Document processing and customer operations AI for ACMES, a Mexican insurance broker.
ACMES operates across complex commercial insurance lines and customer-facing service channels. Volume is high. Documents are dense, unstructured, and policy-specific. Customer operations scale with book size, not with headcount availability. Every workflow touches sensitive data, and every AI decision must be defensible under CNSF oversight and internal audit standards.
Before Saptiva AI, ACMES, like most enterprises in the region, had evaluated hyperscaler-based AI solutions and found the same gap: the technology could do the work, but nothing got it to production.
ACMES approached Saptiva AI with two concrete, measurable bottlenecks and one non-negotiable condition.
Document processing at volume. Policies, endorsements, and claims documents were being manually reviewed and keyed into downstream systems. The work was accurate, but slow. Each additional line of business meant more headcount.
Customer operations at scale. First-tier inquiries - policy questions, coverage clarifications, claims status - required a growing team of human agents. Response times lengthened as book size grew.
The condition: AI had to run under ACMES's compliance framework, not despite it. Data could not leave approved boundaries. Every decision had to be auditable. Any vendor relationship had to satisfy internal risk review before a single document passed through the system.
The deployment is deliberately small - two applications, one orchestration engine, one embedded Forward Deployed Engineer. Not a platform rollout. Not a multi-year transformation. Two workflows, in production, fast.
Extracts, classifies, and structures the data contained in ACMES's policy, endorsement, and claims documents. Outputs flow into the existing systems of record. Exceptions and low-confidence cases route to human reviewers with the model's reasoning attached.
Handles first-tier customer inquiries across the channels ACMES already operates. Resolves the routine. Escalates the complex, with full conversation history and context, to the appropriate human agent. Respects compliance, tone, and disclosure requirements by design.
Sits between both applications and the compute environment ACMES approved. Writes an immutable audit record for every decision. Data flows reviewed by ACMES compliance.
Data and inference remained inside the boundary ACMES's compliance framework permits.
Each workload enters through a Saptiva Studio application carrying its metadata. frIdA checks that metadata against the configuration ACMES set - residency, data class, model eligibility, audit requirements - then dispatches to the approved compute. Every step is logged.
The configuration lives in ACMES's own repository, reviewed by ACMES compliance, versioned like any production configuration. Every decision is on the record. This is the difference between “AI that passed review” and “AI that is continuously reviewable.”
A Saptiva AI Forward Deployed Engineer landed on day one inside the ACMES team, physically, operationally, and organizationally. Not on a call. Not on a ticket queue. In the room.
Discovery, scoping, environment provisioning, the first workflow running, compliance review, and production cutover all took place inside fourteen days against a signed SLA. The FDE stayed for a capability-transfer period after cutover. They left when the ACMES team could operate, extend, and audit the applications without us. Not before.
Consistent with ACMES's disclosure standard, outcomes are described qualitatively. Specific metrics, volumes, and internal KPIs are not disclosed on this page. They remain ACMES's to share on their terms.
Analyst capacity redirected toward exception handling and complex cases.
Human agents focused on escalations with full AI-prepared context.
Compliance and risk teams hold an immutable record of which workload ran where, under which configuration, against which model.
The deployment is now operated, extended, and governed by ACMES, not by a vendor dependency.
Customer quotes are published with explicit written approval, on the customer's timeline, in language the customer's communications team has cleared. We do not draft quotes for our customers. The disclosure posture below is what serious buyers read first anyway.
Several categories of information are deliberately absent from this case study. They are not missing by oversight. They are withheld out of respect for ACMES's confidentiality obligations to their own clients, and to the regulators who oversee them.
ACMES serves specific named organizations under their own confidentiality agreements. We do not name those organizations on this page, in sales conversations, or in any public material.
Volumes, throughput rates, error rates, cost savings, time savings, and internal KPI movement are ACMES's to disclose. Not ours.
Specific business rules, pricing logic, exception-handling policies, and organizational routing are intellectual property belonging to ACMES, not to Saptiva AI.
Sophisticated insurance and financial services buyers recognize this posture as the same standard they hold their own vendors to. A case study that claimed to disclose every detail of a major customer would be less credible, not more.
Mexico's largest private AI lab in higher education, won over global solution partners on architecture and execution speed.
Read the deployment →KYC, conversational agents, and credit origination. Put the group's stalled pilots into production.
Read the deployment →If you operate in insurance, banking, or any other industry, and you're accountable for a workflow that needs production AI under a real compliance framework, a Forward Deployed Engineer will respond within 48 hours.