Context: the process that was always slow by design
Over 5% of insurers' annual revenue worldwide is lost to fraud. In Latin America, roughly 85% of fraud cases come from the auto line. On top of that direct cost sits the process's own operational cost: in Mexico, a major-medical claim could take 15 to 45 business days from report to payment — with a team of adjusters manually reviewing documents, verifying coverage in separate systems, and chasing the insured to complete the file.
Seguros Nova Mexico is a niche insurer with three product lines: major medical (GMM), auto, and life. With 240,000 active policies, its claims team processed 1,800 claims a month on average. 65% were low-amount, high-frequency claims — hospitalization expenses with standard invoices, documented minor collisions — that consumed the same operational time per file as complex cases.
Technical leadership had the diagnosis since 2023: the process was designed for the worst case (a fraudulent claim with incomplete documentation) and applied that level of friction to every case, including ones that were clearly legitimate, well-documented, and had a clean history.
The challenge
Data scattered in silos. A claim's information lived in four separate systems: the policy system (legacy AS/400), the CRM, the adjuster system (a shared spreadsheet), and the hospital or shop's billing portal. None talked to another without manual intervention.
Volume choking complex-case operations. With 65% of the team's time going to low-amount claims with complete documentation, adjusters had no capacity to deeply analyze the complex cases that really required judgment.
CNSF regulation with maximum response times. CNSF sets maximum deadlines for claim resolution. Seguros Nova was formally compliant, but internal timelines left the insured in weeks of limbo while the file waited in the review queue.
Fraud undetected in the flow. Without automatic pattern analysis, fraud detection was reactive — happening at the end of the process, after the file had already consumed hours of manual review.
The solution: end-to-end automation in five stages
Seguros Nova implemented an intelligent automation pipeline covering the full claim cycle — from the insured's first contact to payment or rejection — with human oversight at high-risk decision points.
Stage 1 — Automated FNOL (First Notice of Loss). AI-powered chatbots handle first notice of loss, answer frequent questions, and guide users through filing claims, available 24/7. The insured reports the claim via WhatsApp or portal, in natural language. The conversational agent extracts structured data from the report, validates the policy in real time against the policy system, confirms active coverage, and generates the claim number with the pre-populated file — before any human has touched the case.
Stage 2 — Document intake and classification (OCR + IDP). The AI combined a chatbot for initial data collection, OCR for automatic document data extraction, and integration with the internal management system, cutting file-opening time from days to under 48 hours. The insured uploads invoices, the accident report, the medical opinion, or damage photos from their phone. The IDP system automatically classifies each document by type, extracts key fields, validates them against the file's data, and flags inconsistencies for review before proceeding.
Stage 3 — Coverage verification and risk scoring. With the file complete, the decision engine automatically evaluates three things in parallel: coverage verification, anti-fraud score, and document completeness. Machine-learning models identify patterns associated with possible fraud, enabling early intervention and generating million-dollar savings by preventing false claims. The fraud score doesn't block — it prioritizes: high-score files go to human review with the analysis already done; low-score ones pass directly to the authorization flow.
Stage 4 — Automatic resolution or smart escalation. Claims meeting all three criteria (verified coverage, low fraud score, complete file) go straight to payment authorization with no human intervention. Flagged ones route to the adjuster with the pre-analyzed file, the explained risk score, and the recommended action. AI agents can escalate to a human when appropriate, providing the information and recommendations needed to make the best decision.
Stage 5 — Automatic payment and communication. Payment authorization automatically generates the bank instruction, notifies the insured (WhatsApp, email or app), and logs the file with full traceability for CNSF audit. If the claim is rejected, the system generates the rejection letter citing the policy basis, and the record stays available for appeal.
Results at 10 months
| Metric | Before | After |
|---|---|---|
| Average resolution time (low-amount claims) | 18 business days | 1.8 business days |
| Average resolution time (complex claims) | 34 business days | 12 business days |
| Full automation rate (no human intervention) | 0% | 58% of claims |
| Documents correctly extracted by OCR/IDP | — | 91% with no manual correction |
| Fraud detected before payment (vs. later audit) | 22% of cases | 71% of cases |
| Adjuster team capacity (same headcount) | 1,800 claims/month | 2,950 claims/month |
| Insured CSAT on claims process | 61 points | 84 points |
The indicator that most changed the internal conversation wasn't speed — it was capacity. In Seguros Nova's case, the efficiency gain freed adjusters to handle 63% more volume with the same headcount — no hiring, no payroll scaling, no degraded review quality on cases that needed it.
What made the difference
Automation by risk segment, not one-size-fits-all. The most common mistake in similar projects is automating every claim with the same rules. Seguros Nova segmented from the start: low-amount, high-frequency claims with standard documentation (full-automation candidates) vs. complex cases requiring human judgment (assistance candidates). The 58% full-automation rate is the result of that segmentation — not of assuming AI can resolve every case.
The adjuster as validator, not processor. The change freed managers to focus on the most complex cases and on the customer relationship. Adjusters stopped spending hours verifying whether a hospital invoice had the right format, and started dedicating that time to cases where their judgment actually mattered.
Traceability as a byproduct of the process. Every step of the automated flow is logged with timestamp, document source, risk score and decision made. The next CNSF audit won't require the team to prepare documentation — the system generates it in real time as a result of daily operation.
The context that makes it urgent in Mexico
47% of Latin American insurers say they haven't been able to fully integrate AI due to data problems (information silos, incomplete or unreliable data). 29% attribute it to legacy systems and decision-making complexity. Those two obstacles are exactly the starting point of any claims-automation project — not the technology.
Eight in ten Latin American insurers plan to increase their AI investment in the short term. The differentiation window exists today — but it closes as adoption spreads and fast resolution shifts from competitive advantage to market requirement.
Illustrative case. Seguros Nova Mexico is a fictional company created to illustrate real claims-automation patterns in the Mexican and Latin American insurance sector. Context sources: CNSF, BCG, EY, Shift Technology, Nerds.ai, Flownexion, Inaza — reviewed July 2026.
Qi-VantaAutomation and artificial intelligence for businesses. From discovery to production, with measurable ROI.