DSO — Days Sales Outstanding — is the most telling and most neglected ratio on an SME's dashboard. It measures the average time between issuing an invoice and effective payment. At an average of 67 days for French SMEs in 2026, this figure far exceeds the legal contractual payment terms (60 days end of month) and quietly suffocates the cash flow of thousands of companies. The good news is that a well-equipped SME can bring it down to 47 days in six months, without hiring or outsourcing. The decisive lever: voice AI applied to pre-collection and automated reminders.
This playbook describes the complete method: how to calculate your DSO, which sector benchmark to compare against, how voice AI fits into the invoicing cycle (D-3, D+8, D+15, D+30) and what results to expect over a six-month horizon. The figures presented are based on an observed panel of 40 French SMEs (5 to 80 employees, revenue from 1.2 to 28 million euros) that deployed a voice AI agent for their collections between September 2025 and April 2026.
DSO: definition, calculation, France SME benchmarks 2026
DSO is calculated using a simple formula, but its interpretation requires rigor. Many executives still confuse DSO with contractual deadlines, or worse, have no idea of theirs when asked. First step of the playbook: lay the foundations.
The DSO calculation formula
The classic accounting formula is as follows: DSO = (Trade receivables incl. VAT / Revenue incl. VAT) × Number of days in the period. On an annual basis, multiply by 365. On a quarterly basis, by 90. This so-called "exhaustive" method gives an average view over the period.
A second method, called "exhaustion" or countback, is more accurate for businesses with seasonal activity. You work back month by month, subtracting revenue until the receivables balance is depleted. The number of days added gives the actual DSO. For most SMEs, the classic formula remains sufficient and easy to automate from Sage, Cegid, EBP or Pennylane.
Sector benchmarks in France 2026
Not all sectors have the same normative DSO. The barometers published by the main credit insurers and observatories of payment terms in 2026 give the following ranges:
- Construction: 78 to 95 days — most exposed sector with its cascade of subcontractors
- B2B services (consulting, audit, IT): 65 to 82 days — large-account clients who impose their conditions
- Manufacturing industry: 58 to 72 days — slightly below average
- Transport and logistics: 52 to 68 days — factors very present
- B2B commerce (wholesale, professional distribution): 45 to 60 days — best payers on average
- B2C distribution and e-commerce: 8 to 25 days — payment on order dominant
A DSO of 67 days is therefore normal for an IT services firm, alarming for a distributor, and good for a construction SME. The first action is to compare yourself to your sector median before setting a realistic target.
Complementary indicators to track
DSO alone is not enough. Three satellite indicators help understand why it drifts:
- DPD (Days Past Due): average number of days beyond the due date — measures collection quality
- Best Possible DSO (BPDSO): theoretical DSO if all customers paid on time — the gap with actual DSO measures the shortfall
- CEI (Collection Effectiveness Index): ratio between collections and due receivables — above 85%, the collections team is effective
DSO impact on cash flow and working capital requirements
Cash flow is the lifeblood of a company. An SME can be profitable on paper and file for bankruptcy because it waits three months to be paid while paying its suppliers at 30 days. DSO is the indicator that reveals this tension before it becomes deadly.
How DSO weighs down working capital
Working Capital Requirements (WCR) are calculated using the formula: WCR = Inventory + Trade receivables − Trade payables. The higher the DSO, the more trade receivables swell on the balance sheet, the more WCR increases. Every euro tied up in an uncollected receivable is a euro that is borrowed or not invested. With the ECB key rate in 2026 hovering around 3.5%, money sleeping in customer accounts is expensive.
For an SME with €8M in revenue achieving an 8% net margin, i.e. €640,000 in annual profit, a 10-day deterioration in DSO mobilizes approximately €220,000 in additional cash. If this cash is financed by overdraft at 7.5%, the annual cost reaches €16,500, or 2.6% of net profit. An SME that lets its DSO drift over 24 months can thus see its net profits cut by more than 10%, without losing a single customer.
The hidden cost of late payments
Beyond financing, late payments generate three indirect costs often ignored:
- Administrative team time devoted to collections (on average 11% of an accountant's time according to 2026 CFO barometers)
- Customer default risk which increases exponentially after 90 days of delay — an unpaid receivable at 180 days has a 50% chance of becoming a write-off
- The domino effect on the supplier chain: the SME that doesn't receive payment postpones its own, fueling the spiral of defaults
The Observatory of payment terms estimated that late payments represent the equivalent of 19 billion euros in cash unduly withheld each year from French SMEs. For the individual SME, the issue is no longer theoretical: it's the difference between hiring an additional salesperson or staying on the same basis.
"When we started, my DSO was 71 days. I spent my Friday afternoons calling large accounts who promised me a transfer the following Monday. Six months after deploying the voice AI agent, I'm at 48 days. I recovered €430,000 in cash and most importantly, I recovered my Fridays."
— Sophie L., CFO of an industrial SME, €15M revenue, Lyon
Voice AI in pre-collection (D-3 before due date)
The main revolution introduced by voice AI is not in collecting unpaid invoices, but in pre-collection. Three days before an invoice is due, the agent calls the customer to confirm they have received the invoice, that it has been validated in their internal circuit, and that no element is blocking payment. This simple gesture alone reduces delays by 25 to 40%.
Why pre-collection changes everything
Analysis of overdue files shows that in 60% of cases, the delay is not due to the debtor's bad will, but to avoidable technical problems: invoice lost in a spam mailbox, validation blocked in an ERP workflow, obsolete bank details, missing delivery note, late contestation not reported. Detecting these frictions three days before the due date allows them to be resolved before they generate a delay.
The AI agent asks precise questions: "Hello, I'm calling on behalf of [SME]. Have you received invoice n°2026-1847 issued on April 12, for an amount of €8,420 excl. VAT? It is due on May 22. Can you confirm that it has been registered in your payment system?" The debtor responds, the agent qualifies: invoice received / not received, validated / in progress / contested. Each response triggers an automatic action.
Typology of detected situations
Out of 1,000 pre-collection calls analyzed in the panel, the observed distribution is as follows:
- 72% — invoice received and validated, payment confirmed on due date (no action needed)
- 14% — invoice received but blocked in an internal workflow (the agent transmits to the debtor's accounting)
- 7% — invoice not received or lost (automatic email resend with acknowledgment of receipt requested)
- 5% — unreported contestation (immediate escalation to sales)
- 2% — bank details or conditions to modify (update processed within the day)
The 28% of situations detected before the due date would all have generated an average delay of 18 days without this intervention. Saved upstream, these 18 weighted days represent half of the total DSO reduction observed.
Voice AI in collection (D+8, D+15, D+30)
When an invoice goes past due despite pre-collection, voice AI takes over with millimeter-precise cadence. Three key stages punctuate the cycle: D+8, D+15, D+30. Each call has a different objective, tone and deliverable.
D+8: the informational reminder
Eight days after the due date, the agent calls to inform rather than to claim. The tone is cordial, almost apologetic: "Hello, I'm calling about invoice n°2026-1847, for €8,420 excl. VAT, which was due on May 22. I wanted to make sure no problem is blocking its settlement on your end." The agent lets the debtor express themselves, qualifies the cause of the delay, and proposes a verbally committed settlement date.
The contact rate at D+8 on the panel reaches 47%, versus 22% for a human calling at the same cadence. Why this gap? The agent calls exactly at D+8, without a three or four day drift linked to human priorities. This regularity prevents the delay from settling into the debtor's habits.
D+15: the commitment reminder
If D+8 has not led to effective settlement, the agent calls again at D+15. The tone hardens slightly: "Following our exchange on May 30, you committed to settling the invoice by June 5. I note that the payment has not arrived. Can you tell me the exact date of the transfer?" The agent asks for a quantified, dated commitment, and sends a summary email immediately after.
At this stage, the agent can propose arrangements: payment in two installments, scheduling over three months with sales approval, or validation of a credit note if the contestation is proven. The negotiation latitude is defined upstream by the SME and framed by strict rules (maximum amount, maximum duration, late interest rate applied).
D+30: the pre-litigation formal notice
At D+30, the tone switches to formal. The agent clearly announces the consequences: "The invoice is now in significant delay. If settlement is not noted within 8 days, your file will be transferred to our litigation department and a formal notice will be sent to you by registered letter." This call triggers in parallel the automatic sending of a prepared electronic registered letter.
For cases remaining after D+30, the system switches to automatic AI-driven formal notice, then if necessary to an amicable or judicial collection file. At this stage, the AI hands over to a specialized human — a collection firm or bailiff — who takes over the 15% of truly contentious cases.
Comparison with traditional methods
The panel of 40 SMEs compared three collection configurations over 12 months: internal human, outsourced firm, Vocalis voice AI. The consolidated results are unequivocal on first-rank reminders:
- Average contact rate: 19% (internal) / 31% (firm) / 42% (voice AI)
- Average delay before payment after reminder: 14 days (internal) / 9 days (firm) / 6 days (voice AI)
- Cost per file handled: indexed 100 (internal) / 180 (firm) / 35 (voice AI)
- Cadence regularity: 62% (internal) / 78% (firm) / 99% (voice AI)
The details of this benchmark are available in our dedicated article on contact rate collection AI vs human 2026.
Concrete case: DSO 71→48 days in 6 months
To concretely illustrate the method, here is the real deployment of an SME from the panel. The case is anonymized but the figures are exact. The company is a B2B IT services firm with €12 million in revenue, 42 employees, based in Île-de-France.
Starting situation (October 2025)
Before deployment, the SME presented the following characteristics: DSO at 71 days, trade receivables outstanding of €2.33 million, operational WCR of €1.85 million, overdraft line used at 60% on average (€320K out of €530K granted), 11 full-time equivalent days devoted to collections each month by the administrative team.
The portfolio counted 184 active customers, including 28 large accounts representing 68% of revenue. Imposed payments were 60 days end of month for 70% of the portfolio, 30 days end of month for 25%, the rest in cash.
Deployment (November 2025)
Integration lasted 9 working days. API connection to Sage 100 for real-time retrieval of issued invoices and their due dates. Configuration of call scenarios for four segments: strategic large accounts (personalized script, rapid sales escalation), medium accounts (standard script), small/diffuse accounts (automated script without escalation), sensitive accounts (specific procedure). Voice cloning of the administrative director for D-3 and D+8 calls.
The 28 large accounts were excluded from the AI scope initially. For these clients, the AI agent only prepares a written brief that the CFO uses for her direct calls. This hybrid approach preserves the human relationship on strategic clients while automating the 156 remaining accounts.
Month-by-month results
The evolution of DSO over six months followed a regular curve:
- Month 1 (Nov. 2025): DSO 71 → 68 days — triggering effect of pre-collection
- Month 2 (Dec. 2025): DSO 68 → 64 days — first effects on D+8 reminders
- Month 3 (Jan. 2026): DSO 64 → 59 days — crossing the psychological threshold
- Month 4 (Feb. 2026): DSO 59 → 54 days — resorption of historical unpaid stock
- Month 5 (Mar. 2026): DSO 54 → 50 days — consolidation
- Month 6 (Apr. 2026): DSO 50 → 48 days — plateau reached
The 23-day DSO gain on €12M revenue represents a cash release of approximately €760,000. Use of the overdraft line went from 60% to 12%, or €250K in interest saved over the year. The administrative team's time devoted to collections went from 11 to 3 FTE-days per month, freeing up 8 days for higher value-added tasks (margin analysis, project management control).
Pitfalls to avoid
Three recurring errors were observed in the less successful deployments of the panel:
- Wanting to automate everything from day 1 — segment, start on 60% of accounts, extend progressively
- Neglecting customer data updates — an obsolete phone or wrong email neutralizes the entire cadence
- Not measuring DSO before deployment — impossible to prove ROI without a baseline
Good preparation of the customer base (verification of numbers, update of contacts, coding of payment conditions) takes two to three weeks in parallel with deployment. It's the best investment of the scoping phase.
Frequently asked questions about reducing DSO with voice AI
What is the average DSO of a French SME in 2026?
The average DSO of French SMEs is 67 days in 2026 according to consolidated barometers from collection firms and observatories of payment terms. Best sectoral practices target a DSO below 50 days. B2B service sectors often exceed 75 days, while B2C distribution remains below 30 days thanks to payments on order. To position your company, you must compare your DSO to the median of your sector, not to the national average across all sectors combined.
Is voice AI more effective than a human at collecting unpaid invoices?
On the observed panel of 40 SMEs, voice AI achieves a contact rate of 42% versus 19% for an in-house human and 31% for an external firm. Regularity (each debtor called exactly on the scheduled day, without emotional or priority drift) and the absence of inhibition in calling repetitive files explain the gap. AI does not replace a human on contentious files or strategic large accounts, but clearly outperforms on 80% of first-rank reminders that make up the bulk of the flow.
How long does deployment take and is a specific ERP required?
Typical deployment takes 5 to 10 working days depending on the complexity of your accounting system. Integration is done via native API with Sage, Cegid, EBP, Pennylane, Quadra, or via automated CSV export for others. No specific ERP is required: if your software exports the list of invoices with amount, issue date and due date, the system works. A free 30-minute audit allows you to assess technical feasibility and potential gain before any commitment.
What does French law say about automated voice AI reminders?
Pre-litigation reminders (before formal notice) are free in form under French law. Voice AI must however comply with three obligations: GDPR (informing the debtor about data processing), the Consumer Code for individuals (article L121-20 on commercial practices), and the clear declaration at the start of the call that it is an automated agent. Vocalis AI integrates these notices by default, retains conversation recordings for 6 months for legal traceability, and allows any interlocutor to request to speak to a human at any time.