The dilemma between bringing amicable debt collection in-house or outsourcing it to a specialized agency is as old as the credit management function itself. For thirty years, the default answer for European SMBs was outsourcing, for a simple reason: recruiting, training and running a team of collection specialists cost more than it yielded as long as receivable volume did not justify three to four full-time equivalents. Voice AI has shifted this frontier. It is now possible to bring the call reminder function in-house for volumes far below the historical threshold, without adding headcount. That changes the question without eliminating it.
This article offers an updated decision framework: six criteria, three company profiles, and a hybrid model that, in practice, is gaining ground among the most mature finance leaders.
The outsourced model: amicable agency then legal
Outsourced debt collection includes several types of players, often used in cascade: amicable collection agencies, specialized financial intelligence firms, bailiffs (now judicial commissioners in France) for the legal phase, and large integrated European operators.
Typical players in Europe
- GCollect: digitalized amicable positioning, online debtor journey, useful for SMBs with recurring volume.
- Rubypayeur: platform combining prevention, scoring and collection, with a strong data component.
- Intrum: legacy European player, amicable and legal coverage, group and multi-country vocation.
- Eulis and other regional agencies: relational approach, geographic proximity, high-stakes cases.
- Judicial commissioners: legal phase, injunctions, seizures.
The dominant commercial model is commission: a percentage of recovered amounts, typically between 10 and 25% depending on the age of the receivable, the complexity of the case and the volume entrusted. Some players offer hybrid models (subscription plus reduced commission).
Strengths and limits of outsourcing
Agencies bring established legal expertise, litigation capability, proven scoring tools and a "third party" psychological effect useful on the debtor. In return, the company partially loses control of tone, frequency and projected image; it depends on external reporting; and the cost, as a percentage of recovered amounts, can be high on portfolios with a high recovery rate.
The in-house voice AI model
Bringing voice AI in-house means deploying within the company's own information system a platform of conversational agents that place reminder calls, talk with the debtor, propose settlement terms and feed information back to the credit manager. The human team focuses on complex cases, high-stakes negotiations and quality supervision.
Market players include generic voice agent platforms adapted to debt collection and verticalized solutions. Vocalis AI positions itself in the latter category, with a finance and debt collection orientation.
What voice AI brings compared to outsourcing
- Full tone and brand control: the script, the voice and the callback policy are configured by the creditor itself.
- Predictable unit cost: subscription or per-minute model, decoupled from the recovered percentage.
- Responsiveness: campaign launch within hours, real-time script modification.
- Hourly scalability: simultaneous call capacity without hiring.
- Native data and reporting: every call produces a recording, a transcript, a summary, structured metadata directly usable.
Decision matrix: six criteria
| Criterion | Agency outsourcing | In-house voice AI |
|---|---|---|
| Cost | Commission on collection, variable | Subscription/usage, predictable |
| Brand control | Partial, delegated to the agency | Total, script and voice owned |
| Flexibility | Contractual, modification lead times | Real-time modification |
| Compliance | Agency assumes, but principal co-responsible | In-house, to be tooled (telecom rules, GDPR) |
| Reporting | Periodic aggregated reports | Raw data continuously, BI-integratable |
| Scale | Linear with agency resources | Elastic, peaks absorbed without friction |
No column wins on all criteria. The right choice depends on the company profile and the phase of the receivable lifecycle considered.
Comparative ROI calculation (for reference only)
The figures that follow are hypothetical illustrations, not guaranteed outcomes. They serve to structure reasoning.
Illustration 1: average portfolio
An SMB issues 800 invoices per month, of which 5% slip into significant delay (40 invoices) with an average ticket of $1,200, or $48,000 monthly in arrears. With an agency charging 15% on recovered amounts and a 60% recovery rate, the variable cost is roughly $4,320 per month (15% × $28,800 recovered). A voice AI subscription covering this volume has a fixed cost independent of the recovery rate. The crossover point depends on recovery rate and volume: at high recovery rates and stable volume, the subscription becomes economically advantageous.
Illustration 2: volume effect
At 200 invoices in arrears per month with the same ticket, the agency commission cost grows linearly (roughly $21,600 monthly at constant parameters), while the voice AI cost grows much more slowly. The gap widens in favor of in-house above a certain volume threshold.
Small business case: under 10 employees
Typical profile
Low volume (a few cases per month), no dedicated credit manager, priority on simplicity and low fixed cost. Often heterogeneous receivables: a few small cases, occasionally a significant one.
Recommendation: light hybrid model. Keep a success-fee agency for high-stakes cases and legal escalation, without a heavy subscription. Adopt voice AI in self-service or on a pooled platform for standardized amicable reminders if volume justifies it. Full in-house AI is not always relevant under a certain volume threshold.
SMB case: 10 to 250 employees
Typical profile
Significant and recurring volume (several hundred to several thousand invoices per month), dedicated credit manager or accountant, important brand and customer relationship stakes, often a legacy agency in place for years.
Recommendation: bring amicable reminders in-house with voice AI; keep the agency for legal work and complex cases. This setup simultaneously maximizes brand control (the voice calls "on behalf of the company"), cost predictability (subscription) and legal power (agency as backup). This is the sharpest tipping profile on the European market today.
Mid-market case: 250 to 5,000 employees
Typical profile
Very high volume, structured credit management team, sometimes multiple entities with their own flows, strong group reporting requirements, multi-country compliance, ERP integration (SAP, Oracle, Sage), dedicated DPO.
Recommendation: three-tier architecture. In-house voice AI for all pre-legal work, human team for value-added negotiation and structured arrangements, agency for judicial collection and international. AI becomes a productivity tool for the internal team, which retains control of the strategy. This is also the profile where compliance (telecom rules, GDPR) must be handled with the utmost rigor given the volume processed.
The hybrid model: the winning configuration
Beyond size-based cases, one observation emerges: the winning model is almost never a binary choice. It is almost always hybrid, along a time axis:
- Phase 1 (D+5 to D+45) — automated amicable reminders via voice AI. Multiple contacts, scoring, orientation toward settlement options. Strong automation, low cost, high volumes.
- Phase 2 (D+45 to D+90) — supported human negotiation on cases not resolved in phase 1. Internal team or amicable agency, with context enriched by phase 1 data.
- Phase 3 (beyond D+90) — legal proceedings, judicial commissioner, injunction, seizure. No AI substitution, legal expertise indispensable.
This sequence turns the "in-house or outsourced" dilemma into "which phase do I assign to which tool". The best setups articulate all three, with smooth handoffs between stages.
The real question is no longer in-house or outsourced. It is: which phase of the cycle am I handling, and which tool is most effective for this specific phase?
Pitfalls to avoid
- Double billing: never assign the same case simultaneously to AI and to the agency. Define strict handoff rules.
- Forgotten in-house compliance: bringing things in-house means directly assuming telecom and GDPR obligations. These obligations do not disappear with automation. See our dedicated article on voice AI compliance.
- Under-investment in integration: voice AI disconnected from the ERP loses 70% of its value. Budget integration from day one.
- Choice based on headline cost only: true TCO includes integration, training, compliance, maintenance, script evolution. Think in total cost over 24 months.
- Lack of human supervision: AI without a human supervisor drifts. Plan an "AI supervisor" role within the credit team.
Useful internal resources
To go further: our article on voice AI agent for debt collection, our guide on GDPR and voice AI compliance, the debt collection industry page, our About page, our legal notice.
FAQ
Should I bring amicable debt collection in-house or outsource it in 2026?
The decision depends on six criteria: receivables volume, brand stakes, regulatory sensitivity, internal reporting capability, budget and scale horizon. Voice AI has shifted the frontier: in-house becomes accessible to structures that would have outsourced by default three years ago.
Is an agency (GCollect, Rubypayeur, Intrum) more effective than voice AI?
Agencies excel on litigation, legal proceedings and complex receivables. Voice AI excels on high-volume pre-legal work, scoring and standardized reminders. They are not competitors but tools for different phases of the cycle.
Which company size benefits most from in-house voice AI?
SMBs and mid-market firms with recurring volume (starting at a few hundred cases per month) get the best ratio. Small businesses often benefit from a light hybrid model. Large accounts use both in parallel.
Does voice AI fully replace an agency?
No. It replaces the high-volume amicable call function. It replaces neither judicial collection, nor legal support, nor the trust relationship on complex cases. The winning model is hybrid.
What is the typical cost of voice AI versus an agency?
Commercial models differ. An agency charges a commission (10 to 25% of recovered amounts), voice AI charges per subscription or per minute. The comparative calculation depends on recovery rate, average ticket and volume. For a quantified comparison on your portfolio, a free audit is recommended.