The premium that never arrives

TL;DR

In Argentine insurance, the bad ending of the collections story is not always "the customer refused to pay". Part of the arrears is manufactured by the process: the debit bounced and nobody retried in time, the broker collected but didn't remit, the employer misdeclared payroll, the receipt never reconciled and the policy got cancelled anyway.

I mapped the market's four collection flows and quantified them with four official sources. The workers' comp system fails to collect $45.3 billion pesos every month (8.36% arrears rate, per SRT, the workers' compensation regulator). For every 100 property & casualty policies issued, 19 are annulled, per SSN, the insurance regulator. The median top-50 insurer carries 2 months of premium locked in receivables, and annulments eat 10.6% of issued premium (SSN financial statements).

The pattern that ties it together: friction grows with intermediation and with how small the payer is. The thesis is that process-manufactured arrears are the cheapest to recover and the most expensive to ignore, and that merely accelerating reconciliation frees months of premium trapped on the balance sheet. I propose where to start, what to build and which metrics should govern it.

How money enters an insurer (four flows, four ways to break)

Insurance collections is not one process: it's at least four, with different actors and different failure modes. I surveyed and diagrammed them before touching a single dataset, because a number without its flow explains nothing.

The original diagram, with the working notes from the research, is navigable in Eraser.

1. Automatic debit (personal lines). The debit attempt bounces (insufficient funds, expired card), there are 2-3 blind retries, and if they fail: arrears → notice → suspension → cancellation. Customers often find out when they no longer have coverage.
2. Broker-mediated collections (PAS). The broker collects from the customer in cash, check or transfer, and remits to the insurer whenever they remit. Reconciling what was collected against the core system is manual, with human error and structural delay.
3. Workers' comp (ART). The employer self-declares payroll through form F.931, the monthly sworn payroll declaration filed with AFIP/ARCA, Argentina's tax agency, that determines social-security charges, including the ART premium. The premium varies month to month with that declared payroll, and reconciling declared vs. collected drags under-declaration, error and arrears.
4. Regulated multi-channel (Mexico, as contrast). Direct debit, bank counter, OXXO, SPEI, card: every channel emits a different receipt, all reconciled by hand against the core. If reconciliation doesn't land in time, the policy is cancelled improperly: the customer had already paid.

The common denominator is that between the money and the system that decides whether a policy lives, there is a manual process. That is what this case sets out to measure.

The numbers: four official sources, one problem

45 billion pesos a month: workers' comp arrears as measured by the regulator itself

The SRT publishes monthly how much every employer in the system should have paid (agreed premium) and how much actually came in (collected premium). In April 2026: $542.2 billion pesos agreed, $496.9 billion collected. The gap, $45.3 billion in a single month, is exactly the "Morosidad: 8.36%" the agency itself reports.

The segment split is consistent with the thesis, though there is an alternative reading these figures cannot rule out: smaller payers, more volatile income. The system splits into two universes: productive units (companies and businesses) and private households (families employing domestic staff).

98% of those arrears are companies. Productive units cover 9.5M workers: they agreed $535.9B, paid $491.5B and left $44.4B uncollected. Private households cover 601k: they agreed $6.3B, paid $5.4B and left $0.9B.

The volume sits on one side and the signal on the other. Private households are 6% of the workers and 2% of the arrears, but they fail at nearly twice the rate: 14.25% against 8.29%. That is the segment with the smallest payer and the most manual process, which is exactly what the thesis predicts. Put per head, the SRT reports an agreed premium of $58,730 pesos per covered worker a month. It is the agency's own indicator and sits on a narrower base than the table, because it leaves out self-insured employers: it does not tie back to the aggregates above, nor is it subtractable against the collections one.

For every 100 policies issued in a quarter, 19 are annulled

The property & casualty market annuls 3.14 million policies per quarter against 16.6 million gross issuances (net issuance plus annulments). That is 19 out of every 100. It comes from the SSN's policies, claims and premiums release for Q1 2026.

The pattern by line of business repeats the SRT's. Motorcycles, the most precarious segment, annul 25.6% of what they issue and 18% of their in-force stock per quarter. Auto, the mass-market line, annuls 1.38 million policies per quarter.

The cost lives on the balance sheet: 2 months of premium trapped

The median top-50 insurer carries the equivalent of 2 months of invoiced premium not yet collected, but the number depends on the sales channel more than on the company: it ranges from 0.3 to 4.6 months depending on how the money is collected. On the balance sheet, premium receivables are money already invoiced that hasn't come in yet; divided by an average month of premium, they show how many months of sales are pending, which is any industry's DSO (days sales outstanding), here expressed in months.

The data comes from the quarterly financial statements insurers file with the SSN (March 2026 close, 177 companies). The same source shows annulments in pesos eating 10.6% of premium, with extremes: Seguros Galicia 24.6%, Nación Seguros 24.5%. That median leaves out ten of the fifty: workers' comp and retirement carriers, where annulment does not exist as a concept because coverage is continuous. Counting them as zero, the median across the full top 50 drops to 8.9%.

At the top of the ranking sits everything intermediated: Assurant (sells warranties through retailers), the cooperatives with broker networks (La Segunda, Cooperación Mutual, Río Uruguay) and the bank and institutional channel (Allianz, Galicia, Nación, Provincia Seguros). At the bottom, direct collection: La Caja with automatic debit (0.9 months) and Provincia ART with payroll-based collection (0.3). With an intermediary in the money flow, premium takes 3-4 months to arrive; without one, weeks. It's flow 2 of the diagram, confirmed company by company in the accounting.

One precision: part of those months is commercial design (installment plans, remittance cycles agreed with the broker), not pathology. The channel gradient tells you where to look, not how much to recover.

The year-over-year comparison (same fiscal month, 2025 vs 2026) is flat: median 2.06 → 1.98 months. The problem is structural, not cyclical.

The thesis

Insurers do have a collections problem, but not all arrears are equal. Part of them is manufactured by the process itself: payments that existed and never reconciled, debits that bounced without a smart retry, cancelled customers who wanted to pay. That part is the cheapest to recover, the most expensive to ignore, and the only one product can fix.

Part, not all: with public data you cannot measure what share of the arrears is genuine (customers who can't or won't pay) and what share the process produces. Every number in this case mixes both. The SRT gap includes real insolvency on top of under-declaration and error; the 18.9% annulment rate includes voluntary churn and technical causes; the months of premium in receivables include commercial credit by design, not just pathology. Claiming "the problem is reconciliation" would be a statement these figures cannot prove.

What the figures do show is where the recoverable value per peso invested is concentrated, for three reasons:

1. Asymmetric error costs. This is reasoning, not measurement: public data does not carry the cost of each type of error. Losing a genuine defaulter costs the premium that never came in and coverage stops, so the damage ends there. Improperly cancelling someone who was paying costs everything that customer was going to pay from now on, plus whatever it takes to replace them. Even if manufactured arrears are a minority of the stock, each case costs a good deal more.
2. The working capital does not depend on how much of the arrears is genuine. The top 50 carries $4.13 trillion pesos in premium receivables ($4,130 billion) against a monthly premium of $2.13 trillion (SSN financial statements, March 2026). Accelerating collection by one month frees ~$2.1 trillion once, and that number depends on no rate at all: it's money that was arriving anyway, arriving earlier. Holding the stock, on the other hand, does cost: every point of monthly funding cost is ~$0.5 trillion a year, so each company plugs in its own real rate and gets its own figure. All of this holds even if every peso of arrears were genuine.
3. Tractability. A customer's insolvency cannot be fixed with software; retry timing, receipt matching and self-service payment updates can. As a reference for the recoverable order of magnitude, in digital subscriptions 20% to 40% of churn is involuntary, a benchmark from another industry, useful as an order of magnitude rather than sector data.

Prioritization: where I would start

Four flows, one team: you have to choose. My criterion is volume × automatability × the channel's political friction.

1. Automatic debit, first. Maximum volume (the auto line alone issues 7 million net policies per quarter), zero humans with their own incentives in the middle, and the intervention is pure product: retries with smart timing, multi-channel dunning, self-service payment-method updates. It's the flow where software alone can move the needle.
2. Brokers, second, and as a product for the broker, not against them. It's the most expensive flow (3-4 months of premium trapped at intermediated carriers) and the hardest: the broker is a channel with interests, not a process. The play is not to disintermediate but to give them automatic remittance: make remitting stop being work.
3. Workers' comp, third. The arrears are huge in pesos but the flow depends on employer self-declaration and the tax agency; the lever is early detection of under-declaration, not classic dunning.

What I would build: the collections agent

Three concrete design decisions, each with its why:

• Retries that know when there's money. The blind 48-hour retry is the policy of someone who doesn't look at data. The right timing can be learned: salary-deposit dates, the customer's own bounce history, day of the month. Goal: maximize first-retry approval, not retry count.
• Dunning that offers a way out, not a threat. Between the bounce and the suspension there's a window that today is used to notify and should be used to resolve: a WhatsApp message with a payment link and self-service payment-method update. Cancellation becomes the end of a path with exits, not a timeout.
• Reconciliation where humans only review exceptions. Automatic matching of receipts and remittances against the core; people stop hunting for matches and start resolving what the system flags. Improper cancellation (cancelling a paid policy) becomes an auditable event, not a statistical accident.

How I would govern it

The North Star: % of issued premium collected on time without human intervention. It measures value delivered (the money that comes in and the manual army no longer needed), it's a leading indicator of the financial result, and every flow feeds it with inputs the team can move:

With two guardrails so the metric can't be optimized by breaking something else: complaints about improper charges (the goal is collecting smarter, not harder) and cost per peso collected. Plus a warning about the metric usually crowned in collections: recovery rate rewards the flow failing so it can be recovered. Recovery is an input; the star is the payment going through the first time.

Sources

• SRT: Coverage and financing, latest data (April 2026)
• SSN: Policies, claims and premiums (Q1 2026)
• SSN: Open data, insurance market indicators (2009–2026)
• SSN: Open data, insurer financial statements (March 2025 and March 2026 closes)
• GoCardless: Involuntary churn (subscriptions benchmark)