Three Things Must Be True Before You May Charge a Dollar. Your Subscription Charges on Day Zero.
Debt-relief rules do not cap fees or require them to be reasonable. They forbid collecting any fee at all until a specific debt has been settled, an agreement is in place, and the consumer has paid under it. Recurring billing is structurally incompatible with that gate — and it is the default pricing model for every AI product.
Why this is a product problem, not a legal one. Most compliance findings ask a team to add a disclosure, a log or a gate. This one asks them to change how the company makes money, which is why it needs to be resolved before the pricing page ships rather than after the first enforcement letter. A team that discovers it eighteen months in has a book of business built on a fee structure it cannot keep.
The Fee Gate: All Three, Per Debt, Before Any Money
These conditions are conjunctive and they attach to an individual debt rather than to the programme. Settling one account in a portfolio of nine unlocks a fee proportionate to that account only.
A specific debt has been renegotiated, settled, reduced or otherwise altered
In plain terms: Not the enrollment. Not the plan. One identified account, changed.
How products break it: Products bill on activation, on the first negotiation attempt, or on a plan being generated. Each of those precedes the condition, and none of them is cured by good faith about what happens later.
The consumer has entered into an agreement with that creditor
In plain terms: A concluded arrangement on that account, not a settlement offer received or a proposal drafted.
How products break it: Automated pipelines generate offers at volume, and the temptation is to treat a strong offer as the milestone. The milestone is the agreement, and it belongs to the consumer and the creditor rather than to the product.
The consumer has made at least one payment under that agreement
In plain terms: Money has actually moved on the settled account.
How products break it: This is the condition most products do not track at all. It requires visibility into a payment that often happens outside the product entirely, which means the billing system cannot verify the fact it is legally required to verify before it charges.
Five Triggers, and How Each One Gets Satisfied by Accident
Nobody sets out to build a regulated debt-relief provider. The definitions are broad enough that a general-purpose financial assistant walks into them one feature at a time.
The product renegotiates or reduces a debt
Debt relief service definitions reach any programme represented to renegotiate, settle or in any way alter the terms of payment of an unsecured debt. Interest-rate reduction and payment restructuring are inside; the word 'settlement' appearing nowhere in your marketing is not a defence.
Satisfied by accident when
A generic 'financial assistant' that drafts hardship letters and negotiates payment plans has performed the described service without anyone on the team deciding to enter this market.
The service is sold over the phone or by other covered means
The federal advance-fee rule attaches to telemarketing, and the scope question — whether an inbound call, a callback, a voice agent or a chat-to-call handoff qualifies — determines whether the rule reaches you directly.
Satisfied by accident when
An AI voice agent that returns a missed call is making the call, and a chat flow whose next step is a scheduled phone consultation has built a covered channel into a web product.
A fee is collected before the three conditions are met
Any fee — setup, subscription, credits, priority access — collected before a specific debt is settled and paid, is the violation. Renaming it does not help, and neither does calling it a fee for software.
Satisfied by accident when
Standard SaaS pricing is an advance-fee structure. This is the single most common way an otherwise careful team builds an unlawful product.
Consumer funds are held or directed
Where funds are set aside for settlements, the rules require the account to be at an insured institution, the consumer to own the funds and accrued interest, the consumer to be able to withdraw and cancel at any time without penalty, and the administrator to be unaffiliated with the provider and not fee-sharing with it.
Satisfied by accident when
Holding balances triggers the money-transmission perimeter as well, so the same design decision lands you in two regimes at once.
The service is offered to residents of licensing states
Many states license or register debt adjusters, debt management providers or debt settlement companies, with bonding, fee caps, contract-form requirements and, in several states, an outright prohibition on for-profit debt adjusting.
Satisfied by accident when
A national signup form with no state gate is an offer in every one of them, and offering without registration is frequently its own violation regardless of whether a consumer enrolled.
What the Rule Requires vs What the Chatbot Says
Pre-enrollment disclosures are prescribed in content and timing. A conversational interface optimised for completion produces a reassuring paraphrase instead, and the paraphrase is what the consumer actually received.
| What must be disclosed | What a generated flow says | Why the substitution fails |
|---|---|---|
| The amount of time before the provider will make a settlement offer to each creditor | "We'll start working with your creditors right away." | The requirement is a number of days or months, stated before enrollment. A reassurance is not a disclosure, and a generated one varies per conversation, so there is no consistent artefact to point at later. |
| The amount of money or percentage of each debt the consumer must accumulate before an offer is made | "Just make your monthly deposits and we'll handle the rest." | The threshold is the thing consumers most need to evaluate the offer, and it is the thing an engagement-optimised conversation is least likely to volunteer. |
| That the funds are the consumer's, may be withdrawn at any time without penalty, and what happens on withdrawal | Nothing, usually — this appears in a terms page nobody reads. | Placement matters. These are pre-enrollment disclosures, which means before the consumer commits, in the flow that commits them. |
| That non-payment may result in collection efforts, lawsuits and increasing balances, and may affect creditworthiness | "Most clients see their debt reduced significantly." | The negative disclosure and the positive claim are usually in tension, and a model trained to be encouraging resolves that tension in the wrong direction every time unless the disclosure is rendered outside the model's control. |
| Savings claims stated against the enrolled balance, including accrued fees and charges | "Clients typically save 40 to 50 percent." | Success-rate and savings representations carry substantiation requirements and a prescribed basis for the calculation. A number a model produces because it is a plausible-sounding number is an unsubstantiated claim made individually, at scale, in an unreviewed channel. |
The first-charge test
Take any consumer who has paid you. Find the first charge on their account, then answer three questions with records rather than recollection: which specific debt had been settled at that moment, where is the agreement with that creditor, and what payment had the consumer made under it.
If the first charge predates all three, the fee structure is the finding — and it is the same finding for every customer in the table, which is what turns a compliance issue into a restitution number.
Frequently Asked Questions
We're a software vendor, not a debt-relief company. Does this reach us?
It depends on what the software does and how it is sold, and the safe assumption is that a product doing the substantive work is treated as providing the service. The definitions describe a service represented to renegotiate, settle or alter the terms of an unsecured debt — they do not carve out services delivered as software. If the consumer is your customer, your product communicates with creditors on their behalf, and you charge them for it, you are the provider in every respect that matters, and the fact that no employee touched the account cuts the other way if anything. The genuinely different posture is licensing your tooling to a regulated provider who holds the consumer relationship, holds the licences and makes the disclosures. That posture only survives if it is real: your customer is the provider rather than the consumer, you do not bill the consumer, and you do not present a consumer-facing brand promising debt reduction. Where products get into trouble is a hybrid — vendor framing in the contracts, consumer-facing service everywhere the consumer can see.
Can we charge a subscription if we call it a software fee rather than a settlement fee?
No, and the label is the least interesting fact about the transaction. The advance-fee prohibition is written in terms of requesting or receiving payment of any fee or consideration for a debt relief service before the three conditions are met. The analysis follows what the consumer is paying for and when, so a monthly charge running from enrollment through months of accumulation is fee collection before settlement whatever the invoice line says. Two structures do work. Charge nothing until an individual debt is settled, an agreement is in place and a payment has been made, then charge in proportion to that specific debt — which is the compliant contingent model the rules contemplate. Or charge a different customer entirely: sell to the regulated provider, where your fee is a business expense rather than a consumer fee, and the consumer's fee timing remains the provider's obligation. Everything in between — trial periods, credits sold upfront, freemium with paid negotiation, deferred billing that still accrues from enrollment — is the same structure with more steps.
What is the state licensing layer underneath the federal rule?
It is the layer that most often stops a product first, because it applies to the activity rather than the channel and has no telemarketing predicate to argue about. A majority of states regulate debt adjusting, debt management or debt settlement under some name, and the requirements typically include registration or licensure, a surety bond scaled to volume, fee caps expressed as a percentage of debt or of savings, prescribed contract terms with cancellation rights, trust-accounting rules for consumer funds, and periodic reporting. Several states prohibit for-profit debt adjusting outright, permitting only non-profit agencies, and in those states the compliant answer is not to operate. Because licensure attaches where the consumer is, a national product needs either fifty-state coverage or state gating at signup, and gating has to be enforced rather than disclosed. There is also an unauthorised-practice-of-law question in some states, where negotiating a consumer's debts or advising on the consequences of default is treated as legal work — which an AI negotiator does by design.
Does the advance-fee ban apply to a pure web product with no phone calls?
The federal rule's reach turns on the telemarketing predicate, and that is a live question rather than a safe harbour. Voice agents, callbacks, SMS-to-call flows and scheduled consultations all pull a web product toward coverage, and the analysis is done afterwards, on your actual call logs, by someone constructing the least favourable reading. More importantly, avoiding the federal rule does not leave you unregulated. State debt-adjuster statutes apply to the activity irrespective of channel, and their fee caps and trust rules frequently impose similar or stricter timing constraints. General unfair-and-deceptive-practices authority reaches the conduct at both federal and state level. And practically, a business model that is lawful only because it avoided a phone call is one product decision away from unlawful, made by a growth team that has never read this rule. Design to the substance — no fee before a specific debt is settled and paid — and the channel question stops being load-bearing.
What does compliant AI negotiation actually look like?
Narrower than the pitch deck, and viable. The automation that is clearly fine is the work that does not touch fee timing or the consumer relationship: intake and document handling, debt inventory and verification, creditor identification and routing, drafting communications for review, tracking offers and deadlines, and producing the audit record. The controls that make it defensible are specific. Bill only on a per-debt basis after settlement, agreement and first payment, with each of those three facts recorded as a verified event rather than inferred from a status field. Render disclosures as fixed, versioned, logged artefacts outside the model's output, and capture acknowledgement per consumer with the version shown. Enforce a hard gate on any generated claim about savings, timelines or success rates. Keep consumer funds out of the product, or if you cannot, meet the dedicated-account conditions exactly and accept the money-transmission analysis that comes with holding them. State-gate enrollment. And log every creditor communication verbatim, because the record of what was said on the consumer's behalf is the first thing requested when this goes wrong.
How does this differ from the credit-repair rules we already reviewed?
Different statute, different counterparty, same structural collision with subscription billing — and you can be inside both at once. Credit repair law governs services represented to improve a consumer's credit record, history or rating, and its advance-fee prohibition bars payment until the promised services are fully performed. The debt-relief rules govern renegotiating what the consumer owes, and their fee gate is per-debt with the settlement, agreement and payment conditions described above. A product that both disputes tradelines and negotiates balances is squarely in both regimes, subject to two advance-fee prohibitions with different trigger events, two disclosure regimes and two sets of state licensing statutes. Teams building broad 'financial health' assistants tend to arrive here without noticing, because each feature was added on its own merits. The practical move is to inventory features against both definitions before pricing is set, since pricing is the thing that has to change and it is the hardest thing to change later.
Related Reading
- AI credit dispute letters and credit-repair law — the parallel advance-fee ban on the other side of the consumer's file.
- AI agents, held funds and money transmission — what happens the moment a dedicated account sits inside your product.
- Auto-renewal and cancellation law for AI SaaS — the recurring-billing rules that apply even where the advance-fee ban does not.