Most finance teams track overall DSO and 30-day amicable collection rate. These two indicators are outcomes — not levers. Between open receivables and effective collection runs a precise causal chain: dial → connect → speak to the right person → obtain a promise → see that promise kept. Each link has its rate, its own drift and its own improvement levers. The modern credit manager — whether running a human team, an AI platform, or both — must know these indicators with the same rigor a sales director tracks pipeline.
This article first sets the definitions, then confronts the available 2025-2026 public benchmarks, and finally details the method to measure cleanly and improve concretely. All external figures are tagged as indicative and to verify against your own data.
Precise definitions: RPC, PTP, PKR, containment
The first trap is lexical. Mixing up "answer rate" and "contact rate" leads to erroneous decisions. Here are the operational definitions that enjoy consensus among serious debt collection practitioners.
Right Party Contact (RPC)
RPC designates a call that effectively reached the person responsible for the debt — not a relative, not voicemail, not a wrong number. In B2C, it is the account holder. In B2B, it is the decision-maker authorized to commit the company to a payment (AR manager, CFO, owner depending on size). RPC is calculated as:
RPC = (Useful contacts) / (Attempted calls). A useful contact assumes you obtained at least one substantive sentence from the debtor acknowledging their own identity.
Promise to Pay (PTP)
PTP is the debtor's verbal commitment (recorded or tracked) to settle a specific amount by a specific date — today, within 7 days, in 3 installments, etc. A vague PTP ("I'll see what I can do") must not be counted as such, otherwise your ratios become polluted.
PTP rate = PTPs obtained / RPC. This is the true negotiation performance indicator.
Promise kept (PKR)
PKR measures the conversion between verbal commitment and actual collection. A debtor can promise sincerely and not deliver (aggravated hardship), or promise without intent (avoidance). PKR is the ultimate quality indicator.
PKR = PTPs kept on time / PTPs obtained. Below 55-60%, your promises are low quality; above 75%, your negotiation is solid.
Containment rate (AI)
Containment rate is AI-specific. It measures the share of calls fully handled by AI without human escalation. A containment of 80% means AI resolves 8 interactions out of 10 alone — the remaining 20% go to a human agent.
AI benchmarks 2025-2026
Voice AI agent vendors for debt collection are progressively publishing benchmarks. These figures provide a useful reference point but should be read with three caveats: they are published by vendors themselves (image bias), they often cover North American use cases (different regulatory market than Europe), and they vary widely by debtor sector.
- Containment rate 45-50% — per Retell AI, for reference only, to verify
- RPC uplift of +8 to +12 points vs human team — per Prodigal case studies, to verify
- PTP rate 28-38% on RPC — range reported by US AI platforms, for reference only
- Cost per effective call $0.40 to $1.20 — market order of magnitude, highly dependent on volume and average duration
- 24/7 availability (outside local legal hours) — intrinsic capability, not a "benchmark" per se
Human call-center benchmarks 2025-2026
On the human side, the most-cited European sources are AFDCC (French Association of Credit Managers) and the annual reports of debt collector associations. These studies give orders of magnitude stable over several years.
- B2B Europe answer rate 15-25% — per AFDCC, for reference only
- B2C Europe answer rate 20-30% — per AFDCC, for reference only
- Effective B2B RPC 10-18% — market estimate; the delta with answer rate reflects switchboards, assistants and wrong numbers
- PTP rate on RPC 22-32% — range observed in European amicable collection agencies
- PKR 55-70% — varies widely by debtor profile
- Capacity 40-60 calls/day/agent — with tight discipline and no idle time
- Fully loaded cost per call $6 to $12 — loaded salary + infrastructure + management
Step-by-step measurement methodology
Before comparing AI and human, you must measure cleanly. Here is the standard procedure used by mature collection departments.
Step 1 — Clean the source data
Every attempted call must produce a timestamped record with: case ID, dialed number, technical status (answered, not answered, ring, invalid number), human status (RPC, third party, voicemail, refusal), duration, outcome (PTP, no PTP, refusal, agreed callback). Without this log, any measurement is illusory.
Step 2 — Define window and scope
Choose a window (30, 60 or 90 days) and a scope (portfolio, sector, aging bucket). KPIs only make sense at comparable scope. Comparing one team's RPC on B2C 15 days overdue with another's on B2B 90 days overdue makes no sense.
Step 3 — Compute the cascade ratios
The collection cascade reads top to bottom: Open cases → Attempted calls → Answered → RPC → PTP → PKR → Collection. Each floor has a conversion rate. Isolating the weak floor enables targeted action.
Step 4 — Segment by profile
Same overall averages hide massive gaps. Segment at minimum by: receivable age, amount, debtor sector, last successful contact channel, payment history.
Factors influencing contact rate
RPC is not a constant — it is the outcome of operational choices. Major factors, in decreasing impact order:
Call hour and day of week
For B2B, the 10 a.m. to noon and 2 p.m. to 4 p.m. window Tuesday through Thursday concentrates 60 to 70% of RPCs. Monday mornings and Friday afternoons are generally slow. For B2C, the 6 p.m. to 8 p.m. evening and Saturday morning significantly outperform business hours.
Debtor sector
A small contractor owner is very hard to reach during the day (on site). A services CFO is more accessible. A craftsman is reached in the evening. Adapting hours to the sector sometimes doubles RPC.
Predictive availability scoring
AI platforms train models on each profile's contact history. A debtor already reached at 7:30 p.m. on a Tuesday will be dialed at that time again — not at 10 a.m.
Caller ID used
A number flagged as spam by carriers or by the recipient (via Truecaller, Hiya) sees its answer rate collapse. Number rotation and SIP trunk reputation matter more than people realize.
Conversation opening
The first 5 seconds determine whether the debtor hangs up or continues. A factual, name-based opening ("Hello Mrs. Smith, this is about your file XYZ") significantly outperforms a commercial one ("Hello, I'm calling on behalf of...").
AI vs human call-center comparison table
The following table consolidates the benchmarks discussed above. It is deliberately presented as ranges: no vendor can guarantee a precise figure without measurement on a real portfolio.
| Indicator | Human call-center | AI agent | Source |
|---|---|---|---|
| Answer rate | 15-25% | 35-50% | AFDCC / Retell — indicative |
| Right Party Contact (RPC) | 10-18% | 22-35% | AFDCC / Prodigal — indicative |
| PTP / RPC rate | 22-32% | 28-38% | Case studies — to verify |
| PKR / PTP rate | 55-70% | 60-75% | Market estimate |
| Containment rate | N/A | 45-50% | Retell AI — to verify |
| Daily capacity | 40-60 calls/agent | 500+ simultaneous | Technical capacity |
| Cost per effective call | $6-12 | $0.40-1.20 | Market order of magnitude |
| Conversation quality | Variable by agent | Consistent | Structural |
How to improve your AI contact rate by +20%
Once the voice AI agent is deployed, marginal RPC gains are won through disciplined iteration. The five levers to test in order:
- Predictive scoring by slot — exploit history to call each debtor at their best slot, not an average slot. Typical gain: +6 to +10 points.
- Caller ID rotation — maintain 8 to 15 rotating numbers, monitor spam reputation, replace proactively. Typical gain: +3 to +5 points.
- Schedule diversification — systematically add an evening slot (B2C) or Saturday morning. Typical gain: +4 to +7 points on difficult profiles.
- Automatic disqualification of bad numbers — STIR/SHAKEN detection, directory cross-check, carrier exchange. Typical gain: +2 to +4 points from cleaning the base.
- A/B testing of the first 5 seconds — test 3 to 5 openings, measure continuation rate at T+10 seconds. Typical gain: +3 to +6 points on winning openings.
You only steer what you measure. You only improve what you segment. You only persuade with verified numbers.
Recommended tool / dashboard
A modern collection dashboard consolidates in real time: the RPC → PTP → PKR cascade, breakdown by hour/day/sector, human vs AI channel comparison, and DSO impact projection for the coming weeks. Modern voice AI agent platforms embed this analytical layer; for human call-centers, you generally have to combine routing tool, CRM and in-house BI. See our article Measuring AI agent performance for methodological details.
FAQ
What is Right Party Contact (RPC) in debt collection?
RPC designates a call that effectively reached the person responsible for the debt — not a relative, voicemail or wrong number. It is the most important metric before any negotiation.
Why is the answer rate not the right indicator?
An answered call can reach a third party, a voicemail, an assistant. Only RPC measures whether you spoke to the debtor — a necessary condition for any promise to pay.
What is the average contact rate in B2B debt collection in Europe?
According to AFDCC feedback for reference only, B2B Europe answer rate oscillates between 15 and 25%. Effective RPC is lower — generally 10 to 18%.
How do I improve my contact rate by 20%?
Five levers: predictive scoring by slot, caller ID rotation, schedule diversification, automatic bad-number disqualification, A/B testing of openings.
Are Retell AI and Prodigal benchmarks reliable?
They are figures published by the vendors themselves, so indicative. No external benchmark replaces measurement on your own portfolio over 4 to 6 weeks.
Going further
This article extends our Voice AI and amicable debt collection dossier, our methodological guide Measuring AI agent performance, the Vocalis for debt collection industry page and our About page. For contractual and legal aspects, see our legal notice.