# RESPA Section 8 and AI-Driven Mortgage Referrals: The Anti-Kickback Architecture for Lead Routing, MSAs, and Co-Marketing

*June 26, 2026 · 13 min read · Pranay Shetty*

> Lead-routing scores, AI co-marketing tools, and pay-per-application platforms have rebuilt the mortgage referral economy in vocabulary that maps very cleanly onto the RESPA Section 8 'thing of value' standard. The posture we put in front of marketing and partnerships so an AI workflow does not produce a 12 CFR 1024.14 finding the next exam catches.

## The Old Rule on a New Channel

The Real Estate Settlement Procedures Act's Section 8 anti-kickback prohibition was written for an industry where the referral was a phone call from a real-estate agent to a loan officer, the thing of value was a steakhouse dinner, and the agreement was a handshake. The substantive rule did not change when the channel turned digital, and it has not changed now that the channel is AI. The CFPB's 2015 [Compliance Bulletin 2015-05](https://files.consumerfinance.gov/f/201510_cfpb_compliance-bulletin-2015-05-respa-compliance-and-marketing-services-agreements.pdf) on marketing services agreements is the most recent definitive read, and the agency has continued to bring enforcement actions on the same theory: a payment that compensates a referral, however structured, is a Section 8 violation regardless of the label on the contract.

The lender that buys a "leads platform" that scores and routes consumer inquiries to lenders based on a marketing-fee schedule has bought a referral mechanism. The lender that signs an "AI co-marketing" agreement with a real-estate brokerage where the brokerage's agent assistant recommends one preferred lender to a borrower has signed a referral arrangement. The lender that pays a per-application fee to a comparison shopping platform whose ranking algorithm is opaque is paying for placement that the CFPB will treat as referral compensation if the placement is selective. The Section 8 question is the same one it was in 1974. The architecture has to encode the answer.

We work with mortgage lenders, banks, and non-bank originators that route their lead flow through some combination of in-house AI and third-party platforms. The framework below is the one we hold our customers to before any AI-driven referral, scoring, or co-marketing arrangement turns on, because the cost of a Section 8 finding is high and the operational fix usually requires changing a vendor contract that takes months to renegotiate.

## What the Statute Reaches

[12 U.S.C. 2607(a)](https://www.law.cornell.edu/uscode/text/12/2607) and the implementing regulation at [12 CFR 1024.14(b)](https://www.consumerfinance.gov/rules-policy/regulations/1024/14/) prohibit a person from giving or accepting any fee, kickback, or thing of value pursuant to any agreement or understanding that business incident to or part of a real estate settlement service involving a federally related mortgage loan will be referred to any person. The rule reaches every settlement service the loan touches, which on a typical purchase or refinance means the lender, the title company, the escrow agent, the appraiser, the home warranty provider, the credit-report seller, the flood-certification provider, and several others.

"Thing of value" is broader than money. The CFPB's commentary and the case law read it to include payments above fair market value for goods or services, discounted goods or services, gifts, entertainment, leads, anything that has economic value to the recipient. The "agreement or understanding" element does not require a written contract. A pattern of payment plus a pattern of referral can establish the understanding by inference, which is the posture the agency takes in enforcement.

The exemption institutions misread most often is Section 8(c)(2), which permits payment for goods or services actually furnished or for services actually performed, provided the payment bears a reasonable relationship to the market value of the goods or services. The exemption is real, and the bulletin and the rule both confirm it. The exemption does not cover payments that exceed market value for the services. It does not cover payments where the services are not actually performed. It does not cover payments where the services are nominal cover for what is effectively a referral fee. The institutions that get caught are the ones whose contract documents one set of services and whose invoices and emails reflect another.

## The AI Lead-Routing Pattern That Most Looks Like a Referral

The fastest-growing model in mortgage lead distribution is a platform that collects consumer inquiries, scores them with a proprietary model, and routes the inquiry to one of N enrolled lenders. The platform charges the lenders on some combination of per-lead, per-application, per-funded-loan, and subscription bases. The question we ask before any of our customers buys into this kind of platform is whether the routing is editorial or mechanical, and whether the platform's economic incentive is to steer to the highest-paying lender or to the lender best matched to the consumer's needs.

A platform whose routing model is documented to optimize for consumer match, whose payment structure is uniform across enrolled lenders for similar lead profiles, and whose disclosures to the consumer name every enrolled lender and explain the routing logic is a platform that has built a defense to Section 8 exposure. A platform whose routing model the lender does not see, whose payment varies by lender on a basis the platform cannot tie to market-rate compensation for services, and whose disclosures present the routed lender as a recommendation rather than a routed result is a platform whose first CFPB inquiry will be unpleasant for both the platform and every lender on it.

The CFPB's [2024 advisory opinion on digital comparison shopping platforms](https://www.consumerfinance.gov/about-us/newsroom/cfpb-issues-guidance-to-protect-mortgage-borrowers-from-pay-to-play-digital-comparison-shopping-platforms/) and the Bureau's broader posture on what it called "pay-to-play" arrangements made the line explicit. A platform that presents lender rankings or recommendations in a way that is influenced by lender payments and that does not disclose that influence in a way the consumer can act on is, in the Bureau's view, deceiving the consumer about the nature of the recommendation and steering business to lenders based on payment rather than fit. The Section 8 exposure is the kickback theory. The UDAAP exposure is the deception theory. The two theories run in parallel and a lender exposed on one is usually exposed on both.

The architecture we put around lead intake on the lender side is a vendor-due-diligence file that documents the platform's routing logic in enough detail for the lender's compliance team to assess whether the routing is editorial, the platform's disclosure to the consumer with the exact rendered text, the lender's payment terms with a fair-market-value attestation supported by an independent benchmark, and the lender's ongoing monitoring of the lead-quality and conversion data the platform produces. A lender whose file shows all four is a lender whose Section 8 posture is defensible. A lender whose file is just the platform's master service agreement is a lender whose first CFPB question will be "what did you know about how this platform routes."

## Marketing Services Agreements With an AI Component

MSAs were a stable structure for a decade and then the CFPB pushed back hard in the mid-2010s, brought a series of consent orders, and issued [Bulletin 2015-05](https://files.consumerfinance.gov/f/201510_cfpb_compliance-bulletin-2015-05-respa-compliance-and-marketing-services-agreements.pdf). The Bureau did not declare MSAs per se illegal. It declared most of them to be high-risk arrangements where the actual conduct rarely matched the contract. A lender that pays a real-estate brokerage a monthly fee for marketing services is, the Bureau said, often paying for referrals dressed up as marketing, and the Section 8 exposure is real.

The AI-era MSA we see most often takes the form of a brokerage adopting an AI agent assistant that integrates with a preferred lender's systems and recommends that lender to borrowers the agent is helping. The lender pays the brokerage a fee labeled as a "data services" payment, a "co-marketing technology" payment, or a "platform integration" payment. The conduct is that the AI agent steers consumers toward one lender. The label on the payment does not change what the conduct is. The first question the Bureau will ask in a sweep is what the AI agent says to consumers, and the first thing the Bureau will look at is whether the AI's recommendation logic correlates with the payment structure.

A defensible MSA in the AI era requires four things. The services have to be defined narrowly and concretely with deliverables a vendor could perform if the brokerage did not have any consumer relationship. The payment has to bear a written fair-market-value relationship to the services with an independent benchmark, and the lender has to actually receive what it paid for. The AI agent's recommendations have to be neutral or, if preferential, the preference has to be disclosed to the consumer in language the consumer reads at the point of recommendation, not buried in a privacy policy. The arrangement has to be reviewed annually with the conduct compared against the contract.

We have seen MSAs where the contract said "social media training and lead generation" and the actual deliverable was that the brokerage's AI agent inserted the lender's name into every borrower conversation. That arrangement is a referral fee with a marketing-services label, and the documentation gap is what the Bureau will pull on first.

## Co-Marketing Through Platforms the Lender Does Not Control

A separate pattern is the lender that participates in a multi-lender platform where real-estate agents subscribe and consumers interact with an AI assistant that helps them through the buying process and recommends a lender at the appropriate moment. The lender does not contract directly with the brokerage. The lender contracts with the platform. The platform contracts with the brokerage. The economic flow is from lender to platform to brokerage.

The legal question is whether the lender's payment to the platform compensates referrals from the brokerage. If the platform pays the brokerage out of the lender's fees and the brokerage's AI agent preferentially recommends the lender, the chain of payment and the chain of referrals are aligned, and the Bureau will read the structure as the lender paying the brokerage through the platform. The platform-as-intermediary structure does not insulate any party in the chain from Section 8 exposure. The intermediation makes the conduct harder to see and easier to litigate later, not legal.

The diligence we run before a customer joins a platform like this asks four questions. What is the platform paying the brokerage and on what basis. What does the AI agent say to the consumer at the moment of lender recommendation. How does the AI agent's recommendation correlate with the platform's lender payment data. What disclosure does the consumer see at the moment of recommendation about the financial relationships in the chain. A platform that will not answer these questions for a lender doing diligence is a platform the lender should not be on, because the answers will eventually come out under subpoena and the lender that did not ask will be in a worse position than the lender that asked and got incomplete answers.

## The AI Agent's Own Recommendation Architecture

We build voice and text agents that sit on the lender's own customer surface. The Section 8 question for an agent the lender controls is different from the question for an agent at a brokerage or a platform. The lender's own agent recommending the lender's own products is not a Section 8 issue, because there is no referral and no third party being compensated. The Section 8 question arises when the lender's agent recommends a third-party service provider for any of the settlement services the loan involves.

The agent that recommends a specific title company, a specific home-warranty provider, a specific home-insurance carrier, or a specific real-estate agent has to do so on a basis the rule's exemption frameworks support. The two structures that work are affiliated business arrangements under [Section 8(c)(4) and 12 CFR 1024.15](https://www.consumerfinance.gov/rules-policy/regulations/1024/15/), which require a written disclosure to the consumer at or before the referral, a written estimate of the charge, and a statement of the consumer's right to shop, and a true comparison-shopping mechanism that presents the consumer with options and lets the consumer choose without the lender steering.

We instrument the agent to render the affiliated business disclosure at the moment any affiliated provider is suggested, to capture the consumer's acknowledgment, and to keep the disclosure rendering in the per-call audit file. The agent never recommends a single third-party provider without that documentation, because a single recommendation without disclosure is functionally indistinguishable from a referral arrangement and the Bureau will not give the agent credit for being well intentioned. The agent will say the institution has affiliated providers, will name them, will disclose the relationship, and will tell the consumer they can shop. That is the rendered language. The institution's marketing team does not get to rewrite it without a fair-housing-and-Section-8 review.

The harder version of the question arises when the agent is asked to recommend an unaffiliated provider. A lender whose agent suggests a specific title company without affiliation and without a comparison-shopping mechanism is steering business to that title company on a basis the rule wants documented. The basis the rule accepts is the consumer's explicit request, the title company's market position, or the institution's pre-published list with neutral selection criteria. A basis the rule does not accept is that the title company has paid the lender or its affiliate anything beyond fair-market-value compensation for services actually performed.

## The Sales Funnel and the Outbound Lead

A particularly exposed pattern is the lender that pays a third-party outbound contact center, a real-estate technology vendor, or a CRM platform that delivers outbound contacts to consumers and converts a percentage of them into mortgage inquiries. The fee structure that prices the engagement on a per-funded-loan basis is the structure most clearly tied to referral compensation. The fee structure that prices on a per-contact-attempted basis with no relationship to outcomes is the structure cleanest under Section 8, because the lender is paying for marketing labor performed regardless of whether any consumer applies.

The CFPB has not declared per-funded-loan compensation per se illegal, but the Bureau's posture is that a payment structure tied to closed loans is, in the absence of clear documentation that the payment is for services performed at market value, presumptively referral compensation. We hold our customers to the more conservative posture. The economic value of the marketing labor is what the lender should be buying, not the conversion outcome, because the conversion outcome is the part of the funnel where the rule's concern about steering and undisclosed compensation lives.

The agent that runs on the outbound side has to know which lead source produced each conversation, has to keep that provenance in the per-call audit file, and has to be auditable against the contract that governs the lead source. A lead source whose actual operation differs from its contract gets caught here, and the operational fix is usually to renegotiate the contract or to drop the source. The marketing team does not always like that conversation. The compliance team likes it less when the exam finds the gap.

## The Section 8(c)(2) Defense and the Fair Market Value File

When the lender does pay a third-party service provider for goods or services, the Section 8(c)(2) defense requires that the payment bear a reasonable relationship to the market value of the goods or services. The "fair market value" determination is what the lender's file has to establish, and the file has to be more than the contract. The CFPB has been clear that a contract recital that the payment reflects fair market value does not, on its own, establish that the payment in fact reflects fair market value, because parties to a Section 8 problem can be expected to write the contract in their favor.

The file we recommend the lender's compliance team build for every recurring third-party payment includes an independent benchmark for the services from a non-aligned source, the actual deliverables received with timestamps and the persons responsible for receiving them, the lender's assessment of whether the services were necessary to the lender's marketing or business operations, and a periodic update of the benchmark to confirm the payment still tracks the market. The institutions that have this file in place treat it as part of the vendor-onboarding workflow and refresh it annually. The institutions that build the file only when the CFPB asks for it are building it under conditions that do not favor a clean answer.

The work we wrote separately on [third-party risk management for AI vendors](/blog/third-party-risk-management-ai-vendors-banking-tprm-playbook) describes the vendor file in the broader TPRM context. The Section 8 file is a specialization of the TPRM file for vendors in the settlement-services value chain, and the two are coordinated in the institutions we serve.

## What Section 8 Enforcement Has Looked Like

The Bureau and HUD before it have brought RESPA Section 8 enforcement on co-marketing arrangements with realtors, marketing services agreements with brokerages and builders, lead-generation platforms with selective routing, and captive reinsurance arrangements between lenders and mortgage insurers. The penalties have included restitution to consumers, civil money penalties, and consent orders restricting future arrangements. The CFPB's [enforcement actions database](https://www.consumerfinance.gov/enforcement/actions/) lists the matters by year, and the consistent pattern is that the conduct's economic substance, not the contractual label, decides the outcome.

The institutions that have settled these cases generally agreed to disgorge the payments, to terminate the arrangements, and to implement enhanced compliance management for referral arrangements going forward. The cost of the settlement is rarely the dominant economic harm. The cost of unwinding a multi-year referral arrangement, replacing the lead source it depended on, and rebuilding the production pipeline without the contributions the arrangement was producing is the harder cost.

## How We Architect Around This for AI Workflows

The institutions we serve build an AI referral and recommendation policy that the agent technically enforces and the compliance team monitors. The policy has four pillars.

Every third-party service provider recommendation the agent makes runs through a registered referral pathway with a documented basis (affiliated-business disclosure, neutral comparison shopping, or consumer request). The agent does not invent recommendations off-policy, and the validator stops the conversation if the agent's intended next utterance is a referral the policy does not cover.

Every lead source feeding the agent's conversation queue has a vendor file documenting the payment structure, the fair-market-value benchmark, the routing logic, and the consumer disclosure rendered before the agent connects. The agent reads from the queue and the queue is the lender's central choke point for confirming lead-source compliance.

Every MSA, lead-distribution platform, and co-marketing arrangement runs through an annual conduct-versus-contract audit where the institution's compliance team samples the actual interactions the AI produced and compares them to the contract's deliverables. A drift between the two is the early indicator that the arrangement has become a referral mechanism the contract no longer describes.

The agent's audit file per call includes the recommendation made, the basis under the policy, the disclosure rendered, and the consumer's acknowledgment. The file is the artifact the Bureau will ask for in a sampled inquiry, and the institution that produces it cleanly is the institution whose inquiry does not escalate.

## The Honest Read

Section 8 is one of the older statutes the AI mortgage workflow runs into and the rule has aged remarkably well. The economic substance of a referral arrangement is what the rule cares about, and the AI channel did not change the substance. What it changed is the speed at which arrangements can be set up, the opacity of the routing logic that makes them work, and the volume of consumers exposed to any given arrangement. A Section 8 violation that would have produced a few hundred affected loans in the phone-and-fax era can produce thousands in the AI era because the same routing logic runs at scale.

The lenders we work with treat Section 8 as a structural question that gets answered at the contract layer and the recommendation-policy layer, not as a script question the AI handles in conversation. The script is the symptom. The contract and the policy are the cause. The Bureau will read both. The institution that has both right does not have a Section 8 exam finding to defend. The institution that has a clever script in front of a referral arrangement the contract does not support has the worst version of the finding to defend, because the script will read as deception added on top of the kickback.

We have written separately on the [TRID disclosure clocks](/blog/trid-loan-estimate-closing-disclosure-ai-agents-timing), the [adverse-action notice machinery](/blog/adverse-action-notices-ai-credit-decisions-ecoa-reg-b), and the [TCPA outbound consent layer](/blog/tcpa-ai-voice-mortgage-banking-2026-field-guide) that sit alongside Section 8 in the mortgage origination compliance stack. The rules interact, the audit files share evidence, and the architecture that gets one of them right tends to get the others right because the discipline is the same. The discipline is to encode the rule at the workflow layer rather than to manage to it at the script layer, and to keep the audit file in the shape the regulator will ask for before the regulator asks.

---

_Source: [https://www.seiright.com/blog/respa-section-8-ai-mortgage-referrals-kickback](https://www.seiright.com/blog/respa-section-8-ai-mortgage-referrals-kickback) · Sei AI_