Taking an assessment rather than buying one? This page is written for employers. Here is the page for candidates.

Collections and retention · Mid level

How to assess a Third-Party Debt Recovery Specialist

Everything in that list is either a credential check or a post-hire control. The licence proves the person sat an exam; the monitoring catches the breach after a real consumer has already experienced it. Nothing in the process, at any point before the first live call, observes the candidate at the specific fork where the money and the rule separate. And the compliance module often teaches the fork wrongly: it teaches 15 U.S.C. 1692c(c), under which a cease request must be in writing, and leaves agents believing that a spoken "stop calling me" can be ignored. It cannot. Regulation F at 12 CFR 1006.14(h) prohibits a collector from continuing to use a medium the consumer has asked them not to use, and the CFPB has explained that a consumer asking a collector to "stop calling" bars further telephone communication. An agent who has been trained on the statute and not the regulation will fail this in production and will pass every quiz you give them.

This is the seat the Fair Debt Collection Practices Act was written about. Unlike the first-party arrears agent, a third-party recovery specialist collects debts owed to another, which puts them inside the statutory definition of "debt collector" and inside every duty that follows from it. The consequence for hiring is that a single sentence spoken in the wrong order on a single call is a compliance event with a private right of action attached, and the person most likely to say that sentence is a competent, motivated agent nine minutes into a call that is going badly.

Four behaviours are cleanly observable in a transcript and are worth naming individually, because each corresponds to a specific rule and a specific slice of the complaint volume. The first is the disclosure. Section 1692e(11) requires the collector to disclose that they are attempting to collect a debt and that information obtained will be used for that purpose, in the initial communication — including where that initial communication is oral. On a call that opens with a consumer already shouting, the disclosure is the easiest thing in the world to skip or to mumble past, and the CFPB's complaint data shows the consequence: among consumers complaining about written notifications, 47 percent reported the notification did not disclose that it was an attempt to collect a debt.

The second is the dispute. Section 1692g(b) requires a collector who receives a written dispute within the validation period to cease collection of the disputed amount until verification is obtained and mailed. Disputes on a call do not arrive as legal notices; they arrive as "that's not mine" and "I already paid that." The CFPB reports that attempts to collect a debt not owed has been the predominant complaint issue every year since 2013, with 60 percent of those consumers saying the debt is not theirs and 28 percent reporting identity theft. The assessable moment is the turn immediately after the consumer says it: does the agent record the dispute and change track, or does it get processed as an objection and answered with a rebuttal?

The third is the stop request, and it is the sharpest fork available anywhere in this corpus. Under 12 CFR 1006.14(h) a consumer's request that the collector stop using a medium binds the collector in that medium — a spoken "stop calling me" ends telephone contact, without any writing. In a transcript this is a single-turn, unambiguous, binary observation: after the consumer says it, does the next turn ask another question about payment? Nothing else about a candidate is this cleanly measurable, and no interview has ever measured it. It is worth adding that call-frequency compliance is not a defence here: the CFPB's supervisory findings describe collectors whose call volume sat within the seven-calls-in-seven-days presumption at 12 CFR 1006.14(b)(2)(i) but who placed over 100 calls after being asked to stop, and examiners found the presumption overcome.

The fourth is the accuracy of stated consequences. Section 1692e(5) prohibits threatening any action that cannot legally be taken or is not intended to be taken. In practice this is not agents threatening arrest; it is the softer, commoner version — implying a court process the client has no intention of funding, or letting a consumer believe a credit consequence is automatic when it is discretionary. That is a content judgment on a sentence, and a human reviewer can make it from a transcript in seconds once the sentence has been surfaced.

One further note for buyers. Some states require debt collector licensing or registration and some do not; that is an employer-verified gate that sits entirely outside any simulation. Savvanta ranks handling ability and never verifies a credential. And as with every voice role in this family, the scoring criteria are behavioural — disclosure made, dispute honoured, stop request honoured, consequence accurately stated — never accent, dialect or manner of speech.

What the job actually needs

ceasing collection activity the moment a debt is disputed
honouring a stop request in the medium in which it was made
making the required disclosure on a call that is hostile from the first second
keeping to verifiable statements about consequences
staying on the ask without escalating tone

How people fail in this seat

asks one more question after acknowledging a stop request
continues pressing after the consumer says the debt is not theirs
omits or buries the debt-collection disclosure
implies legal action that will not be taken
discusses the debt with whoever answers the phone

What most employers do instead

Licence and registration checks where the state requires them, an induction compliance module with a test, call monitoring after hire, and a scripted role-play against a supervisor.

The assessment

About 32 minutes end to end.

The systems it runs in

An agency collections platform holding the consumer account as placed by the client: the balance and placement, a contact-attempt log, per-medium contact permissions and stop flags, a dispute marker with validation and verification tracking, promises and payment arrangements, and the scheduler that sets the next action. Finvi describes Velosidy as a collections platform for third-party collection agencies and Simplicity as a collections workflow platform for small and start-up agencies; C&R Software describes Debt Manager as an all-in-one collections solution used by banks, debt buyers and other creditors on past-due accounts. This design depends on that shape more than any other in the corpus, because the two events it is built around — the oral dispute and the medium-specific stop request — are duties that persist after the call ends, and a duty that lives only in a transcript does not persist at all.

Finvi Velosidy
Finvi Simplicity
C&R Software Debt Manager

Any agency or debt-buyer platform with an account, a contact-attempt history, channel-level contact permissions, a dispute state and a next-action scheduler. Where a buyer's platform enforces medium-specific suppression or call-frequency controls itself, the fixture uses theirs and the candidate is observed working with the guardrail rather than around it; where it does not, ours is used and the buyer is told so, because whether the system stops the agent or the agent stops themselves changes what the score means. Rebuilt against the buyer's instance where a sandbox exists.

Working speed is not scored. Speed is the mechanism of the breach here, not a constraint to be balanced against quality. The whole design turns on the stop request arriving at minute nine, with a payment one question away, precisely because the fast close and the compliant move come apart at that moment. Scoring time to resolution, calls per hour or collection rate would reward the turn the rubric treats as a 1, and the adverse_impact section already warns that a deployer who reintroduces productivity metrics alongside this assessment restores the exact incentive it tests resistance to. The only timings reported are positional — where the disclosure fell, and what the turn after the stop request contained.

What the candidate actually does

Task	What happens
The stop request that arrives at the worst possible moment live_call · 13 min	An AI consumer on a charged-off credit card balance now owned by the candidate's employer, so the candidate is a debt collector within 15 U.S.C. 1692a(6) and every duty that follows applies. The call is hostile from the first second, which makes the initial oral disclosure required by 1692e(11) the easiest thing in the world to skip. At around minute five the consumer says the account is not theirs. At around minute nine, after the candidate has re-established the conversation and the consumer has softened and is discussing a figure, the consumer says not to call this number again. That placement is the whole point of the task: the stop request lands at the moment of maximum commercial temptation, with the payment one question away, and 12 CFR 1006.14(h) makes the correct move a single-turn binary that a reviewer can read in the transcript without interpretation.
The account a colleague picks up cold written_artifact · 8 min	The account as the candidate leaves it: the note, the dispute state, the per-medium contact permissions, and the next action they schedule. The cheap version reads "consumer uncooperative, retry Thursday" and takes ninety seconds. The correct version records the dispute and what specifically was disputed, sets the telephone medium to stopped rather than describing the request as a mood in prose, and schedules nothing that would breach it. Requiring the flag and the scheduled action, not only the sentence, is what makes this task observe a duty rather than an intention: a candidate can say the right thing on the call, write a sympathetic note about it, and still book a call for Thursday, and only the account state distinguishes that candidate from one who did the job. The task is scored against the candidate's own transcript, so the record cannot be better than the call it describes.
Six lines, permitted or not judgment_scenario · 9 min	Six single sentences a collector could say next, each with the surrounding facts. The candidate marks each permitted or not permitted and gives the operative rule in one sentence. The items are chosen so that the intuitive answer is wrong at least three times. A spouse answers and asks what the call is about. A team leader instructs that a spoken stop request can be ignored because the statute says writing. A call is placed at 8.30pm where the consumer is and 11.30pm where the agent is. A collector says legal options will be looked at, on a client portfolio that is never litigated below a threshold stated in the brief. A fourth call in seven days is placed to a consumer who asked the collector to stop, and the agent reasons that seven in seven is compliant. A consumer disputes the debt on the phone and asks for verification, where both intuitive answers are wrong.

The mark scheme

Each criterion is scored 1 to 5 against written anchors, and every score is reported with the excerpt that earned it. A criterion marked floored is reported as a finding rather than averaged into the total. The first is open; open any other to read its anchors in full.

Makes the required initial disclosure on an oral call that opens badly (15 U.S.C. 1692e(11), failure to disclose in the initial communication, and if that communication is oral in that initial oral communication, that the collector is attempting to collect a debt and that any information obtained will be used for that purpose)weight 0.2The disclosure is made in the opening exchange, in full, and the candidate holds the line to complete it before responding to the consumer's opening h…

1 The transcript contains no statement that this is a debt collector attempting to collect a debt, or the statement appears only after a payment amount has already been discussed.

3 The disclosure is present and complete but delivered as one run-together formula while the consumer is talking over it, and is not repeated when the consumer later asks who this is.

5 The disclosure is made in the opening exchange, in full, and the candidate holds the line to complete it before responding to the consumer's opening hostility, then confirms the consumer heard it.

Treats an oral dispute as a change of track rather than an objection to rebut (15 U.S.C. 1692g(b), cessation on a written dispute until verification is obtained and mailed, read together with 1692e(2)(A) and 1692e(10) on false representation of the character, amount or legal status of a debt)weight 0.2Establishes what specifically is disputed, states accurately that a written dispute inside the validation period obliges the collector to stop until v…

1 The next turn rebuts the dispute and repeats the demand, for example by asserting that the account shows in the consumer's name, and collection on the full amount continues unchanged.

3 Records that the consumer disputed it, then carries on asking for payment on the same amount in the same call without telling the consumer what disputing does or does not do.

5 Establishes what specifically is disputed, states accurately that a written dispute inside the validation period obliges the collector to stop until verification is obtained and mailed and that an oral one does not, explains how to make it, and stops asserting the disputed amount as owed.

Honours a spoken stop request in the medium in which it was made, in the turn it is made (12 CFR 1006.14(h), a debt collector must not communicate or attempt to communicate through a medium if the person has requested that the collector not use that medium)weight 0.2Stops using the telephone in that turn, states back what will and will not now happen, offers a permitted alternative medium once without pressing, an…

1 The turn immediately after the request contains a further question or statement about payment, or the candidate tells the consumer the request has to be in writing.

3 Acknowledges the request and stops pressing, but then reopens the ask while closing, for example by asking to take the card details already discussed, or leaves a telephone follow-up scheduled on the account.

5 Stops using the telephone in that turn, states back what will and will not now happen, offers a permitted alternative medium once without pressing, and the account note records the request as binding on telephone contact.

States only consequences that will actually follow (15 U.S.C. 1692e(5), the threat to take any action that cannot legally be taken or that is not intended to be taken)weight 0.15States what the client actually does, marks what is discretionary as discretionary, and declines to fill the silence when the consumer supplies a wors…

1 Says or implies that legal action, a court process, wage attachment or an automatic credit consequence will follow, where the brief says the client does not litigate this portfolio or that the consequence is discretionary.

3 Speaks in the passive and the conditional so that no specific claim is made but the consumer is plainly left believing something the brief does not support, and the candidate does not correct it.

5 States what the client actually does, marks what is discretionary as discretionary, and declines to fill the silence when the consumer supplies a worse consequence themselves.

In the judgment items, identifies the operative rule rather than the familiar one (12 CFR 1006.14(h) against 15 U.S.C. 1692c(c); 1692c(a)(1), local time at the consumer's location; 1692c(b), third parties; 12 CFR 1006.14(b)(2)(i), the seven-in-seven presumption)weight 0.15Right answer and right operative provision on at least five items, including naming Regulation F rather than the statute on the stop request, and comp…

1 Marks the spoken stop request as ignorable because the statute requires writing, or treats the seven-in-seven presumption as a defence to continued calling after a stop request, or tells the spouse what the call concerns.

3 Reaches the right answer on four or more items but cites the wrong provision or gives no reason, so the answer is recall rather than reasoning and will not transfer to an item not on the list.

5 Right answer and right operative provision on at least five items, including naming Regulation F rather than the statute on the stop request, and computing the calling hour at the consumer's location rather than the agent's.

The account note lets a colleague act without re-reading the callweight 0.1Records what was disputed, the medium the stop request binds, the permitted next action, and nothing about the consumer's character.

1 Records a characterisation of the consumer rather than the two events, and schedules a call that the stop request forbids.

3 Records both events but not what was disputed or which medium the request covered, so the next agent has to guess.

5 Records what was disputed, the medium the stop request binds, the permitted next action, and nothing about the consumer's character.

How it is scored

Weighted mean of the six criteria, each scored 1 to 5 against the anchors, reported with the time-marked excerpt that earned each score. The three criteria drawn directly from statute or regulation are also reported unweighted as a conduct panel, because a buyer in this seat usually wants to see the stop-request score on its own rather than averaged with commercial ability. Whether a 1 on the stop-request criterion is a hard stop is the deployer's decision and should be made in writing before the first candidate sits the assessment, not after seeing the names.

Integrity

monitored session, call and written tasks in one unbroken sitting
the account note is scored against the candidate's own transcript
one live follow-up question asking the candidate to state what changed in the call at the point the consumer asked them to stop calling
timing anomalies between call end and note submission
rotated variants, in which the stop request, the dispute and the hostile opening are reordered between sittings so that a described version of the scenario does not transfer to the next candidate

The log describes what happened. It does not produce a cheating verdict — the follow-up conversation is the control, because a statistical accusation is not something we would ask a reviewer to defend.

What you receive

full call transcript with the disclosure, the dispute and the stop request time-marked
the account note and next-action plan as submitted
the account state at the end of the session, being the dispute marker, the per-medium contact permissions, and the next action scheduled with its channel and date
the six judgment answers with their stated operative rules
per-criterion score with the excerpt that earned it

Who decides

Required. Two decisions must be made by a person and cannot be delegated to the score. The first is whether a given sentence threatened an action not intended to be taken, which is a content judgment about a specific sentence against a specific client brief, and a reviewer makes it in seconds from a surfaced excerpt. The second is what to do with a candidate who scored 1 on the stop-request criterion but well on everything else, which is a policy decision about risk appetite, not a measurement question. The reviewer is looking at three time-marked turns: the opening exchange, the turn after the dispute, and the turn after the stop request. The ranking entitles the buyer to conclude that this candidate did or did not do these four things in this conversation. It does not establish that the candidate holds any licence or registration, which several US states require and which no simulation can verify, and it is not a substitute for post-hire call monitoring.

What this does not measure

Nothing in this design reads accent, dialect, fluency, vocabulary, vocal register or manner of speech, and the anchors are written so that no criterion can be satisfied by sounding calm or authoritative. The riskiest criterion for that leakage is the disclosure one, because a fluent speaker can deliver a formula more smoothly, so the anchors were written around completeness, placement and repetition rather than delivery, and a 5 is reachable by a candidate who speaks slowly or hesitantly. The assessment does not measure collection rate, call duration, or any productivity metric, and a deployer who reintroduces those alongside it will restore exactly the incentive the design is testing resistance to. It deliberately does not measure resilience to abuse as a construct in its own right: the AI consumer is hostile because that is what makes the disclosure fork real, not because tolerating hostility is being scored, and the scenario contains no personal abuse of the candidate. The judgment task is text-heavy and time-boxed, which is the most likely source of a construct-irrelevant disadvantage for second-language readers and for candidates with dyslexia; extended time should be available on request without a diagnosis, and deployers should compare judgment-task scores against call-task scores by group, since a gap that appears only in the written half is the signature of a reading-speed effect rather than a competency difference. Prior third-party collections experience will advantage candidates on the judgment items in a way it does not on the call, which is worth watching if the buyer intends to hire people new to the industry.

This is the one seat in the corpus where a single sentence, spoken in the wrong order on a single call, is a compliance event with a private right of action attached, and where the person most likely to say it is a competent and motivated agent nine minutes into a call that is finally going well. The design is built around that timing rather than around rule knowledge, because rule knowledge is not the scarce thing. Every candidate for this seat has passed a compliance module. The module is where the problem starts.

The stop-request task exists because the standard training is wrong in a specific, teachable, dangerous way. Modules teach 15 U.S.C. 1692c(c), under which a cease-communication notice must be in writing, and agents come away believing a spoken request can be worked around. Regulation F closed that: 12 CFR 1006.14(h) prohibits communicating through a medium the person has asked the collector not to use, with no writing requirement, so a spoken instruction to stop calling ends telephone contact at the turn it is given. An agent carrying the statute-only belief will pass any quiz you set and will fail in production in their first month. Putting the request at the point where a payment is one question away is what turns that belief into an observable behaviour, and the observation is binary: the reviewer reads the next turn.

The dispute task is designed so that both reflexes are wrong, which is unusual and is why it earns its weight. The reflex trained by production is to treat a dispute as an objection and rebut it, and the reflex trained by an anxious compliance culture is to promise that everything stops. Neither is right. Under 1692g(b) the cease-collection duty is triggered by a written dispute inside the validation period; telling a consumer on the phone that collection will now stop when it will not is itself a false representation about the legal status of the debt. The 5 anchor therefore requires the candidate to say the accurate thing, which is more demanding than either shortcut and which almost nobody does.

The consequence criterion is included because the failure it catches is not the one buyers picture. Agents do not threaten arrest. They speak in the passive, in the conditional, about options being looked at, and let the consumer supply the worst case themselves. That is why the 3 anchor describes the conditional-mood version rather than treating it as partial credit toward a 5, and why the criterion cannot be scored without a client brief that states what the portfolio actually does. A reviewer with the brief and the excerpt makes this call immediately; a scoring model without the brief cannot make it at all, which is the clearest reason human review is required here rather than recommended.

One boundary is worth restating on the page, because it is the question buyers in this market ask first. Several US states require debt collector licensing or registration. Savvanta ranks handling ability and never verifies a credential, and any deployment must keep licensure as a separate employer-owned gate. What the assessment adds is the thing the licence does not cover: the licence proves somebody sat an exam, and post-hire call monitoring catches the breach after a real consumer has already experienced it. This design is the only point in the process that observes the fork before a consumer is on the other end of it.

Sources

Every figure on this page is traceable. Where a claim could not be sourced it is stated qualitatively instead.

US Bureau of Labor Statistics, Occupational Outlook Handbook, Bill and Account Collectors, 2025, https://www.bls.gov/ooh/office-and-administrative-support/bill-and-account-collectors.htm
Consumer Financial Protection Bureau, Fair Debt Collection Practices Act: CFPB Annual Report 2025, November 2025, https://files.consumerfinance.gov/f/documents/cfpb_fdcpa-2025-annual-report_2025-11.pdf
15 U.S.C. 1692g(b), duty to cease collection on a written dispute until verification is obtained and mailed, https://www.law.cornell.edu/uscode/text/15/1692g
15 U.S.C. 1692e(11), required disclosure that the communication is from a debt collector attempting to collect a debt, and 1692e(5), threats of action that cannot legally be taken or is not intended to be taken, https://www.law.cornell.edu/uscode/text/15/1692e
15 U.S.C. 1692c(a)(1) and 1692c(c), convenient calling hours of 8am-9pm local time and cessation of communication on written notice, https://www.law.cornell.edu/uscode/text/15/1692c
12 CFR 1006.14(b)(2)(i), Regulation F presumption of compliance at no more than seven calls in seven consecutive days and none within seven days of a telephone conversation, https://www.law.cornell.edu/cfr/text/12/1006.14

See what the employer actually receives. A full report for one role, with every score shown beside the excerpt that earned it, conduct findings reported rather than averaged, and a reviewer sign-off required before any decision. No form.

Read a sample report Or talk to us about this role