making process"; and expects due diligence on third-party AI systems, including contract terms that provide audit rights.

The bulletin is addressed to insurers, and whether a vehicle service contract is regulated as insurance varies by state. Insurers do take part in this business — issuing contractual liability policies, writing GAP and mechanical breakdown insurance, and reinsuring dealer programs. Where a program is backed by an insurer, it is reasonable to ask your provider whether and how the bulletin is addressed in your state.

Voice agents: the TCPA

In a declaratory ruling released 8 February 2024 (FCC 24-17), the FCC confirmed that the Telephone Consumer Protection Act's restrictions on "artificial or prerecorded voice" calls apply to AI technologies that generate human voices, so such calls require the called party's prior express consent. Inbound and outbound use raise different questions. If an AI voice agent is used, or planned, for outbound renewals, claims follow-up or marketing, a question to ask is how prior express consent is obtained and recorded.

06How to tell real AI from marketing

"AI-powered" is not a specification. It can describe a large language model summarizing a phone call, a statistical model trained on a million deals, a rules engine that has existed for a decade, or a feature on a roadmap. None is inherently better. The problem is not knowing which one you are buying.

Start with the first question below. A system that recommends to a trained person who decides raises different questions — practical and, as section 05 shows, regulatory — from one that decides, even when both are described in the same words.

07How CareGard approaches this

Disclosure first: this is the section where we describe ourselves, and everything in it is company-reported.

CareGard administers its programs on TRONIX, its program administration platform. As CareGard describes it, TRONIX provides dealer and agent portals, configurable reporting down to the VIN level, OEM integrations, and the program administration functions this guide has discussed — rate quoting, contract issuance, cancellation management and reinsurance accounting.

This guide does not describe any artificial intelligence capability at CareGard, and nothing in it should be read as a claim that we have one. We would expect the questions in sections 03 and 06 to be put to us, and would rather be asked them than not.

The questions in this guide apply to CareGard as much as to anyone.

08How we built this, and what we could not verify

Every vendor description here comes from the company's own product page, press release or filing, or trade coverage of that release, and is attributed. Ownership statements were checked against acquisition announcements. Regulatory statements come from the FTC, the Federal Register, the C.F.R., the NAIC, a state insurance bulletin, the FCC and the White House; where we could not reach a primary source, we relied on published legal summaries, said so, and kept to high-level, dated points. We tested no product and verified no vendor's performance figures.

What we could not verify, specifically:

• MenuVantage's acquisition history. CDK Global markets the product today as CDK MenuVantage Platinum. We did not locate the announcement of how it came to CDK, and state no date.
• The current ownership of every vendor. Acquisitions are stated as of their announcement dates. We did not verify the current ownership of J.D. Power or of StoneEagle's retail unit.
• Colorado SB 26-189, California SB 766 and Texas HB 149 bill text. We could not read the enacted text, so we give only dates and a high-level description attributed to published legal summaries, and do not describe specific provisions.
• Whether particular vendor AI features are generally available. Where a release said "coming," or did not distinguish shipped from planned, we said so.
• Industry-wide adoption. We found no reliable data on how many administrators use AI in claims, or how many dealers use menu recommendation engines. We have not estimated.

This guide ranks no vendor and no administrator, including CareGard. It is not legal advice. The AI rules described here are changing quickly — several changed between December 2025 and May 2026 — and should be checked with counsel before you rely on them. We intend to review this guide annually. If something here is wrong, we would like to know.