Written for collections teams and any outbound operation measured on RPC. The levers in order of impact, the compliance walls around them, and how to prove lift with your own data.
How do you improve right party contact rate?
Fix the data before the dialer: score every number, keep skip traces current, and correct phone-type flags. Then test calling windows inside Reg F and state limits, protect your caller ID from spam labels, and keep wrong-number dispositions out of your connect counts. Run one controlled change per week and read results at account level. Data quality moves RPC more than any dialer setting.
Right party contacts divided by connects, plus a per-account view, and a formula everyone can see.
Ask five collections operations to define right party contact rate and you will get five different formulas. The one we hold clients to is simple: right party contacts divided by connects. A connect is a call answered by a live human being. Not dials, not ringouts, not voicemail systems. If answering machine hits leak into your denominator, your rate is understated and every timing analysis built on it is standing on sand.
Run a second view alongside it: right party contacts per account per month. The rate view tells you how efficient each conversation opportunity is. The per-account view tells you whether the portfolio is actually being reached, which is the thing your client, or your board, is really paying for. A floor can hold a flattering RPC rate while quietly failing to touch half its accounts, and the per-account view is what catches it.
Owning the definition matters more than it sounds. If one team counts a spouse who takes a message as an RPC and another does not, month-on-month comparison is fiction, and sooner or later someone prices a portfolio or staffs a shift off that fiction. Write the formula down: which disposition codes form the numerator, which system events form the denominator, what gets excluded. Publish it where the floor can see it, and when you change it, restate history. A number in Friday's deck that cannot survive the question 'connects by whose definition' is decoration, not a metric.
Number scoring, skip-trace recency and phone-type flags outrank any pacing tweak.
Most RPC problems are list problems wearing a dialer costume. When we audit a collections floor with a sliding contact rate, the pacing settings are rarely the story; the story is numbers that stopped belonging to the debtor months ago and nobody noticed. Three levers do most of the work, in rough order of return:
US operations have one more tool worth the admin: the FCC-established Reassigned Numbers Database at reassigned.us. Query a number with the date you last verified it and the database reports whether it has since been permanently disconnected, which is exactly the 'right number, wrong human' failure that produces polite strangers, wasted attempts and TCPA exposure. It is not a substitute for skip tracing; it is a cheap filter in front of it.
All of this belongs inside a written waterfall: which number gets tried when, what evidence retires a number, and how wrong-number feedback flows back to your data vendor. We cover the mechanics in our outbound list strategy guide; the short version is that a list with a memory outperforms a list with a refresh button.
Our free 17-question Five9 Health Check covers list hygiene, pacing, caller ID reputation and disposition mapping, the same levers this article walks through. Ten minutes to complete, and you get a straight scorecard back.
Find your portfolio's patterns inside the windows the rules allow.
Every collections manager has a theory about the best time to call debtors, and plenty of conference decks will sell you a universal magic hour. The honest answer is that timing patterns are real, but they are yours. They differ by debt type, employment mix, region and placement age, and a pattern lifted from someone else's portfolio transfers about as well as their staff rota would.
The workable method is unglamorous. Build an attempts-by-outcome matrix, hour of day against day of week, split by your two or three biggest segments. Set a minimum cell size before you read anything into a cell, and hunt for lift against your own portfolio average rather than against a published best hour. On the floors we run, the differences between segments within one portfolio are frequently bigger than any across-the-board time-of-day pattern, which is exactly why borrowed benchmarks mislead.
Timing tests live inside a regulatory box, and the box is smaller than the clock suggests. Regulation F presumes that calls before 8 a.m. or after 9 p.m. at the consumer's location are inconvenient (12 CFR 1006.6), and separately presumes compliance at no more than seven calls within seven consecutive days per debt, with no call within seven days after a telephone conversation about that debt (12 CFR 1006.14). Both presumptions are rebuttable, which cuts both ways. States layer their own rules on top: Massachusetts, for one, caps collection contact at two communications per seven-day period under 940 CMR 7.04, and counts texts toward it. Operators typically derive the dialing window from area code plus mailing address, take the stricter reading when the two disagree, and cap attempts per debt rather than per phone number. These rules get amended and enforcement priorities move, so build cadence logic against the current text rather than a blog post, and involve qualified counsel where the exposure justifies it. Our Regulation F dialing guide works through the configuration detail.
Your best-scored number dialed at the ideal hour still fails if the handset says Spam Likely. Right parties screen calls harder than third parties do, so labeling does not just cut your connect rate; it skews the residual answers toward exactly the people you did not want. The RPC damage is done before your agent hears a click.
Carrier analytics engines label numbers on volume patterns, answer and complaint behavior, and signing attestation, and they judge each number on its own record. A collections shop that rotates DIDs weekly to outrun labels is teaching those engines to distrust everything it owns, because fast rotation reads as a scam signature, not a remedy. The durable program is duller: fewer numbers, sane daily volume per DID, numbers registered with the analytics providers, display monitored across the major carriers, and remediation filed whenever a label is wrong. We wrote up the full sequence in our spam-likely remediation guide, including which registrations actually move labels.
One measurement note. When labels bite, RPC per connect can hold steady while RPC per account collapses, because the calls still being answered come from your least-screened contacts. If the rate looks fine but coverage is dying, check your caller ID display before you blame the list.
Answering machine detection is a bet. You spend a small share of live answers, misread as machines and dropped or greeted with a beat of dead air, to save agents from listening to voicemail greetings. On a fat, fresh consumer list the bet can pay, because live answers are plentiful and agent minutes are the scarce commodity.
Collections lists invert that math. A deeply worked portfolio is thin: the accounts still in play answer rarely, and a meaningful share of the answers that do come are the right party you have been chasing for weeks. Every AMD false positive on that list destroys the exact event the whole operation exists to produce, and the person you just hung up on is now warier of the next attempt. No detector is perfect, whatever the datasheet implies, and the cost of its errors scales with how precious each live answer is.
Our working rule from the floors we run: the thinner and higher-value the segment, the less AMD it should carry. Turn it off, or run a mode that passes uncertain answers to an agent, on late-stage and high-balance work, and accept that agents will eat more voicemail greetings. Reserve full AMD for early-stage, high-volume passes where connects are cheap. The Five9 AMD guide covers the specific modes and where each belongs.
RPC is a voice metric, but a voice-only cadence fights with one hand behind its back. The supporting channels do two jobs: they make the eventual call expected, and they give the right party a lower-friction way to surface themselves.
Letters still earn their postage in collections. So does the limited-content message, Regulation F's prescribed voicemail: a business name that does not indicate debt collection, a request for a return call, the name of a person or people the consumer can ask for, and a callback number, with nothing beyond the short list of optional extras the rule itself permits. It is classed as an attempt to communicate rather than a communication, but the CFPB's own FAQs confirm those calls still count toward the seven-in-seven call frequency presumption. It is a way of using your permitted attempts better, not a way of getting more of them.
Text is the sharpest tool and the easiest to cut yourself on. Under Regulation F, every electronic communication needs a clear and simple opt-out method (12 CFR 1006.6(e)). Under the TCPA rules at 47 CFR 64.1200, autodialed or prerecorded texts to mobile numbers need prior express consent, and the current text requires revocation requests to be honored within a reasonable time not exceeding ten business days. Consent provenance in collections is frequently murky, portfolios change hands, and this corner of the rules has been amended repeatedly through 2025, so verify against the live regulation and take proper advice before a single message leaves the platform. Operators who do have that consent foundation typically use text sparingly and specifically, for example a message the day before that names the number which will call tomorrow. On the floors we support, that expectation-setting does more for next-day answer behavior than any script change we have tested, though we will not dress the observation up with a percentage.
Five common corruptions and the cheap audits that catch them.
Before you credit any initiative with an RPC lift, make sure the number itself is telling the truth. The most common corruption is the wrong-number disposition. Handled inconsistently, it hits you twice: agents who code wrong numbers as vague no-contacts leave dead numbers in the pool to be redialed, inflating connects and sinking the rate, while agents who over-use the code suppress numbers that were actually right. Either way you end up retuning timing to fix a hygiene problem.
| Trap | What it does | Fix |
|---|---|---|
| Voicemail counted as a connect | Inflates the denominator, understates RPC, and sends timing analysis chasing ghosts | Define connect as live-answer events only and audit how the dialer maps AMD results to dispositions |
| Wrong-number codes never fed back | Dead numbers get redialed for weeks; connects rise, RPC falls, agents get blamed | Auto-suppress the number on a confirmed wrong-number code and trigger a re-trace on the account |
| Third-party contacts coded as RPC | Flatters the rate and hides a reach problem | Tighten the numerator definition and QA a monthly sample of RPC dispositions against recordings |
| Counting per call on multi-number accounts | Double-counts contact on the same account and overstates coverage | Report the per-account view deduplicated by account and day |
| Definition changed mid-quarter | The trend line becomes fiction precisely when someone is judging it | Version the definition and restate history on every change |
A monthly disposition audit, thirty recordings pulled at random and checked against their codes, costs an afternoon. The setups we audit that skip it are usually the ones most certain their RPC number is accurate.
Everything above only compounds if the floor can test it without borrowing a data science team. The loop below is deliberately small enough to survive contact with a busy month.
Two or three honest cycles a month is enough. Floors that plateau on RPC have usually stopped testing rather than run out of ideas, and the plateau arrives quietly: the same cadence, the same windows, the same slowly staling data, all defended by last year's test results. None of this is glamorous. In our experience it is the entire difference between floors that lift RPC and floors that hold meetings about it.
Right party contact rate is the share of connected calls answered by the person you are actually trying to reach: right party contacts divided by connects, where a connect means a live human answered. Track it alongside right party contacts per account per month, which shows portfolio coverage. Write down exactly which disposition codes count in the numerator and which events count as connects, because inconsistent definitions make the trend meaningless.
The honest answer is that no reliable public benchmark exists, because the rate depends heavily on debt age, placement history, data quality and how each shop defines a connect. Published figures rarely disclose their definitions, which makes comparison misleading. Benchmark against your own baseline instead: fix the definition, measure a clean month, then judge every change against that. A rising trend on a stable definition is worth more than any borrowed number.
There is no universal best hour, and anyone selling one is guessing with confidence. Timing patterns are real but portfolio-specific, varying by debt type, employment mix and region. Build an hour-by-day outcome matrix from your own attempts, segment it, and test shifts against your own average. All of it sits inside the rules: Regulation F presumes calls before 8 a.m. or after 9 p.m. at the consumer's location are inconvenient, and some states restrict further.
Regulation F creates presumptions rather than a hard cap. A collector calling no more than seven times within seven consecutive days about a particular debt, and not within seven days after a telephone conversation about it, is presumed compliant; exceeding those thresholds is presumed a violation, and both presumptions are rebuttable. Some states are stricter, for example Massachusetts caps contact at two per seven-day period. Check the current rule text and take advice from qualified counsel for your own program.
Yes, provided recency and feedback are managed. A number traced eighteen months ago is a guess, so re-trace on triggers such as wrong-number dispositions, returned mail or long no-answer streaks rather than on a fixed calendar. Score vendors on verified right-party hit rate, not on the volume of numbers returned. US operations can also query the FCC-established Reassigned Numbers Database to catch numbers disconnected and reassigned since they were last verified.
Sparingly. AMD misreads some live answers as machines, and on a thin, heavily worked collections list every live answer is scarce and disproportionately valuable, so false positives destroy the exact events you need. In our experience the thinner and higher-value the segment, the less AMD it should carry: off or agent-pass modes for late-stage and high-balance work, full AMD only on early-stage, high-volume passes where connects are cheap.
Check the denominator before the strategy. Voicemail systems counted as connects, wrong numbers left in the dialing pool, and third-party conversations coded as right-party contacts all move the rate without anything real changing. Spam labeling produces a subtler version: RPC per connect can hold while RPC per account collapses, because the remaining answers come from your least-screened contacts. Audit dispositions against call recordings monthly before retuning timing or pacing.
Only on a solid consent and compliance foundation. Regulation F requires a clear and simple opt-out in every electronic communication, and TCPA rules at 47 CFR 64.1200 require prior express consent for autodialed texts, with revocations honored within at most ten business days under the current text. Consent provenance is often murky on purchased portfolios, so verify before sending. Where the foundation exists, a day-before text naming the number that will call tends to lift next-day answers.
We run collections and outbound floors on Five9 daily, as a dedicated pod rather than a ticket queue. Bring your RPC trend and your list setup, and we will tell you honestly which lever is worth pulling first.
Talk to us