Tez I Solutions

Industries · Mortgage and lending

How we handle your borrowers’ data

Your pipeline is full of applicants who never closed — and every one of them is a consumer under federal privacy law. An outside vendor working that list is the first thing your examiner will ask about.

Applies to
Non-bank mortgage lenders and brokers; home equity and personal-loan lenders
Covers
Lead reactivation and AI services
Current as of
October 2026 · V1
Book a call

Free download

Get the Mortgage and lending documentation as a PDF.

Tell us where to send follow-up questions, and the download starts straight away.

We use these details only to send you this document and to follow up with you about it — by email, or by phone. We won't add you to a mailing list, sell them, or share them with anyone. See our privacy policy.

Part 1

Lead reactivation

How your list is checked, ranked and worked — what leaves your side, what never does, and the rules your industry adds.

What we take, where it goes, and who sees it.

For compliance, operations and production leaders

Lead reactivation works the borrowers your firm has already generated — people who asked about a loan or started an application, were followed up once or twice, and then were not. We rank them by likelihood to book a conversation and work the list from the top, so your loan officers spend their time on the people most likely to come back.

Every one of those records is regulated: by the Gramm-Leach-Bliley Act and the FTC Safeguards Rule, by RESPA and Regulation Z, by fair lending law, and by federal and state telemarketing law. This document sets out, control by control, how the engagement keeps your firm inside those rules — and which decisions stay with your firm. It is written to be read by your compliance officer and the Qualified Individual who runs your information security program, and filed with your vendor due diligence.

Sections 02 to 06 are the plain-English part. They show the whole system in pictures, explain the four things the ranking looks at, and follow one example lead from your CRM to your loan officer’s desk. The sections after them are the detail your compliance team will check.

01

Commitments

Ten commitments, before the detail

  • 01 · Direct identifiers never reach an AI provider — Names, phone numbers, emails, property addresses and loan numbers stay in a secure environment. The AI provider sees a random ID and a few short facts, nothing more.
  • 02 · Nothing goes out without your approval — Your compliance officer or designated approver approves every template before use. Our review sits on top of yours, never in place of it.
  • 03 · Every message is kept as your record — Outreach goes through your own channels, or a copy of every message, sent and received, is delivered into your systems as it happens.
  • 04 · No evidenced consent, no text message — Consent, opt-outs, Do Not Call status and quiet hours are checked at the moment of sending. Opt-outs are permanent.
  • 05 · Your data serves your engagement only — No model training, no reuse across clients, no sale, no outside enrichment, and no credit-report data in the ranking.
  • 06 · Breach notice within 72 hours — As soon as possible and no later than 72 hours — within 24 hours for firms licensed in New York, whose own clock to DFS is 72 hours.
  • 07 · Return and destruction, certified — At the end of the engagement your data, including the identity key, is returned or destroyed, and we confirm it in writing.
  • 08 · No rates, no terms, no loan advice — Messages invite a conversation with your loan officer. No rates, payments, loan terms, pre-approval claims or product recommendations.
  • 09 · A fee structure your compliance team can approve — A flat fee, or a results-based fee only where counsel approves it in writing. How RESPA and Regulation Z treat each is set out in section 14.
  • 10 · Evidence on request — Every control here has something you can inspect. We complete your vendor due diligence, including your Safeguards Rule review, during onboarding.

02

The whole picture

The system on one page

The system works in three zones. Names and contact details stay on your side. Our scoring works only with a random ID and a few short facts. The results come back to your side, where your firm approves and sends.

Three zones: your firm keeps identity, Tez scoring sees random IDs and facts, results return to your firm YOUR FIRM · NAMES AND CONTACT DETAILS STAY HERE G2  Check the list Opt-outs, Do Not Call Swap names for IDs The key stays with you G3  Notes into facts Answers from a fixed list TEZ SCORING · RANDOM IDS AND FACTS ONLY Score and rank Four factors, plain arithmetic G4  Explain A reason and a short summary Pick the next step Channel and timing YOUR FIRM · RESULTS LAND HERE Match back ID to your CRM record G5  Your approval Your compliance team signs off G6  Send In your name, from your channels Outcomes (booked, applied, funded) come back by random ID and improve your firm’s ranking only. G1  The agreement and intake review clear before any data moves. G7  Retention and destruction.

G1 — the agreement and intake review — clears before any data moves. G7 covers retention and destruction.

Dashed gold-edged boxes, marked G1 to G7, are compliance gates. Nothing passes a gate until it clears. Section 03 lists what each one checks.

03

Step by step

From our first conversation to the first message

Seventeen steps in five phases. The gate marks show where a compliance check has to clear before the work moves on.

Onboard (before any data)

  • 01 First conversation — How your firm works its pipeline today
  • 02 Agreement — Roles, purpose, subprocessors, 72-hour notice G1
  • 03 Intake review — Lead sources, consent, licensed states G1
  • 04 Secure environment — In your environment, or our workstation

Prepare (on your side)

  • 05 Map your stages — Your loan statuses to eight shared stages
  • 06 Check the list — Opt-outs, Do Not Call, loans in process, complaints G2
  • 07 Swap names for IDs — A random ID; the key stays with you
  • 08 Notes into facts — Fixed-list answers; no credit data G3

Rank (Tez scoring)

  • 09 Score — Intent, Opening, Reachability, Fit
  • 10 Tier and reasons — A to D, with the top reasons
  • 11 Short summary — Why call, and an opening line, checked G4
  • 12 Next step — Channel by consent, and timing

Act (on your side)

  • 13 Match back — The ranked list appears in your CRM
  • 14 Your approval — Your compliance team approves the wording G5
  • 15 Send — Re-checked at sending; your name and NMLS ID G6

Learn (ongoing)

  • 16 Record outcomes — Booked, applied, funded; a held-back group
  • 17 Improve and report — Your ranking only; results you can check G7
GateWhenWhat has to be true to passWho decides
G1Before any dataSigned agreement; lead sources, consent wording, opt-out records and privacy-notice delivery reviewed; your licensed states, any state that licenses this work, and your fee model confirmedYour firm and Tez
G2As the list arrivesOpt-outs, Do Not Call, loans in process, complaints, bought or prescreened leads who never responded to you, and anyone hostile are set aside and never rankedSoftware, on your rules
G3Before anything leaves your sideOnly approved facts cross: no names, contact details, notes, loan numbers, credit data, exact figures or exact datesSoftware, checked by Tez
G4After a summary is draftedNo personal data, no rates or terms, no banned topics, nothing the facts do not supportSoftware, then your loan officer
G5Before any outreachYour compliance officer or designated approver approves the wording and the batchYour firm
G6At the moment of sendingConsent for that channel, Do Not Call, quiet hours in the lead’s time zone, opt-outs, your licence in the lead’s stateSoftware, on your channels
G7Throughout, and at the endOutcomes kept by random ID, used for your firm only, returned or destroyed at terminationTez, certified to you

04

The ranking

How the list is put in order

Every lead is checked first, then looked at through four factors. The factors combine into one score, and the list is sorted by it. The sorting is plain arithmetic, not an AI model, so every score can be explained in a sentence.

Leads pass a gate, are scored on four factors, and produce an order, a reason and a next step G2  Checked first Opt-outs, Do Not Call, loans in process Set aside Never ranked or contacted THE FOUR FACTORS · BUILT FOR EVERY LEAD Intent How far they got, and how recently Opening Why they paused, and what has changed Reachability Can we reach them, with consent Fit Can your firm still serve them Score and rank Plain arithmetic, no AI model Value tier Optional, your call Order and tier A to D, best first Why this lead Top reasons, in one line Next step Channel, timing, angle
FactorThe question it answersWhat it looks at, for a lender
IntentHow serious were they, and how long ago?How far they got: inquiry, prequalification, application, approval without closing. Time since the last real conversation. Whether they replied before. How many follow-ups were tried.
OpeningIs there a real reason to get back in touch?The reason they paused: rates, not ready to buy, a purchase that fell through, another lender. A condition they named, such as “once we sell our current place.” Whether that moment has now passed.
ReachabilityCan we reach them, and are we allowed to?A working email or phone. Consent on file for each channel. Do Not Call status, and whether the three-month inquiry window has passed.
FitCan your firm still serve them?A state you are licensed in. Whether they asked about a product you still offer. Nothing about their credit, income or assets.

How the ranking gets better

  • 01It starts from your team’s experience — Your loan officers tell us how often each kind of paused borrower comes back. Those estimates set the first odds. Until your results come in, the scores are educated estimates, and we say so.
  • 02It corrects itself from your results — As conversations are booked and applications taken, the odds for each reason move toward what happened at your firm.
  • 03It becomes a model of your firm alone — After about 300 recorded outcomes, the ranking is fitted to your firm’s results. It is never trained on, or pooled with, another firm’s data.
  • 04A held-back group keeps us honest — With your agreement, about one lead in ten is held back and not contacted for a set period. Comparing the two groups shows what the outreach actually changed, rather than what would have happened anyway.

05

A worked example

One lead, from your CRM to your loan officer

Jordan Hale is fictional. His record is shaped like a real one. Follow it through the four cards, left to right and top to bottom.

1 · Your CRM record (stays with you)

FieldValueNote
nameJordan Halekey table
phone(512) 555-0163key table
emailjordan.h@example.comkey table
loan IDLN-2026-04417key table
statusWithdrawn by borrowerbecomes a stage
noteWants to sell his condo before buying; listing it this summerbecomes facts
amount$385,000 requestedtier, if opted in
creditScore from the application pullnever used
last contactApril 2, 2026becomes days

2 · What we receive (leaves your side)

FieldValueNote
IDs_4Hn8tQ1crandom
stageApplication submittedIntent
days since187Intent
reasonTimingOpening
conditionYes, waiting on an eventOpening
reachEmail yes · text no consentReachability
fitLicensed state yes · product offered yesFit
value tier2 of 3, only if you opt inoptional

Never leaves your side: his name, phone, email, loan ID, the note itself, the loan amount, his credit score, exact dates.

3 · What comes back (back to you)

FieldValueNote
IDs_4Hn8tQ1c
score68 · tier A
reasons+ named a condition · + application submitted · − 187 days since contact
reason lineApplication submitted; paused for an event that should have passed.fixed wording
summaryHe paused until something he was waiting on had happened. By now it likely has. A good opener is to ask whether his plans have moved.AI, checked
next stepEmail, Thursday evening his time

4 · In your CRM (stays with you)

FieldValueNote
recordJordan Hale · tier A · 68matched back
fieldsReason line, summary and next step added to his record
draft“Hi {first_name}, …” with his name filled in only at sending, and your NMLS ID in the footerplaceholder
flagsText blocked, no consent · Do Not Call check requiredG6

No text consent on file, so the next step is email. His application is over three months old, so a Do Not Call check comes before any call. His loan officer sees his name here, in your own CRM, for the first time.

  • 01His note became a few facts, and his credit stayed out — “Selling his condo” left only as “waiting on an event.” His credit score was never exported: it was pulled for his application, not for marketing. If your compliance team approves a short list of event types, the summary can be more specific without the note ever leaving.

06

The suggestion

How the “why call” note is written and checked

Each lead comes back with a reason line in fixed wording and a short summary drafted by an AI model, both written only from the facts in card 2 of section 05. The AI writes; it does not decide. The order comes from arithmetic, and your firm approves anything sent.

Facts feed a fixed-wording reason line and an AI draft; both pass a check before a loan officer sees them, then approval and sending The facts and reasons Random ID, no names or figures Your rules Script, never-say list Reason line Fixed wording, filled from the facts AI draft Short summary and an opening line G4  Checked before anyone sees it Personal data, rates and terms, made-up facts Passes Goes to your loan officer, in your CRM Fails Reason line only, flagged for review G5 + G6  Approve, then send Your sign-off; checks at sending

The four checks at gate G4

  • No personal data — The draft may not contain a name, number, email or address. Names appear only as a placeholder, filled in on your side at sending.
  • No rates, no terms, no promises — No rates, payments, loan amounts, “pre-approved” or “guaranteed”, and nothing suggesting we are their current lender or servicer. The draft invites a conversation, nothing more.
  • Nothing made up — Every statement in the summary must trace back to one of the facts it was given. A statement that does not is treated as a failure.
  • Your never-say list, and fair lending — Phrases and topics your firm rules out are blocked, and so is any wording that could discourage someone on a prohibited basis.

07

Roles

Your firm stays the financial institution

Under the Gramm-Leach-Bliley Act your firm is the financial institution and Tez I Solutions is a service provider. You share consumer information with us under the service-provider exception in Regulation P (12 CFR 1016.13), on a contract that limits our use of it to this engagement; the reuse and redisclosure limits in 1016.11 then bind us. Applicants who never closed are consumers under the rule — whether or not credit was extended — so the dormant list is covered in full. The FTC Safeguards Rule (16 CFR 314.4(f)) requires your firm to choose service providers that can protect it, bind them by contract and assess them periodically. This is how responsibility divides.

ObligationYour firmTez I Solutions
Privacy notices and opt-outsOwns, including delivery of your initial privacy noticeHonours your suppression and opt-out lists
Approval of communicationsCompliance approves before useSubmits templates; never alters approved wording
Information security programOwns, through your Qualified IndividualOperates the controls in sections 10 and 11; reports to your Qualified Individual on request
Licensing and advertising rulesOwns, state by state; supplies your NMLS ID and required wordingApplies them to every template; sends only into states where you are licensed
How leads were obtainedWarrants it in the agreementRecords the consent source and enforces it at send time
Records of messagesOwnsSends through your channels, or delivers a copy of every message
Breach notice to consumers, the FTC and statesOwns; FTC within 30 days when 500 or more consumers are affectedNotifies you within 72 hours, or 24 for New York licensees, and supports the investigation
Vendor due diligencePerforms, documents and repeats periodicallyCompletes questionnaires and supplies evidence
Access and deletion requestsReceives and instructsExecutes within your deadline and confirms
Return or destructionInstructsExecutes and certifies in writing

08

Data

What we take, and what we never take

The least data that lets the list be ranked and worked. Where a figure carries signal, we keep a tier and drop the figure.

DataHow it is handledReaches the AI model
Name, phone, email, addressesHeld only in the encrypted key table on your side; used to address a message when it is sent.No
Loan and account numbers, SSN, birth date, IDsExcluded from the export. If one appears in a note, the record is quarantined and the value removed.No
Credit scores, reports, denial reasonsNot requested, not used. Pulled for an application; the Fair Credit Reporting Act limits them to that purpose.No
Requested loan amountOnly after a fair lending review your team approves: a three-level tier, made on your side. Income, assets and property value are never used.No
Inquiry, application, contact datesConverted to intervals — “14 months since last contact.”Interval only
Source, stage, purpose, loan officerKept as categories; loan officer names replaced with codes.As categories
Call and loan notesRead on your side and turned into fixed-list answers (section 09). The note itself never leaves.Fixed-list answers only
Consent and opt-outsKept as an append-only history and enforced at send time.No
Age, sex, race, marital status, ZIP codeNot requested, not used to rank, not sent.No

09

The privacy layer

Where AI sits, and what it sees

StationWhat happensHolds or sees
01 · Your firm — Your CRMFull records stay here. Only the agreed fields are read.Holds: everything, as today
02 · Secure env. — Separation, and notes into factsA local model with no outside connection answers fixed questions about each note. The key lives only here.Holds: identifiers, key table, notes
03 · Tez scoring — Ranking, then a short summaryRanking is arithmetic in our own code. Anthropic’s Claude words a summary from the facts.Sees: random ID, fixed-list facts
04 · Your channels — Approved, then sentNames filled in on your side; consent checked; sent in your name, with your NMLS ID.Holds: the message, as a firm record

Notes are the most sensitive part of a record and the most useful. So they are read where they already are, and only short answers leave. Five layers make sure of it.

LayerWhat it does
1 · Field allow-listOnly approved columns are read. Structured identifiers are separated by rule, not by a model’s judgement.
2 · Identifiers stripped firstPhone numbers, emails, loan, account and card numbers, Social Security numbers, credit scores, dollar amounts and dates are removed from a note before the local model reads it.
3 · Fixed questions, fixed answersThe local model answers set questions — why they paused, whether they named a condition, which kind of event — by choosing from a list your compliance team has seen. It cannot write free text.
4 · Unsure means a person decidesAn answer the model is not confident about is not guessed. The record is held for a person to review. The process fails closed.
5 · Final scanEvery record is scanned again before it leaves. A record carrying anything outside the approved fields is held back, not sent.

10

Where it runs

Two ways to run the secure environment

The secure environment is where identifiable data lives during the engagement, and your firm chooses where it sits. In both options the key table never leaves it, and the AI provider receives only a random ID and fixed-list facts.

Inside your environmentOn a Tez-operated workstation
Where raw data sitsYour infrastructure; it never leaves your perimeterAn encrypted workstation operated by Tez, in a separate encrypted volume per firm
AccessYou control it. We use a named, least-privilege account you can revoke at any timeOne named individual, with multi-factor authentication
NetworkYour controlsOutbound limited to the AI provider and your sending channels
LoggingYour logging stackAccess log kept for the engagement, provided on request
SuitsFirms with an IT team or managed IT providerFirms without capacity to host it

Workstation standard, when Tez operates it

  • Encrypted at rest — Full-disk encryption, plus a separate encrypted volume for each firm’s data.
  • Dedicated and locked down — An OS account used only for client data, multi-factor sign-in, screen lock, current patches, endpoint protection.
  • No cloud sync — Client data is never stored in a synced folder or consumer cloud backup, and is excluded from sync by configuration.
  • Wiped at the end — The firm’s volume is destroyed at termination and the destruction certified in writing.

11

Security and incidents

The controls, and what happens if one fails

ControlEvidence you can ask for
Written information security program, modelled on the FTC Safeguards Rule and Massachusetts 201 CMR 17.00The program document
Encryption in transit and at restConfiguration record
Multi-factor authentication on every account that touches your dataAccount inventory
Access limited to named individuals, reviewed quarterlyAccess list and review log
Access to your data loggedLog extract
Background check for anyone with access, on your requestWritten confirmation
Raw data held only for the life of the engagementRetention schedule
Return or destruction within 30 days of terminationSigned certificate
Cyber liability insurance, where your contract requires itCertificate of insurance
Written incident response plan, reviewed each yearThe plan and its last review

If something goes wrong

  • 01 · Contain — Stop sending, revoke the affected credential, isolate the system.
  • 02 · Notify you within 72 hours at the latest — Within 24 hours for New York licensees, so your 30-day FTC and 72-hour DFS clocks stay intact. Which records, what window, what we have done.
  • 03 · Preserve evidence — Logs, send records and the consent history are kept intact for your investigation.
  • 04 · Support your notice, never pre-empt it — Notice to borrowers, to the FTC and to states is your firm’s decision. We never contact your borrowers about it.

12

Advertising and fair lending

Your marketing rules cover this work

Using a vendor does not move responsibility: the federal ban on unfair, deceptive or abusive practices reaches service providers too. The CFPB has narrowed its supervision and withdrawn guidance since 2025, but the rules below are unchanged, and states can enforce them.

  • 01No trigger terms — Regulation Z — No rate, payment, downpayment, term or finance charge, so nothing triggers the disclosures of 12 CFR 1026.24, or 1026.16 for home equity lines. Terms, if wanted, come from your team.
  • 02No misrepresentation — Regulation N — No “pre-approved”, “guaranteed” or government-sounding claims, and nothing suggesting the message comes from the borrower’s current lender or servicer (12 CFR 1014.3). Every message names your firm.
  • 03Your licence on every message — Your licensed name and NMLS ID appear as your state rules require — Connecticut requires the identifier on all solicitations. Messages go only into states where you are licensed.
  • 04Fair lending — ECOA and the Fair Housing Act — Nothing that would lead a reasonable person to expect denial or worse terms on a prohibited basis (Regulation B 1002.4(b), as amended in 2026; 24 CFR 100.75). The ranking uses no protected characteristic or stand-in (section 04).
  • 05An approval trail you can produce — For every template: the AI draft, the approved wording, who approved it and when — kept at least 24 months, as Regulation N requires.
  • 06Rails the drafting stops at — Rates, terms, credit, complaints and anything like loan advice go to your loan officer. Outreach never asks for an SSN, account details or documents, and every draft passes G4 (section 06).

13

Consent

No evidenced consent, no text message

Under the Telephone Consumer Protection Act the party that sends a message can be liable for it directly. So consent is enforced in software at the moment of sending, not left to policy.

  • 01 · A consent history, not a checkbox — Each lead’s consent is an append-only record of source, wording, channel and date. Contactability is derived from it, never overwritten.
  • 02 · Checked when the message sends — Consent, suppression, Do Not Call and quiet hours are rechecked at sending, not at list-building. Yesterday’s opt-out holds today.
  • 03 · Aged leads without evidenced consent — Email, or your loan officers directly — not text. An inquiry or application over three months old is no established business relationship for Do Not Call.
  • 04 · Do Not Call and reassigned numbers — Checked against the National Do Not Call Registry and the FCC Reassigned Numbers Database. Results dated and stored.
  • 05 · Quiet hours and state law — 8am to 9pm in the lead’s local time, narrower where required. Florida, Oklahoma, Maryland, Washington and Texas SB 140 rules applied by state.
  • 06 · Opt-out everywhere, permanently — STOP, a request in any words, or one made to your loan officer is honoured on every channel and survives any re-import. Email unsubscribes within 10 business days.

14

How we are paid

Two fee models, and how each is treated

We offer two models. Fees are set in conversation and are not in this document. What belongs here is how each model is treated under the rules that apply to your firm — for mortgage and home equity lending, chiefly RESPA section 8 and the Regulation Z loan-originator rules — so your compliance team can choose with that in view.

Model A: Flat monthly fee

A fixed fee for operating the service. It does not depend on applications, loans funded or loan revenue.

  • RESPA section 8. Payment for services actually performed, at market value and not tied to the volume or value of business, is what section 8(c) and 12 CFR 1024.14(g) permit.
  • Regulation Z. We advertise your services in your name and never take applications or discuss terms; the official commentary to 1026.36 does not treat advertising another’s services as loan origination.
  • The simplest model to approve, document and supervise.

Model B: Results-based fee

A share of revenue attributed to reactivated loans, measured from your CRM’s own records.

  • RESPA section 8. A payment that rises with funded loans is evidence of an agreement to refer (12 CFR 1024.14(e)), and referrals are not compensable. For an outside vendor, that is the core section 8 risk.
  • Regulation Z and licensing. Pay tied to loans the outreach influenced can make the payee a loan originator, whose pay may not be based on loan terms (1026.36(d)); loan revenue moves with rate and fees. Connecticut ties contingent pay to broker licensing. Tez I Solutions holds no NMLS licence.
  • Personal loans fall outside RESPA; state broker licensing can still apply. Model B is offered only where your counsel approves it in writing for the states involved.

15

Subprocessors and contract

Who else touches the data, and what we sign

SubprocessorWhat it holdsWhen
Anthropic (Claude API)A random ID and fixed-list facts only. No names, contact details, notes or exact figures. Not used for training by default under Anthropic’s commercial terms; zero data retention requested where offered; can run through your own cloud provider insteadEvery engagement
Your CRM, loan origination and messaging providersMessages and records, as they do todayEvery engagement, under your contracts
Messaging provider operated by TezPhone numbers and message contentOnly if you do not send through your own texting channel, and only with a copy of every message delivered to you
Operator console hostingNamed in your agreement, with what it holdsWhere your team uses our console

The full list is an exhibit to your agreement. We give 30 days’ notice before adding or changing a subprocessor.

What the agreement contains

  • Roles and purpose — Your firm as financial institution, Tez as service provider; use limited to this engagement, as Regulation P 1016.13 and the Safeguards Rule require.
  • No secondary use — No training, no sale, no reuse across clients, no enrichment, no credit data, without written instruction.
  • Security exhibit and incident notice — The controls in sections 09 to 11 as binding terms, with the access, encryption and notice terms New York’s Part 500 asks for.
  • Consent warranty — Your firm’s representation of how the leads were obtained, and which came from bought or prescreened sources.
  • Privacy requests — Access, correction and deletion within your deadlines, including state privacy laws for data outside the GLBA exemption.
  • Audit, return and destruction — Audit and questionnaire rights, including your periodic Safeguards Rule assessment; return or certified destruction at termination.

16

Regulatory map

Each rule, and where it is addressed

RuleWhat it requires hereSection
FTC Safeguards Rule, 16 CFR 314Security program, vendor oversight, FTC notice07 10 11
Regulation P 1016.11 and 1016.13Use limited to the contracted purpose07 15
FCRA; Homebuyers Privacy Protection ActNo credit-report data; no trigger leads08 13
RESPA section 8, Regulation X 1024.14No payment for referrals14
Regulation Z 1026.36Loan originators and their pay14
Regulation Z 1026.24, 1026.16; Regulation NTrigger terms; misrepresentation; records06 12
State licensing and NMLS ID rulesLead-generator licences; NMLS ID03 12 14
ECOA, Regulation B; Fair Housing ActNo discouragement; no prohibited basis04 08 12
Dodd-Frank sections 1031 and 1036Unfair, deceptive or abusive practices12
TCPA and the Do Not Call rulesConsent, revocation, Do Not Call, quiet hours13
State telemarketing lawsState consent and calling-hours rules13
CAN-SPAMUnsubscribe, address, accurate headers13
NYDFS Part 500; Mass. 201 CMR 17.00Vendor security, notice; written program10 11 15
State comprehensive privacy lawsRequests for data outside the GLBA exemption15
NIST IR 8053, SP 800-188De-identification practice08 09

17

References

Every requirement, traced to its source

Each commitment in this document rests on a primary source: a statute, a regulation, a regulator’s own guidance, or a court decision. They are listed here so your compliance team can check our reading against the text. Links were current in October 2026.

Privacy and data securityRelied on forSections
Gramm-Leach-Bliley Act, 15 U.S.C. 6801–6809 (law.cornell.edu/uscode/text/15/chapter-94)The federal basis for protecting consumers’ nonpublic personal information07 08
FTC Safeguards Rule, 16 CFR Part 314 (ecfr.gov/current/title-16/chapter-I/subchapter-C/part-314)Service-provider oversight (314.4(f)); encryption, MFA, incident response; FTC notice for 500 or more consumers within 30 days (314.4(j))07 10 11
FTC, Safeguards Rule amendment, 88 FR 77499, November 2023 (federalregister.gov/d/2023-24412)The notification requirement and its May 13, 2024 effective date07 11
FTC, Safeguards Rule: What Your Business Needs to Know (ftc.gov/business-guidance/resources/ftc-safeguards-rule-what-your-business-needs-know)Plain-language reading, including mortgage brokers and lenders in scope07 11
Regulation P, 12 CFR Part 1016 (ecfr.gov/current/title-12/chapter-X/part-1016)Service-provider exception (1016.13); reuse and redisclosure limits (1016.11); applicants as consumers (1016.3)07 15
NYDFS Cybersecurity Regulation, 23 NYCRR Part 500 (dfs.ny.gov/cybersecurity)Covered entities include NY-licensed mortgage bankers and brokers; service-provider policy (500.11); 72-hour notice (500.17)01 11 15
Massachusetts 201 CMR 17.00 (mass.gov/regulations/201-CMR-1700-standards-for-the-protection-of-personal-information-of-ma-residents)Written security program, encryption on laptops and in transit, service-provider contracts10 11
New York SHIELD Act, N.Y. Gen. Bus. Law 899-aa and 899-bbReasonable safeguards and breach notice for New York residents’ data11
California Consumer Privacy Act, Cal. Civ. Code 1798.100 et seq., incl. 1798.145(e); amendments narrowing GLBA exemptions in Oregon, Minnesota, Montana (Oct 2025), Connecticut (Jul 2026)Data-level GLBA exemptions; access, deletion and no-sale duties for data outside them15
Credit reports, settlement and payRelied on forSections
Fair Credit Reporting Act, section 604, 15 U.S.C. 1681b (law.cornell.edu/uscode/text/15/1681b)Consumer reports used only for a permissible purpose; prescreened firm offers08
Homebuyers Privacy Protection Act, Pub. L. 119-36, September 5, 2025 (congress.gov/119/plaws/publ36/PLAW-119publ36.pdf)Limits on mortgage trigger leads, effective 180 days after enactment13
RESPA section 8, 12 U.S.C. 2607 (law.cornell.edu/uscode/text/12/2607)No fee or thing of value for referrals of settlement business; services actually performed14
Regulation X, 12 CFR 1024.14 (ecfr.gov/current/title-12/chapter-X/part-1024/subpart-B/section-1024.14)Referral defined; volume-linked payment as evidence; reasonable value14
CFPB, RESPA FAQs, October 7, 2020 (files.consumerfinance.gov/f/documents/cfpb_respa_frequently_asked_questions.pdf)Marketing services versus referrals; payments based on referrals prohibited14
CFPB Advisory Opinion, Digital Mortgage Comparison-Shopping Platforms, 88 FR 9162, February 2023 (files.consumerfinance.gov/f/documents/cfpb_respa-advisory-opinion-on-online-mortgage-comparison-shopping-tools_2023-02.pdf)Pay that varies with placement or volume as evidence of a referral fee; still listed as current guidance14
Regulation Z, 12 CFR 1026.36, and official commentary (consumerfinance.gov/rules-policy/regulations/1026/interp-36)Loan originator includes referring for compensation; advertising another’s services excluded; no pay based on loan terms14
SAFE Act, 12 U.S.C. 5102 (law.cornell.edu/uscode/text/12/5102)Federal loan originator definition; licensing through NMLS by state14 18
State licensing, advertising and fair lendingRelied on forSections
Conn. Gen. Stat. 36a-485, 36a-486, 36a-498d, 36a-498h (cga.ct.gov/current/pub/chap_668.htm)Lead generator licensing since 2018; contingent pay treated as broker activity; NMLS ID on all solicitations12 14 18
Cal. Bus. & Prof. Code 10131(d) and 10137, Real Estate Law 2026 (dre.ca.gov/files/pdf/relaw/relaw.pdf)Soliciting borrowers for compensation is broker activity; no pay to unlicensed persons14
California DFPI, interpretive opinion on brokers under the California Financing Law (dfpi.ca.gov/wp-content/uploads/sites/337/2019/03/OP-7408-4-27-16.pdf)Clerical work not paid by commission or tied to a loan closing may not need a broker licence; a fact-specific test14
Regulation Z, 12 CFR 1026.24 and 1026.16 (ecfr.gov/current/title-12/chapter-X/part-1026)Trigger terms in closed-end and home equity advertising12
Regulation N, Mortgage Acts and Practices — Advertising, 12 CFR Part 1014 (ecfr.gov/current/title-12/chapter-X/part-1014)No misrepresentation of source, preapproval or government affiliation; 24-month records06 12
Dodd-Frank Act sections 1031, 1036 and 1042, 12 U.S.C. 5531, 5536, 5552 (law.cornell.edu/uscode/text/12/5536)Unfair, deceptive or abusive practices by covered persons and service providers; state enforcement12
CFPB, withdrawal of guidance documents, 90 FR 20084, May 12, 2025 (federalregister.gov/d/2025-08286)The 2025 narrowing of CFPB guidance12
Regulation B, 12 CFR 1002.4(b) and 1002.6; final rule, 91 FR 21620, April 22, 2026 (federalregister.gov/d/2026-07804)Discouragement in advertising; no prohibited basis in evaluating applicants; effective July 21, 202604 12
Fair Housing Act, 42 U.S.C. 3605; 24 CFR 100.75 and 100.500 (ecfr.gov/current/title-24/subtitle-B/chapter-I/subchapter-A/part-100)Lending and advertising discrimination; the discriminatory-effects rule, still in force while HUD’s 2026 proposal to remove it is pending04 08 12
Telemarketing and emailRelied on forSections
Telephone Consumer Protection Act, 47 U.S.C. 227 (law.cornell.edu/uscode/text/47/227)Consent for marketing texts and calls; statutory damages; sender liability13
FCC rules, 47 CFR 64.1200Prior express written consent, Do Not Call, quiet hours, revocation of consent13 18
Telemarketing Sales Rule, 16 CFR Part 310Established business relationship: 18 months after a transaction, 3 months after an inquiry or application13
National Do Not Call Registry; FCC Reassigned Numbers Database (donotcall.gov · reassigned.us)The two checks run before any number is contacted13
Facebook, Inc. v. Duguid, 592 U.S. 395 (2021)Scope of the federal autodialer definition13
McLaughlin Chiropractic Associates v. McKesson Corp. (U.S. Supreme Court, June 2025)Courts are not bound by FCC interpretations of the TCPA13 18
Florida Telephone Solicitation Act, Fla. Stat. 501.059; Texas SB 140 (2025), Tex. Bus. & Com. Code ch. 302State rules that treat texts as solicitation, applied by the lead’s state13
CAN-SPAM Act, 15 U.S.C. 7701 et seq.; FTC compliance guide (ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business)Unsubscribe within 10 business days, physical address, accurate headers13
De-identification and AIRelied on forSections
NIST IR 8053, De-Identification of Personal Information, 2015 (nvlpubs.nist.gov/nistpubs/ir/2015/NIST.IR.8053.pdf)Quasi-identifiers, re-identification risk, why pseudonymization is reversible08 09
NIST SP 800-188, De-Identifying Government Datasets, 2023 (csrc.nist.gov/pubs/sp/800/188/final)Generalization of figures into bands; de-identification as risk reduction08 09
NIST SP 800-122, Guide to Protecting the Confidentiality of PII, 2010 (csrc.nist.gov/pubs/sp/800/122/final)Classifying which fields are identifying and how sensitive each is08
Anthropic, Commercial Terms of Service (anthropic.com/legal/commercial-terms)No training on commercial API inputs and outputs by default09 15

18

Limits

What this document is not

  • 01Not legal advice — Your compliance officer and counsel decide how these rules apply to your firm. This document tells them what we do so they can.
  • 02Not a SOC 2 report — We do not hold a SOC 2 attestation. We complete your due diligence questionnaire and provide the policies and evidence behind each control instead.
  • 03Not a lender, broker or loan originator — Tez I Solutions holds no NMLS licence, takes no applications, quotes no rates or terms, gives no loan advice and never places a borrower with any lender. Where a state may license this kind of work — Connecticut licenses lead generators — its residents are left out at G1 until your counsel clears it.
  • 04Not permanent — Current as of October 2026. Rules in this area are moving — the FCC’s consent revocation rules, HUD’s disparate-impact rule and the CFPB’s own structure among them. We review this document at least once a year and whenever a rule it relies on changes.
  • 05Not a promise of results — A score says which borrowers to contact first, not who will apply or close. Until your firm’s own outcomes are recorded, the odds behind it are estimates drawn from your team’s experience.

Part 2

AI services

How we handle your data inside your AI tools.

AI services puts a standing AI engineer across the AI tools your business already pays for — ChatGPT, Claude, Copilot, Gemini and the AI inside the software you use. We connect those tools to your CRM, inbox, calendar and files, build the workflows that do the repeat work, set up your processes and brand voice inside them, train your team, and look after the accounts. The tools stay yours, under your own contracts with each vendor. That changes how data protection works. In lead reactivation we keep names and contact details away from the AI provider by design. Here, a tool connected to your CRM will see the records it needs to do its job, names included. So the protections in this document are about something else: whose account it is, what each connection can reach, what the vendor's terms allow, who approves a workflow before it goes live, and how our access ends. Sections A1 to A6 are the same for every business we work with. Your industry's page adds sections A7 and A8: what your industry's rules mean for your AI tools, and where those rules come from.

A1

Commitments

Ten commitments, before the detail

  • 01 · Your accounts, on business terms — We work only in accounts your business owns, on business plans. Never in a personal or consumer account — ours, yours or an employee's.
  • 02 · A named user, never a shared password — You add us as a named user with our own sign-in and multi-factor authentication. We never ask for a shared password, and you should never send one to us by email.
  • 03 · The least access that does the job — Each connection reaches only the data its workflow needs. Admin rights are used for set-up tasks, logged, and reduced afterwards.
  • 04 · Nothing goes live without your approval — Every connection, workflow and set of instructions is approved by you, in writing, before it is switched on. So is every change to one that is already running.
  • 05 · Automated where it is routine, a person in the loop where it is not — Routine, rule-based work can run on its own once you have approved it: sorting and tagging, summaries for your team, filing, and fixed-wording notices such as an appointment reminder from a template you signed off. Anything that needs a decision or a judgement — a reply written for a particular customer, a price, a term, a commitment, or anything your industry's page marks for review — has a person on your side in the loop before it goes anywhere. Each workflow's data sheet (A3) states which kind it is. Where your industry's page requires a person to review every customer message, that rule applies instead.
  • 06 · Your data stays in your tools — We build and test inside your workspace. We do not copy your business data into Tez's own AI accounts. Where we need examples outside your tools, we use made-up records.
  • 07 · Your data serves your business only — No reuse across clients, no sale, no training of anything by us. Prompts, instructions and workflows built from your material are yours and are not reused for anyone else.
  • 08 · Vendor data settings confirmed in writing — During onboarding we read each tool's current data terms, set its controls with you, and give you a written record. We re-check it whenever a vendor changes its terms.
  • 09 · Incident notice within 72 hours — As soon as possible and no later than 72 hours, or sooner where your industry's page says so.
  • 10 · Access removed, everything handed over — At the end, our access is removed, every credential we created is revoked or rotated, and every workflow is documented and handed over. We confirm it in writing within 30 days.

Evidence on request: every control in this document has a record you can inspect, listed in section A5. If you send us a security questionnaire, we complete it during onboarding.

A2

Accounts and access

Whose accounts, and how we get in

The AI tools, the systems they connect to and the accounts that run them all belong to your business. We are given access to them; we never hold them. That is what makes the work portable: if we part ways, everything keeps running and nothing has to be moved out of our name.

Nothing in this section starts before you have decided to work with us. The first session is a conversation. We do not ask for logins, exports or files beforehand.

Who owns what

ItemWho owns itHow we handle it
AI subscriptions and workspacesYour business — in its name, on its payment methodWe administer them for you. If a new tool would genuinely pay for itself, we say so and why, and it goes on your account, not ours.
The owner or super-admin roleYour business keeps it, with at least one person on your side who is not usWe take the lowest role that does the work. Admin rights for a set-up task are logged and reduced when the task is done.
Our sign-inA named user for each Tez person, in your directory or invited as a guestMulti-factor authentication on. No shared logins. Never a password by email, text or chat.
Connections — API keys, app authorisations, service accountsCreated in your accounts, owned by youEach one is named, listed in the access record (A5), scoped to its workflow, and stored where you control it — never pasted into a document or a message.
Personal and consumer AI accountsTheir ownersNot used for business work. If the audit finds business work in personal accounts, we tell you and recommend moving it to business seats. We never connect your systems to a personal account.

How access is set up

  • 01 · You decide what we need — We list the systems and the role each one needs, and why. You agree or strike items out.
  • 02 · You add us — Usually by adding us as a user, on a call where we walk through it together. A password is never sent.
  • 03 · We confirm the safeguards — Multi-factor authentication on, role as agreed, sign-in recorded in the access record.
  • 04 · Reviewed every quarter — We go through the access record with you and remove anything no longer needed.
  • 05 · Removed at the end — Section A5 sets out how access ends.

Roles

ObligationYour businessTez I Solutions
Contracts with AI vendorsHolds them; the vendor processes your data under themReads the data terms and records them for you (A4)
Who on your team uses which toolDecidesSets up the seats and permissions you decide
Approval of workflows and changesApproves before anything goes liveProposes, documents and builds; never switches on without approval
Review of customer-facing outputYour people review before sendingBuilds the review step into the workflow
Industry rules — licensing, records, supervision, privacy noticesOwns themBuilds to them as set out on your industry's page (A7)
Security of the access we holdOverseesOperates the controls in A5
Notice to customers or regulators after an incidentDecidesTells you within 72 hours and supports the investigation
Privacy requests from your customersReceives and instructsHelps find and remove the data in the tools we set up, within your deadline

Where a state privacy law reaches your business, we act as your service provider or processor for the access we hold, and the agreement says so.

A3

What the tools can see

Connections, scopes, and what stays out

An AI tool connected to one of your systems can see whatever that connection lets it see. An assistant that searches your files can surface anything the person using it is allowed to open — Microsoft says of its own Copilot that it "only surfaces organizational data to which individual users have at least view permissions." So the work starts with permissions, not prompts.

Before anything is connected

  • Who can see what — Before an assistant is pointed at a shared drive, inbox or CRM, we review who can open what. Folders that are open to everyone but should not be are fixed first, with your approval.
  • One data sheet per workflow — Every workflow gets a written sheet: what it reads, what it writes, who sees the result, and where a person reviews it. You approve the sheet before we build.
  • Add-ons are vendors too — A plug-in, connector, custom GPT or agent from a third party is another company handling your data. None is installed without your approval and a record of its data terms.

What a data sheet looks like

Illustration — not real client data.

WorkflowReadsWritesWho sees the resultHuman in the loop
Morning summary of new inquiriesCRM records created in the last 24 hours: name, source, service asked about, notesOne summary in the office manager's inboxOffice managerRead before anyone acts on it
First-draft reply to a quote requestThe single inquiry being answered; your approved reply templates and brand voiceA draft in the shared inbox — never sentThe person replyingEdited and sent by a person
Meeting notes to CRMThe transcript of a meeting the attendees agreed to recordA note on the matching CRM recordThe account ownerChecked before it is saved

What stays out by default

Unless your written decision says otherwise, after we have shown you what the vendor's terms and your industry's rules say, these are kept out of every prompt, instruction file and connection:

  • Social Security numbers, tax IDs, driver's licence and other government ID numbers
  • Full bank account and card numbers
  • Passwords, access codes, gate, door and alarm codes
  • Health information
  • Anything your industry's page (A7) marks as restricted

Where a connection would expose one of these — a folder of scanned IDs, a CRM field holding card numbers — the connection is narrowed to leave it out, or not built.

Before a workflow goes live

CheckWhat has to be true
Data sheetApproved by you, in writing
AccessThe connection reaches only what the sheet lists
TestRun on made-up records, or on a small set of real ones you choose, with the results shown to you
Human in the loopMarked on the data sheet. In place wherever a decision or judgement is involved: a reply written for a particular customer, a price or a term, or anything your industry's page marks for review. Routine, rule-based steps you have approved may run without one
OwnerA named person on your side who can pause it
DocumentationWritten up and stored where you keep it

AI tools make mistakes. Every workflow is built on the assumption that a draft can be wrong, which is why the review step is part of the design rather than a suggestion.

A4

Model providers and training

What the vendors do with your data, and how we check

When your team or a workflow uses ChatGPT, Claude, Copilot or Gemini, the vendor processes that data under its contract with your business. Tez is not a reseller or partner of any AI vendor, and not a party to those contracts. Whether a vendor trains its models on your data, how long it keeps prompts, whether its staff can review them, and where it processes them are set by the vendor's terms and by the plan you are on. Those terms change. So we do not promise what a vendor will do. We promise what we will do.

What we do

  • 01 · List every tool — Each AI tool and AI feature you pay for, the plan it is on, and who holds a seat.
  • 02 · Read the current terms — For each one we record what the vendor says about training on your data, retention, human review, where data is processed, and which admin controls exist. Each entry carries the date and the address of the page we read.
  • 03 · Set the controls with you — Where a tool offers a choice — training opt-outs, retention periods, chat history, which add-ons are allowed — we recommend the most protective setting that still does what you need, and you decide.
  • 04 · Give you the record — The settings record is yours, in writing. Your compliance contact can file it with your vendor reviews.
  • 05 · Re-check when terms change — When a vendor changes its data terms, we re-read them, update the record and tell you if anything moved. We also re-check the whole record at least every six months.

What the main vendors said when we checked

Read on each vendor's own page on 6 October 2026. This is the vendor's statement to its customers, not a commitment by Tez. If it changes, the vendor's current terms govern, and we update your record.

VendorWhat its page saidWhere
Anthropic — Claude, under its Commercial Terms"Anthropic may not train models on Customer Content from Services." The Commercial Terms cover Anthropic's API and the business offerings that reference them, not its consumer plans. Effective June 17, 2025.anthropic.com/legal/commercial-terms
OpenAI — APIData sent to the API is not used to train or improve OpenAI's models unless you opt in, as of March 1, 2023. Abuse-monitoring logs are kept for up to 30 days by default.developers.openai.com/api/docs/guides/your-data
OpenAI — ChatGPT business plansRead during onboarding against OpenAI's published terms for your plan, and recorded the same way.openai.com/enterprise-privacy
Microsoft — Copilot"Prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation LLMs." Copilot only surfaces data a user already has permission to view.learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy
Google — Gemini in Google WorkspaceYour content "is not human reviewed or otherwise used for Generative AI model training outside your domain without permission."knowledge.workspace.google.com/admin/gemini/generative-ai-in-google-workspace-privacy-hub
AI inside your other softwareEach vendor's own terms — your CRM, phone system or accounting software. Recorded the same way during onboarding.Each vendor's data or AI terms page

Consumer plans and personal accounts

Consumer plans run on different terms from business plans — Anthropic's Commercial Terms, for one, state that they do not cover its consumer offerings. We do not do business work in consumer or personal accounts, and we do not rely on a consumer plan's settings to protect business data.

Our own tools

We use AI tools of our own to write code and documentation. Your business data does not go into them. Anything we build for you is built and tested in your workspace, or against made-up records.

A5

Records, security and incidents

What we keep, how it is protected, and what happens if something fails

Records you can inspect

Records describe what we built and changed. They live in a location your business owns, and they do not contain your customers' data.

RecordWhat it holds
Access recordEvery Tez sign-in and every connection we created: system, role or scope, date added, date reviewed, date removed
Settings recordEach tool's plan, data terms as read on a given date, and the controls set (A4)
Workflow registerEvery live workflow, with its approved data sheet (A3), its owner on your side and its documentation
Change logEvery change to a live workflow or connection: what changed, why, when, who approved it
Training recordWho was trained, on which tools and processes, and when

Changes to something already running

Every change to a live workflow is described, approved by you, tested, logged and reversible — the previous version is kept. If a vendor update breaks a workflow, or a workflow starts doing something it should not, we may pause it at once to stop harm, and tell you the same day. We do not change what a workflow does without your approval.

Security controls

ControlEvidence you can ask for
Written information security program, modelled on Massachusetts 201 CMR 17.00The program document
Multi-factor authentication on every account we use to reach your systemsAccess record
Access limited to named individuals, reviewed with you quarterlyAccess record and review log
Tez devices: full-disk encryption, screen lock, current patches, endpoint protectionConfiguration record
Your data is not stored on Tez devices or in our cloud storage; where a task needs a file locally, it is deleted when the task endsWritten procedure
Background check for anyone with access, on your requestWritten confirmation
Cyber liability insurance, where your contract requires itCertificate of insurance
Written incident response plan, reviewed each yearThe plan and its last review

If something goes wrong

An incident includes a workflow sending or showing data to the wrong person, a connection reaching more than its sheet allows, a suspected compromise of a Tez sign-in, and a vendor incident we learn of that affects your workspace.

  • 01 · Contain — Pause the workflow, revoke the affected credential or connection, and stop the exposure.
  • 02 · Notify you within 72 hours at the latest — Which systems and data, what window, what we have done. Sooner where your industry's page sets a shorter time. Updates follow as we investigate.
  • 03 · Preserve evidence — Logs, the change log and the vendor's audit records are kept intact for your investigation.
  • 04 · Support your notice, never pre-empt it — Whether customers or regulators are told is your decision. We never contact your customers about it. A vendor's own incident notice comes to you under your contract with that vendor.

When the engagement ends

StepWhat happens
Access removedEvery Tez user and role removed from every system, checked against the access record
Credentials endedEvery key, app authorisation and service account we created is revoked, or rotated and handed to you
Nothing in our nameAnything that ended up in a Tez name is transferred to you first — by design there should be nothing
Handover packThe workflow register, data sheets, settings record, change log and plain-English run notes, so your team or the next provider can keep it running
Written confirmationWithin 30 days of the end of the engagement

Everything we build is yours — documented and portable. If we part ways, it keeps working and your team knows how it runs.

A6

Limits

What this document is not

  • 01 · Not legal advice — Your counsel or compliance officer decides how your industry's rules apply to your business. This document tells them what we do so they can.
  • 02 · Not a SOC 2 report — We do not hold a SOC 2 attestation. If you send a security questionnaire, we complete it and provide the policies and evidence behind each control instead.
  • 03 · Not the vendor — We do not run the AI vendors' systems or set their terms. Their security, retention, outages and incidents are governed by your contract with them. We are not a reseller, partner or certified provider of any AI vendor.
  • 04 · Your obligations stay yours — Using AI tools does not move your duties to us: licensing, supervision, record keeping, privacy notices and your staff's use of AI remain your business's responsibility. We build to them, as your industry's page sets out.
  • 05 · Not a promise about AI output — AI tools can be wrong, and confidently so. That is why anything that needs a judgement has a person in the loop, and why no workflow makes a decision your business would otherwise make.
  • 06 · Not permanent — Current as of October 2026. Vendor terms and the rules on AI change often. We review this document at least once a year, and whenever a vendor term or rule it relies on changes.

A7

Mortgage lending

What your industry's rules mean for your AI tools

A mortgage business's loan origination system, inbox and shared drive hold some of the most regulated consumer data there is: applications, credit reports, income and asset documents. Federal law treats a lender or broker as a financial institution that must oversee every vendor reaching that data — AI vendors and outside engineers included. And whatever an AI tool drafts for a borrower is your advertising, under rules on trigger terms, misrepresentation and fair lending. These are the rules that bear on your AI tools, and what we do about each.

RuleWhat it means for AI toolsHow we handle it
FTC Safeguards Rule, 16 CFR Part 314 [1][2]Mortgage lenders and brokers must oversee service providers, require safeguards by contract, use multi-factor authentication, and notify the FTC within 30 days of a qualifying incident involving 500 or more consumers. AI vendors and Tez are service providers.Incident notice to you within 72 hours at the latest. The settings record and access record (A4, A5) are written for your vendor oversight file. Multi-factor authentication on every Tez sign-in.
New York 23 NYCRR Part 500 [3]New York-licensed mortgage bankers and brokers need a third-party service provider policy and must notify DFS within 72 hours.For firms licensed in New York, we tell you within 24 hours, so your own clock stays intact.
Regulation P, 1016.11 and 1016.13 [4]Applicants are consumers. Information shared with a service provider may be used only for the contracted purpose.Our use is limited to your engagement, in writing. We record whether each AI vendor's terms allow any other use, and set the controls with you.
Fair Credit Reporting Act, section 604 [5]Credit reports may be used only for a permissible purpose. Feeding credit data into a general AI workspace or a marketing workflow is a use.Credit reports and scores are kept out of every AI connection unless your compliance team approves a specific, permissible use in writing.
Regulation Z 1026.24 and 1026.16; Regulation N [6][7]A rate, payment or term in an ad triggers required disclosures. Regulation N bars misrepresenting preapproval, government affiliation or the source of an offer, and requires records for 24 months. AI tools write "you're pre-approved" and "rates as low as" freely.Rates, payments and preapproval language are blocked in drafts unless your approved disclosure goes with them. Approved versions are kept where your firm keeps advertising records.
NMLS ID and state licensing rules [8]State licensing rules require your NMLS ID on advertising; Connecticut, for one, requires it on all solicitations.Your company and originator NMLS IDs are set in the tools' instructions and every template.
ECOA and Regulation B; Fair Housing Act [9][10]Advertising may not discourage applicants on a prohibited basis, and the Fair Housing Act reaches lending and advertising. An AI tool that sorts, targets or words outreach differently by neighbourhood, age or family status can create that risk.No AI tool targets, ranks or tailors outreach on protected traits or close proxies for them, such as neighbourhood. Any workflow that sorts borrowers goes through your fair lending review before it goes live.
Dodd-Frank sections 1031 and 1036 [11]Unfair, deceptive or abusive practices are barred for lenders and for service providers to them.Drafts are reviewed by a person at your firm before they reach a borrower. No workflow sends to borrowers without review.
State privacy laws narrowing the GLBA exemption [12]Data outside the federal exemption can carry state access and deletion rights.We help locate and remove that data in the tools we set up, within your deadline.

In practice

  • A workflow may draft a rate-watch follow-up for a loan officer to send. It may not state a rate or payment unless your complete disclosure goes with it, and it never says "pre-approved."
  • An assistant may help processors find the right condition letter or checklist. It is not connected to the folders where credit reports and bank statements sit unless your compliance team approves that scope.
  • A meeting or call-note tool may summarise a borrower conversation into your LOS. Income and asset figures stay in the LOS record, not in a general AI workspace.
  • During onboarding we check for AI tools staff use on borrower files without approval, and list them for your compliance lead.

A8

Sources

Where these rules come from

1. FTC Safeguards Rule, 16 CFR Part 314 — https://www.ecfr.gov/current/title-16/chapter-I/subchapter-C/part-314 2. FTC, Safeguards Rule: What Your Business Needs to Know — https://www.ftc.gov/business-guidance/resources/ftc-safeguards-rule-what-your-business-needs-know 3. NYDFS Cybersecurity Regulation, 23 NYCRR Part 500 — https://www.dfs.ny.gov/cybersecurity 4. Regulation P, 12 CFR Part 1016 — https://www.ecfr.gov/current/title-12/chapter-X/part-1016 5. Fair Credit Reporting Act, section 604, 15 U.S.C. 1681b — https://www.law.cornell.edu/uscode/text/15/1681b 6. Regulation Z, 12 CFR Part 1026 — https://www.ecfr.gov/current/title-12/chapter-X/part-1026 7. Regulation N, Mortgage Acts and Practices — Advertising, 12 CFR Part 1014 — https://www.ecfr.gov/current/title-12/chapter-X/part-1014 8. Conn. Gen. Stat. 36a-485 to 36a-498h — https://www.cga.ct.gov/current/pub/chap_668.htm 9. Regulation B, 12 CFR 1002.4(b) and 1002.6; final rule, April 22, 2026 — https://www.federalregister.gov/d/2026-07804

  • 01Fair Housing Act regulations, 24 CFR Part 100 — https://www.ecfr.gov/current/title-24/subtitle-B/chapter-I/subchapter-A/part-100
  • 02Dodd-Frank Act sections 1031 and 1036, 12 U.S.C. 5531, 5536 — https://www.law.cornell.edu/uscode/text/12/5536
  • 03Gramm-Leach-Bliley Act, 15 U.S.C. 6801–6809 — https://www.law.cornell.edu/uscode/text/15/chapter-94

Questions your counsel or compliance team wants answered first?

We go through this document on the free first session, and complete your security or due-diligence questionnaire during onboarding.

All industries