Case studies

Measurable outcomes, not promises.

Each case documents the real problem, the technical solution, the stack used, the metrics achieved and the team assigned. If your challenge looks like one of these, let's talk.

Industry
Pharmaceutical distribution
Year
2024

Digital transformation of Aceso, a pharmaceutical distributor

100 %
Billing automated
IA
Purchase forecasts
Tiempo real
Risk monitoring
Reactiva → Predictiva
Operation mode

The challenge

Distributor with manual admin processes, intuition-based purchase decisions and reactive risk monitoring. The operation was growing but visibility over where risk lived and when to restock relied on spreadsheets and personal experience.

The solution

Three-layer integrated platform: automated billing connected to the ERP, an AI model for purchase forecasting (anticipates SKU- and supplier-level demand with statistical confidence) and a real-time risk monitoring dashboard with alerts and automated course-correction proposals.

Stack

Pythonscikit-learnPostgreSQLn8nFastAPIGrafanaDocker

Timeline & team

Multi-phase program · senior squad + client team

“Numoru led the digital transformation of Aceso. They integrated billing, built AI-driven purchase forecasting and deployed risk monitoring with real-time course correction. We moved from reactive to predictive operation.”

Industry
Fintech
Year
2025

73 % fraud reduction for LATAM fintech with real-time ML

−73 %
Fraud (90 days)
−74 %
Chargebacks
+$2.1M
Annualized impact
8 ms
p99 latency

The challenge

Fintech processing 2 million monthly transactions. Fraud represented 4.2 % of volume, more than double the industry benchmark. Manual rules blocked legitimate transactions (18 % false positives) while letting sophisticated patterns through.

The solution

Three-layer detection system: feature engineering (147 transactional, device, behavioral and network features), XGBoost model (AUC-ROC 0.96, p99 latency 8 ms) and a decision system combining rules, model and human review with adaptive thresholds.

Stack

PythonXGBoostKafkaPostgreSQLRedisDockerKubernetes

Timeline & team

4 months · 2 senior engineers

Read full case →
Industry
Insurance
Year
2025

Over 12,000 monthly hours automated for a regional insurer

12 000 h
Hours / month saved
< 90 días
Positive ROI
85 %
Processes automated
0
Layoffs

The challenge

Insurer with repetitive manual processes for quoting, policy validation and regulatory reporting. Admin team spent ~12,000 hours/month on tasks analysts described as "non-value-adding".

The solution

Intelligent automation platform combining RPA, OCR (structured PDF extraction), NLP models for claim classification and n8n orchestration. Human-in-the-loop approval workflow for gray-area cases (confidence < 0.85).

Stack

PythonspaCyTesseract OCRn8nPostgreSQLRedisDocker

Timeline & team

6 months · squad of 3 + client analyst

Industry
Marketplace / e-commerce
Year
2024

6× faster to market: marketplace moves from 6-week releases to daily releases

6×
Faster to market
< 5 min
Median deploy time
< 4 %
Rollback rate
99.95 %
Uptime

The challenge

Marketplace growing 80 % YoY stuck on monthly / bi-monthly releases. Each deploy was an 8-hour manual event with ~30 % rollback rate. Product velocity was limited by infrastructure velocity.

The solution

Migration to cloud-native architecture on Kubernetes with full CI/CD (GitHub Actions + ArgoCD), feature flags (GrowthBook), observability (OpenTelemetry + Grafana) and automated quality gates (tests, performance, security scan).

Stack

KubernetesGitHub ActionsArgoCDGrowthBookOpenTelemetryGrafanaTerraform

Timeline & team

5 months · squad of 4 + client platform team

Questions about the cases

Why are client names omitted?+

Most projects are under NDA. We share metrics, architecture and stack because those are verifiable. Names and references are provided under NDA in formal proposals.

How were these metrics calculated?+

Each metric was calculated by comparing pre-implementation state against the state 90 days post-production, using the same KPIs declared in the proposal. Metrics are auditable internally by the client.

Do these cases translate to other industries?+

Yes. ML detection applies to fintech, e-commerce, telco and insurance. Intelligent automation applies to any high-volume admin process. Cloud-native migration applies to any digital product whose velocity is blocked by infra.

How long does a similar case take?+

Between 4 and 6 months to reach production with measured KPIs. Validated POCs are delivered in 6-8 weeks.

Your case here in 6 months?

Tell us your challenge. The discovery call is free and no-commitment.

Start a conversation