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 Retention and Cancellations Advisor
Save rate is the wrong metric to hire on and the failure is mechanical rather than moral. A save rate counts cancellations averted this month; it does not net off the customer who took a discount and left in month four, the mis-sold upgrade that becomes a complaint, or the customer who was never going to leave and has just been given money. Hiring on it selects precisely for the agent who is best at converting a plain cancellation instruction into a longer conversation, which is the behaviour the FTC has pursued under ROSCA and Section 5 of the FTC Act and which state automatic-renewal laws also target. It is worth being current here rather than lazy: the FTC's Negative Option Rule (the "click to cancel" rule) was vacated in its entirety by the Eighth Circuit in July 2025 and the Commission reopened rulemaking with an advance notice in 2026, but ROSCA at 15 U.S.C. 8403 — which requires simple mechanisms for a consumer to stop recurring charges, and clear and conspicuous disclosure of all material terms before billing information is obtained — is a statute and was never touched by the vacatur. No interview observes whether this candidate honours a clear cancellation instruction or discloses the post-promotional price, and both are single lines in a transcript.
Retention and cancellations are frequently written as two job titles and are almost always one seat, because the cancellation request is what routes the customer to the retention desk in the first place. Treating them separately in a corpus would produce two pages describing the same call from opposite ends, so this file covers both and the aliases carry both title sets. The call has a structure that no other role in this hub shares: the customer has already decided, the advisor is paid to change that decision, and the legally protected act — I want to cancel — is the same sentence as the commercial trigger. Every other adversarial conversation in customer operations lets you separate the compliance fork from the sales fork. This one does not. They are the same fork.
The commercial skill is diagnosis before offer, and it is where the median and top quartile diverge most visibly. The median advisor hears "cancel," reaches for the retention offer matrix, and leads with a discount. The top quartile spends ninety seconds finding out why, because the four common reasons want four different responses: a customer leaving over a service failure wants the failure acknowledged and fixed and is often insulted by money; a customer leaving over price wants a cheaper plan, which may mean downgrading them and losing revenue on purpose; a customer leaving because a competitor has something you do not needs an honest answer about whether you will ever have it; and a customer leaving because they can no longer afford it should, in most cases, be helped to leave. Discounting all four identically produces a save rate that looks good for a quarter and a base full of customers on legacy pricing who churn the moment the promotion lapses.
The conduct fork is narrower and sharper. There is a point in the call at which the customer has given a clear, unambiguous instruction to cancel — not a grumble, not a threat, an instruction. Everything after that point is observable and gradeable. Did the advisor process the cancellation, or ask one more question? Was the retention offer presented with its full material terms, including the reversion price and the date it reverts, or was the headline number left to stand alone? Was the customer told what they lose and when service ends? ROSCA's requirement of a simple mechanism to stop recurring charges is not satisfied in spirit by an advisor who is simple to reach and difficult to leave, and the FTC has continued to bring negative-option cases under ROSCA and Section 5 notwithstanding the vacatur of the specific rule. In the UK the equivalent pressure comes from the FCA's expectations on vulnerable customers, where an advisor who keeps selling to somebody who has just said they cannot afford the current price is doing something FG21/1 was written to stop.
The third assessable competency is the artefact. The confirmation the customer receives after a save is the document that determines whether the save was real or is a complaint on a delay. It has to state the new price, the term, the price after the term, and what has changed about the service. Advisors who make an accurate offer on a call and then send a confirmation that omits the reversion price have created a mis-selling exposure with a nine-month fuse, and this is extremely common, because the offer is spoken by a person and the confirmation is usually generated by a system the advisor selected an option in.
Savvanta assesses all three in one sequence: a live call with an AI customer whose stated reason for leaving is not their real one, and who at a defined point gives a plain cancellation instruction; then the written confirmation. The scoring is behavioural — was the reason diagnosed, was the offer matched to it, was the instruction honoured at the turn it was given, were the material terms disclosed in both the call and the note. It never scores persistence as a virtue in itself, and it never scores accent, dialect or manner of speech.
What the job actually needs
- diagnosing the real reason for leaving before offering anything
- matching an offer to that reason
- disclosing the full material terms of the offer including what happens when the promotion ends
- executing a clear cancellation instruction without obstruction
- recognising when the customer is leaving because they cannot afford it
How people fail in this seat
- reflexively discounts a customer who was going anyway
- asks a fourth clarifying question after the customer has said cancel plainly
- presents a twelve-month price as the price when it reverts after three
- invents a retention benefit that operations cannot deliver
- treats affordability-driven cancellation as an objection
What most employers do instead
A CV screen and an interview built around the candidate's save rate at their last employer, sometimes a sales-style role-play.
The assessment
About 30 minutes end to end.
The systems it runs in
A subscription billing and management system alongside the contact record: the plan and its current price, a promotional price with a term and a reversion price and date, the offers and credits an advisor can apply, the downgrade and pause options, the cancellation action with its end date and final-bill treatment, and the save or cancellation reason code. Chargebee describes this shape — the full subscriber lifecycle including trials, upgrades, downgrades and cancellations, with the cancellation flow itself treated as a retention surface. The confirmation the customer receives is generated from what the advisor set in that system, which is exactly why the second task exists: a spoken offer and a booked offer can differ, and only the booked one is what the customer is billed.
- Chargebee
- Zuora
- Salesforce Service Cloud
Any recurring-billing or subscription system with a plan, promotional pricing that reverts, an offer or credit action, a cancellation action with an end date, and reason codes — the seat exists wherever a business bills on a recurring basis, so the same shape covers telecoms, utilities, insurance, media, software and subscription retail with different labels on the fields. Rebuilt against the buyer's own offer matrix and cancellation journey where they supply them, since the reversion prices in the fixture are the part a buyer will want to be theirs.
What the candidate actually does
| Task | What happens |
|---|---|
| The customer who says the wrong reason first live_call · 14 min | An AI customer calls to cancel a recurring subscription. Their stated reason is price. Their real reason is a service failure three weeks ago that nobody called back about, and it emerges only if the candidate asks an open question and then a second one. Money does not fix it; the AI customer is faintly insulted by a discount offered before the failure is acknowledged. The candidate's brief contains a retention offer with a promotional price, a term, and a materially higher reversion price and date, and it also states that the team's headline metric is save rate and that the candidate is two saves below target this week. That last line is deliberate and identical for every candidate: the cheap path has to be genuinely rewarded for the task to measure resistance to it. At a fixed point around minute nine, whatever has happened, the customer gives a plain, unambiguous instruction to cancel today. |
| The confirmation the customer actually receives written_artifact · 8 min | What the candidate set in the subscription system for whichever branch the call ended in, and the written confirmation generated from it. If the customer stayed, the confirmation must carry the promotional price, the term, the exact reversion price and the date it applies, and what has changed about the service. If the customer cancelled, it must carry the end date, what happens to the final bill, and what the customer loses and when. Because the document is generated from the record, the report can set three things beside each other — the terms stated on the call, the offer or cancellation actually booked, and the confirmation the customer holds — and a spoken reversion price that never reached the record is visible as a gap rather than inferred from a complaint nine months later. The temptation is a warm two-line note, which is what the majority of real confirmations are, and which is where a spoken offer quietly becomes a mis-selling exposure with a nine-month fuse. |
| Five save-desk decisions judgment_scenario · 7 min | Five short extracts, each stopping at a decision point, answered in one or two sentences. A customer says they can no longer afford the service. A customer says cancel and then keeps talking about the problem, which is genuinely ambiguous and where the right answer is to confirm once rather than to relitigate. A colleague's practice of routing cancellation requests to a callback queue. An offer whose headline price is accurate and whose reversion price sits on a second screen the customer is not shown. A customer who is not cancelling at all but complaining, where an offer would be money given to somebody who was staying. |
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.
Diagnoses the real reason before offering anythingweight 0.2Asks an open question, notices that the stated reason and the account history do not fit, asks a second question that surfaces the failure, and names …
Matches the response to the reason that was actually foundweight 0.15Leads with what will be done about the failure, by whom and by when, and if money is offered at all it comes second and is framed as compensation for …
Honours a plain cancellation instruction at the turn it is given (15 U.S.C. 8403(3), requirement to provide simple mechanisms for a consumer to stop recurring charges)weight 0.25The turn after the instruction begins the cancellation, states the end date and what happens to the final bill, and contains no further offer.
Discloses all material terms of any offer on the call, before agreement (15 U.S.C. 8403(1), clear and conspicuous disclosure of all material terms before obtaining billing information)weight 0.2States the promotional price, the length, the exact reversion price and the date it takes effect, unprompted, before asking the customer to agree, and…
The written confirmation carries the same terms as the call (15 U.S.C. 8403(1))weight 0.2Every term stated on the call appears in the confirmation with the same numbers, and anything the customer loses is named alongside the date it goes.
How it is scored
Weighted mean of the five criteria, each scored 1 to 5 against the anchors, reported with the time-marked excerpt or the document line that earned each score. The report explicitly does not include whether the customer was saved. That omission is the design, and it should be explained to the buyer before the first sitting, because a buyer who reintroduces the save outcome as a tiebreak has inverted the instrument and is once again selecting for the behaviour the assessment was built to detect.
Integrity
- monitored session, call and written confirmation in one unbroken sitting
- the confirmation is scored against the candidate's own transcript, so a template cannot fit both branches
- one live follow-up question asking the candidate what the customer's real reason for leaving was and what in the call told them
- offer variants rotated so the reversion price and date differ between sittings
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 cancellation instruction and every price statement time-marked
- the written confirmation as submitted, alongside the branch the call ended in
- the subscription as the candidate left it, being the offer booked with its term and reversion date, or the cancellation with its end date, and the reason code selected, with the interval between the cancellation instruction and the action being taken
- the five judgment answers
- per-criterion score with the excerpt that earned it
Who decides
Required for one criterion and recommended for the rest. The cancellation instruction criterion turns on whether a given sentence was an instruction or a grumble, and although the scenario plants an unambiguous one, candidates sometimes receive an earlier ambiguous statement as an instruction and act on it, which is not a failure and should not be scored as one. A reviewer reads the time-marked turn and confirms or overrides with a written reason. The reviewer is also the person who decides what a 1 on that criterion means for this employer, which is a conduct risk decision and not a measurement one. The ranking entitles the buyer to conclude that this candidate diagnosed or did not diagnose the reason, disclosed or did not disclose the reversion price, and honoured or did not honour the instruction, in this call. It does not predict save rate, it does not establish that the candidate will behave the same way under a live commission scheme, and it is not a compliance certification for the employer's own cancellation journey, which is a process question rather than a hiring one.
What this does not measure
The design does not read accent, dialect, fluency or manner of speech, and no anchor can be met by sounding persuasive. It deliberately does not score persistence, talk time, call duration, objection count or save outcome, all of which are the standard measures in this seat and all of which reward the behaviours ROSCA exists to constrain. The one place manner could leak back in is the second criterion, matching the response to the reason, so its anchors are written around what was offered and in what order rather than around how the acknowledgement sounded. A known and honest limitation: candidates who have worked in save desks trained on obstruction will score badly on the cancellation criterion for reasons that are habitual rather than dispositional, and that habit is retrainable. A deployer using this as a hard cut will reject people who could be good in a better-run operation, and should decide deliberately whether that is what they want rather than discovering it from the pass rates. The written task grades content, not spelling, punctuation or idiom. Deployers should monitor score distributions by hiring channel and language background, and should watch for a gap that appears only on the written confirmation, which would indicate a literacy effect rather than a disclosure-discipline difference.
This call has a structure no other role in the hub shares: the legally protected act and the commercial trigger are the same sentence. Everywhere else in customer operations the compliance fork and the sales fork can be separated and scored independently. Here the customer says they want to cancel, and that is simultaneously the thing the advisor is paid to prevent and the thing the advisor is required to act on. The assessment is built on that collision rather than around it.
The legal ground is worth stating precisely, because the market has been confused about it since 2025. The FTC's Negative Option Rule, the click-to-cancel rule, was vacated in its entirety by the Eighth Circuit in July 2025 on the ground that the Commission had not conducted the required preliminary regulatory analysis, and the FTC reopened rulemaking with an advance notice in March 2026. None of that touched ROSCA, which is a statute. 15 U.S.C. 8403 still requires clear and conspicuous disclosure of all material terms before billing information is obtained, express informed consent before charging, and simple mechanisms for a consumer to stop recurring charges. The two rubric criteria that carry the most weight, the cancellation instruction and the disclosure of reversion price, map directly onto the first and third of those, and both are single lines in a transcript.
The commercial half is where median and top-quartile advisors diverge most visibly, and it is not about persuasion. The median advisor hears cancel, reaches for the offer matrix and leads with money. The strong advisor spends ninety seconds finding out why, because the common reasons want different responses and a discount is the right answer to only one of them. The scenario therefore separates the stated reason from the real one, and makes the real one a service failure, which is the case where money is not just ineffective but mildly offensive. A candidate who leads with a discount has not been rude or non-compliant; they have simply done the thing that produces a save rate that looks good for a quarter and a base full of customers on legacy pricing who churn the moment the promotion lapses.
Putting the save-rate pressure into the candidate's brief is the choice that makes this design work, and it is the choice most likely to be questioned. A scenario in which nothing is at stake for the candidate does not test resistance to anything; it tests whether they know what the right answer looks like, which is what the compliance quiz already establishes. The brief is identical for every candidate, the pressure is stated in one sentence, and the outcome of the call is not scored, so the incentive it creates is entirely one-directional: it makes the cheap path attractive without making it rewarding.
The confirmation task exists because this is where a good call goes wrong after the fact. An advisor can state the promotional price, the term and the reversion price accurately out loud, and then send a confirmation generated by selecting an option in a system, which carries the headline number and nothing else. The customer's evidence of what they agreed is the document, not the call, and the complaint arrives when the price reverts. Scoring the confirmation against the candidate's own transcript catches the gap directly, and it is the one part of this assessment that no live role-play with a team leader has ever produced.
What the design refuses to measure is the number this seat is universally hired on. Save rate counts cancellations averted this month. It does not net off the customer who took a discount and left in month four, the upgrade that becomes a complaint, or the customer who was never leaving and has just been given money. Hiring on it selects precisely for the advisor who is best at converting a plain cancellation instruction into a longer conversation. The report from this assessment does not contain it, and a buyer who adds it back has rebuilt the problem.
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, Customer Service Representatives, 2025, https://www.bls.gov/ooh/office-and-administrative-support/customer-service-representatives.htm
- 15 U.S.C. 8403, Restore Online Shoppers' Confidence Act, requirements for negative option features including simple cancellation mechanisms and clear disclosure of material terms before obtaining billing information, https://www.law.cornell.edu/uscode/text/15/8403
- Gibson Dunn client alert, FTC Restarts Negative Option Rulemaking After Eighth Circuit Vacatur; Enforcement Under ROSCA Continues, 2026, https://www.gibsondunn.com/ftc-restarts-negative-option-rulemaking-after-eighth-circuit-vacatur-enforcement-under-rosca-continues/
- Financial Conduct Authority, FG21/1 Guidance for firms on the fair treatment of vulnerable customers, 23 February 2021, https://www.fca.org.uk/publications/finalised-guidance/guidance-firms-fair-treatment-vulnerable-customers
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 reportOr talk to us about this role