Debt collection is no longer the silent cost centre it once was. In 2026, persistent inflation, deteriorating customer balance sheets and pressure on working capital have made DSO (Days Sales Outstanding) one of the most closely watched indicators in executive committees. The picture is harsh: according to the Atradius 2026 survey, over 38% of French SMEs experience payment delays of more than 60 days, and 1 in 4 SMEs considers unpaid invoices a direct threat to their operations. Debt collection has become a genuine strategic issue, as much as an operational one.
The historic reflex — outsourcing collections to an external agency, or hiring another collections officer — is now reaching its limits. Agencies charge success fees of 12% to 25% on recovered amounts, often damage the customer relationship, and only step in at the pre-litigation stage. Human officers, meanwhile, top out at 40-60 calls per day, with an actual reach rate that rarely exceeds 25%. That is precisely the space voice AI fills: an agent capable of reminding customers from D+1, in their own language, at scale, within a strictly controlled legal framework.
1. Collections in 2026: 38% of SMEs paralysed by unpaid invoices
Debt collection in 2026 looks nothing like it did five years ago. Three structural shifts compound. First, corporate insolvencies have risen since the end of the "whatever it takes" era: the Banque de France recorded more than 67,000 insolvencies over the last twelve months, above the pre-pandemic average. Second, persistent inflation has made inter-company credit more expensive: every extra day of DSO mechanically costs more in financial expenses. Third, finance teams face a chronic shortage of credit management talent.
In this context, DSO becomes a direct proxy for financial health. An SME with €25M in revenue that moves from 65 to 45 days of DSO releases around €1.4M in immediate cash. That cash funds overdraft reduction, investment, or simply holding the line under customer pressure. Conversely, a 10-day DSO drift is enough to turn a profitable SME into one under stress. Collections is no longer a back-office topic, it is a CFO lever.
Why traditional methods can no longer keep up
Legacy tools — Excel, ERP with a basic collections module, automated reminder emails — are running out of steam. Reminder emails have an open rate below 22% and a collapsing reply rate. SMS is read more, but does not allow qualification, negotiation or instant payment. And human calls, despite their effectiveness, are no longer economically viable for small invoices: at €38 fully loaded per outbound call, dunning an unpaid €480 invoice three times already eats 24% of margin.
Meanwhile, litigation services are buried under cases that prevention would have avoided. A 2026 AFDCC study reports that 71% of receivables that go to litigation could have been collected amicably with a reminder within 5 days of due date. The problem is not customer bad faith, but the structural delay in contacting them. That is where an automated late payment recovery for SMEs radically changes the game.
2. Voice AI for collections: what it does legally
A voice AI agent dedicated to collections is not a robo-dialer playing recorded messages. It is a conversational system that answers or places calls, correctly identifies its counterpart, states the reason for the call, negotiates a payment plan or proposes an immediate payment, and closes the exchange with a written summary. All under the supervision of your accounting team, with a complete and auditable call log.
Automatic identification and qualification
Before any call, the agent cross-references your accounting data (invoice, due date, outstanding amount, reminder history) with customer data. On pickup, it checks it is speaking to the right contact (accounting, manager, executive), identifies itself clearly on behalf of your company, then states the reason: "I am calling about invoice no. 2026-0418 for €4,720 issued on 12 March and due on 11 April." The AI is steady, polite, never emotional — a real asset on tough cases.
Real-time negotiation and payment collection
The agent operates with the rules you have set: maximum payment plan of 4 months, minimum first instalment of 30%, permitted early-payment discount, etc. It proposes, listens to the answer, adjusts within thresholds. Once an agreement is reached, the agent sends a Stripe payment link or a GoCardless mandate live by SMS and email. More than 41% of agreements obtained by Vocalis AI in collections lead to payment within 24 hours of the call — a number unreachable for a traditional agency.
Native multilingual
An export-oriented SME's customer book typically contains German, Dutch, Spanish, Italian clients. A French agency will send them letters in French, which will end up in a pile. The Vocalis AI voice AI agent for collections speaks 40 languages natively, with local pronunciation and cultural codes. A Polish customer will be reminded in Polish, by a local voice, with the applicable legal references — a payment driver that multiplies international recovery rates by 2.3.
3. Legal framework: CPCE R124-1 and GDPR applied to collections
The question comes up in every CFO conversation: "is it legal?". The short answer is yes, provided three bodies of law are respected: the French Code of Civil Enforcement Procedures (CPCE), the General Data Protection Regulation (GDPR) and the European AI Act, which came into force in 2026. Vocalis AI applies all three by default, but it is essential to understand what they prohibit and allow.
Article R124-1 CPCE: the hard obligations
The CPCE governs amicable debt collection. The essential obligations, applicable whether the reminder is human or automated, are: the agent must identify itself (creditor company name, agent's name — or the AI), state the amicable nature of the collection, indicate the exact amount due and the breakdown (principal, interest, fees), respect the time slots (between 8am and 9pm, never on Sunday or public holidays), refrain from unlawful pressure or threats. The French Supreme Court confirmed in 2024 that these rules also apply to automated reminders. The CPCE R124 legal framework for AI collections (FR) is now fully scriptable into a voice AI.
GDPR: minimisation and traceability
Voice AI processes personal data (identity, contact details, financial data). It therefore falls under GDPR. The legal basis is generally contract performance or the creditor's legitimate interest. Three key obligations: (1) inform the person, at the start of the call, that the conversation is recorded and conducted by an automated agent; (2) limit the retention of recordings (CNIL recommendation: 6 months maximum for collections calls); (3) allow the person to exercise their rights of access, rectification and objection. Vocalis AI provides a GDPR-ready log exportable on demand.
AI Act: transparency about automation
Since February 2026, the AI Act requires that a counterpart be informed when interacting with an AI system. Concretely, the voice AI for collections must say so explicitly at the start of the call: "Hello, I am the automated assistant of company X…". This transparency does not hurt performance: on the Vocalis AI panel, the acceptance rate after explicit disclosure remains above 87%, because counterparts appreciate the consistency and the absence of emotional pressure typical of a stressed human agent.
4. End-to-end workflow: from DSO D+1 to formal notice and payment
Moving to AI collections is not just plugging a tool into the ERP. It means structuring a continuous workflow, from the day of invoicing to the possible formal notice, with clear escalation thresholds. Here is the standard workflow deployed at SMEs supported by Vocalis AI.
D-3 before due date: courtesy call
Three days before the due date, the agent places a short call (50 seconds on average) to the customer's accounts manager to check the invoice has been received and validated. This call has two virtues: it acts as a psychological nudge ("the invoice is on the radar"), and it identifies blockers upstream (missing PO, dispute on quantities, wrong recipient). Observed stat: 22% of disputes are detected at this stage and resolved with no impact on DSO.
D+1: first amicable reminder
The day after due date, if payment has not been received, the AI sends its first official reminder. Courteous tone, clear identification, reminder of the amount and due date, and an immediate option: wire payment, Stripe SMS payment link, or commitment to pay within 7 days. At this stage, 54% of cases are settled within the next 72 hours. For more on this, read the dedicated guide on reducing DSO with voice AI.
D+8: firm reminder with consequences disclosed
If payment has still not been received, the AI places a second call at D+8. The tone stays professional but firm, the agent recalls Article L441-10 of the French Commercial Code and the calculation of late interest (statutory rate plus 10 percentage points + €40 flat-rate indemnity). This is also when the AI proposes a formalised payment plan if the customer raises a temporary difficulty.
D+15 to D+30: automatic formal notice
Without payment or a payment plan agreement, the AI triggers an automatic AI formal notice sent as an electronic registered letter (LRE) with acknowledgement of receipt. The document is generated from a validated legal template, dated, electronically signed, and kept for the legally required duration. The LRE costs around €2.90, versus €12 for a paper registered letter, with the same legal value.
D+45: escalation or litigation
Beyond 45 days without agreement, the AI hands the case over to human escalation. Your CFO receives a summary sheet with the full history (all calls, transcripts, commitments made, payment links sent, LRE openings). The file is then ready to go to your lawyer or be handed to a bailiff. This is precisely where the collection agency vs AI debate finds its answer: the AI handles 100% of amicable, the agency only takes on tough, well-prepared cases on which its success fee is fully justified.
"Before Vocalis AI, my two collections officers spent 60% of their time on routine reminders. Today the AI handles D+1 to D+30, they only step in on complicated cases. My DSO went from 71 to 49 days in four months. Mathematically, I recovered €1.7M in cash."
— CFO of an industrial SME, €25M revenue, 110 employees, Rhône-Alpes region
5. Real-world case, retail CFO: DSO 72 → 52 days in six months
To anchor the method in the real world, take the case of a specialty retail SME (professional hospitality equipment), €18M revenue, 320 active customers, initial DSO at 72 days, two-person credit management team. The finance department wanted to go below 55 days without hiring. Here is how the deployment unfolded.
Month 0: audit and configuration
Audit of the customer book: segmentation into four tiers (key accounts, recurring SMEs, occasional VSEs, at-risk clients). Definition of reminder rules per tier, time slots, languages to enable (French, Italian, German for border customers), payment plan thresholds (max 3 months, first instalment 40%), and escalation rules. Connection to the ERP (Sage 100) and CRM (HubSpot). LRE templates set up. Total duration: 9 business days.
Month 1: supervised launch
The AI goes live on 30% of the book, in supervised mode: every call is reviewed asynchronously by the accounting lead, who validates or adjusts parameters. Result over the month: 412 calls placed, 287 useful contacts (70% reach rate), 162 payments obtained, 38 payment plans signed. End-of-month DSO: 66 days (-6 days).
Months 2-3: 100% rollout and industrialisation
The AI is extended to the whole book. Human officers focus on escalated cases and strategic key accounts. Over the two cumulative months: 1,940 calls, 1,312 useful contacts, 743 payments, 184 payment plans, 47 automated formal notices sent, 12 files moved to litigation. End-of-month 3 DSO: 58 days.
Months 4-6: stabilisation and fine-tuning
At this stage, the AI has learned the specific book. Scripts are refined by segment, time slots optimised per tier (VSEs respond better between 2pm and 4pm, key accounts between 9am and 11am). Marginal cost per AI call is €0.42, vs. €38 for a human call. End-of-month 6 DSO: 52 days, i.e. -20 days over the period, releasing €985,000 of cumulative cash.
Frequently asked questions from CFOs and credit managers
Is voice AI debt collection legal in France?
Yes, provided that article R124-1 of the Code of Civil Enforcement Procedures (CPCE), the GDPR and the AI Act are respected. The agent must identify itself, state the reason for the call, disclose that it is automated, avoid unlawful pressure, and respect the legal time slots (between 8am and 9pm except on Sundays and public holidays). Vocalis AI is configured by default to block any out-of-scope reminder and provides a call log exportable on demand to the French data protection authority (CNIL).
How many reminders can an AI agent handle per day?
A voice AI agent handles 300 to 500 outbound calls per day in parallel multi-channel, with no fatigue or drop in quality. By comparison, a human collections officer is capped at 40-60 calls per day, with a real reach rate below 25%. The AI can also place reminders 24/7 across international time zones, while respecting locally applicable hours.
Can the AI negotiate a payment plan?
Yes, within the limits you set in advance: maximum duration (3 to 6 months typically), minimum instalment amount, optional early-payment discount. The agent proposes, listens to the counter-proposal, adjusts within thresholds. When an agreement is reached, it sends a Stripe payment link or GoCardless mandate live by SMS. Any threshold breach triggers an automatic escalation to your CFO with a full summary.
What is the typical ROI for an SME switching to AI collections?
Across a panel of 18 SMEs (revenue €5M to €80M) that deployed Vocalis AI for collections between late 2025 and early 2026, the average DSO reduction is 28% (from 71 days to 51 days on average). This releases between €240,000 and €1.8M of cash depending on the size of the receivables book. The total cost is fully amortised within the first two months thanks to the cash released, with no extra hires.