Portfolio

Five practices that turn data into business decisions

GalacticaIA is an integrated, software-agnostic data consultancy. It connects strategy with the platform, models, commercial intelligence, and agentic systems that put each decision into operation.

The point of entry is the problem to be solved. No practice is a prerequisite for another, and each engagement brings in only the cross-cutting capabilities required.

The five practices

01 · Direction, operating model and trust

Data Strategy & Governance

Translation of business priorities into an executable data strategy and the governance model that sustains it. It integrates roadmaps, operating models, ownership, quality, policy and adoption.

Problems it solves

Scattered initiatives, competing metrics, decisions with no accountable owner, and data that cannot withstand scrutiny.

Outcome

Clear priorities, operable accountability, and an investment path tied to business decisions.

The practice draws on the DAMA-DMBOK(opens in a new tab) (DAMA International) and DCAM(opens in a new tab) (EDM Council) frameworks.

Deliverables

Capability assessment and roadmap prioritized by value, risk and feasibility
Data strategy, architecture principles and operating model
Roles, decision forums, policies and ownership for domains and data products
Catalog, lineage, quality and critical definitions in operation
Adoption, training and metrics that track program progress

From priority to operable capability

Strategy and governance: priorities, capabilities, owners and continuity

Priority

What the business needs

Growth and experience
Efficiency and productivity
Risk and trust
New data products

Strategy

Which capability is needed

Prioritized use cases
Target architecture
Roadmap and investment

Operating model

Who decides and delivers

Owned domains and products
Forums with explicit authority
Roles and ways of working

Control

How it stays in place

Quality and critical definitions
Applied, traceable policies
Adoption and progress metrics

Decision traceability

Each priority connects to a capability, an owner, an investment and a measure of progress. Governance operates as the accountability system of this practice.

02 · Data architecture and engineering

Data Platforms & Lakehouse

Construction of the platform where the data resides: layers, sources, ingestion, reconciliation, processing engines and storage, on the existing cloud. The design allows later migrations without rebuilding.

Problems it solves

Isolated sources, fragile pipelines, unpredictable costs, and teams rebuilding the same table again and again.

Outcome

A scalable, observable platform that serves reliable data to analytics, products and AI.

“Many mature organizations suffer from fragmented data repositories. Storing and maintaining those troves can eat up between 15 and 20 percent of the average IT budget.”

McKinsey & Company, Reducing data costs without jeopardizing growth, July 2020.Source

Deliverables

Layered architecture —landing, curation, consumption— with explicit promotion criteria between them
Ingestion and reconciliation of the sources the business actually uses every day
Storage model and table format, chosen for the expected queries
Traceable migration plan from what already exists, without turning off what is in use
Storage and compute cost optimization

The platform, from the inside

Layered architecture: ingestion, promotion between layers and infrastructure

Sources

Core business systems and ERP
Application events and digital channels
Files, spreadsheets and external providers
Documents and unstructured data

The layers, and what it takes to move from one to the next

Landing

Raw

Data arrives exactly as the source has it, untouched. If it has to be reprocessed tomorrow, it is reprocessed from here.

Faithful copy of the source
Full history

To move up: the schema is declared and the load can be repeated without duplicating.

Curation

Reconciled

Names, dates and keys aligned across systems; duplicates resolved by a written rule, not by hand.

Keys reconciled across systems
Types and units normalized
Quality rules that run on every load

To move up: the quality rules pass and there is an owner who answers for the table.

Consumption

Data products

Tables with an agreed definition, ready for the business and for models to use without recomputing them every time.

One definition per metric
Served to dashboards, models and APIs

Who queries it

Dashboards and reporting
Models and experimentation
Applications and APIs

What it runs on

Object storage
Open table format
Query engine decoupled from the data

The data lives in open storage and the engine that queries it is chosen separately: the engine can be swapped without moving the data, which is what prevents lock-in to a vendor. It is also where the bill is decided: how it is partitioned and compacted directly affects the cost of every query.

03 · Models, experimentation and optimization

Advanced Analytics & Machine Learning

Analytical solutions that explain, predict and optimize decisions. The scope spans framing and experimentation through modeling, temporal validation, deployment and monitoring.

Problems it solves

Unreliable forecasts, manual decisions that do not scale, and models that never reach production or cannot be validated.

Outcome

A quantified decision, a model evaluated against a clear baseline, and a mechanism to operate and monitor it.

“On average, only 48% of AI projects make it into production, and it takes 8 months to go from AI prototype to production.”

Gartner, Gartner Survey Finds Generative AI Is Now the Most Frequently Deployed AI Solution in Organizations, May 2024.Source

Team experience · Fintech

Credit scoring for lending, anti-money laundering (AML) detection, churn prediction and revenue analytics.

Deliverables

Forecasting, propensity, classification, scoring and anomaly detection
Optimization models and scenario simulation
Experiment design and causal measurement
Feature engineering, temporal validation and explainability
Deployment, performance monitoring and retraining where appropriate

From problem to operable decision

Model lifecycle: decision, validation and post-launch performance

01 · Frame

Define the decision

Agreement on which action will change, for whom, and under which constraint.

Target and horizon
Error cost and baseline
Success criterion

02 · Model

Build and validate

The model is trained against the past without leaking information from the future.

Reproducible features
Temporal split and backtesting
Explainability and bias

03 · Operate

Prioritize the action

The score becomes a queue, rule, forecast or scenario that someone uses.

Thresholds and constraints
Action and owner
Workflow integration

Validation

Model and business metrics are reported together, compared with a baseline and with the cost of getting it wrong.

Monitoring

Performance, drift, feature quality and action outcomes are observed after deployment; the model is not left behind.

04 · Customers, marketing and digital channels

Customer & Digital Intelligence

Turning customer and digital-channel signals into measurable commercial actions: who to prioritize, what to offer, which experience to fix, and which intervention actually caused a change.

Problems it solves

Segments that never drive action, campaigns measured by correlation, opaque funnels, and competing versions of customer behavior.

Outcome

Growth and retention decisions with audiences, actions, owners, and causal measurement when the case supports it.

“Companies that grow faster drive 40 percent more of their revenue from personalization than their slower-growing counterparts.”

McKinsey & Company, The value of getting personalization right—or wrong—is multiplying, November 2021.Source

Team experience · Fintech

Customer analytics for hyper-personalization and next best offer.

Deliverables

Actionable segmentation: each segment with its behavior, risk, eligible offer and action
Churn prediction trained only on what was known at the time, prioritized by customer value
Next-best-offer, already filtered by who is allowed to receive it and who gave permission to be contacted
Digital funnel and journey with quantified friction points
Campaign incrementality and end-to-end digital ecosystem measurement

A nine-segment grid by risk and value; each cell shows size, churn and expected profit per customer. Only three cells have positive value — Actionable segmentation: nine cells with behavior and economics. Only three pay for the contact; deciding who not to call is part of the deliverable too. Reproducible artifact over synthetic data.See code, data and tests(opens in a new tab)

Digital-origination funnel from product visit to first transaction, quantifying conversion and loss at every step — The funnel locates the largest loss and translates a five-point improvement into additional first transactions. This is a sensitivity scenario over synthetic data, not a client forecast.See code, data and tests(opens in a new tab)

05 · Agentic systems and operable products

Agentic AI & Data Products

Data products and agentic frameworks that bring intelligence into the workflow. Includes the definition of tools, memory, evaluation, observability and authority boundaries so the system is useful and controllable.

Problems it solves

AI prototypes disconnected from the business, agents with unclear limits, and analytics that never enters the workflow where decisions happen.

Outcome

An operable product with users, a data contract, quality criteria, and an explicit boundary between automation and human decisions.

“Over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls.”

Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, June 2025.Source

Deliverables

Data AI-readiness assessment
Data-product design, consumers, contract and service level
Agentic framework with tools, memory, evaluation and observability
Decision flow with human approval and a full trail

Proprietary engine

Nodriza

GalacticaIA's agents are built on Nodriza, its proprietary engine for agents on governed data and process automation.

Governed data: Each agent takes meaning from the data catalog and values from the database, with the permissions of its role.
Traceable figures: Each answer states the query that produced the figure, its glossary definition and what could not be verified.
Agnostic: Database, catalog, industry and language model are interchangeable; agents are configured, not programmed.
Tested: Each version is measured against reference question sets.

Who decides, and who prepares the decision

Agent fleet under human authority

Human authority · outside the automated loop

A person decides

Sets the objective, deliberates with the evidence gathered, and authorizes. No agent coordinates them, and none can bypass them.

Authority boundary · only what needs judgement goes up

Orchestrator

Coordinating agent

Takes the objective, distributes it across the relevant agents, gathers what each contributes along with its source, and synthesizes it into a traceable recommendation.

It neither approves nor executes. What does not need judgement is resolved below and never reaches the person; what does, goes up with the evidence behind it.

Catalog

Inventory agent

Finds what data exists, describes it, and keeps it current when the source changes.

Its sources

Source-system metadata, catalog, schema history

Reports to the orchestrator

Classification

Personal-data agent

Proposes which columns are personal data and of what kind, for a person to confirm.

Its sources

Data samples, known patterns, previously validated classifications

Reports to the orchestrator

Policy

Rule-enforcement agent

Checks actual usage against declared policy and issues an alert when they do not match.

Its sources

Standing policies, access logs, quality and retention rules

Reports to the orchestrator

Trail

What data went in, which agent touched it, with what source, what it recommended and who decided — with a date and an owner, case by case. Without that, the decision cannot be defended six months later, which is when it gets questioned.

Entry case

Data Monetization

Internal monetization, third-party monetization and an agentic layer, built on the five practices of the portfolio.

See Data Monetization

COMMERCIAL EVIDENCE

Cases with impact on decisions

Client work is anonymized and carries no figures that cannot be substantiated. Synthetic artifacts and technical implementations appear separately: they prove method and scope, not a company's results.

Data Platforms & Lakehouse

Telecommunications operator · LATAM

Problem: A SaaS data-warehouse architecture under cost pressure, and a migration that could not lose traceability.
Intervention: Design and delivery of the move to a lakehouse, incorporating lineage and control over the data journey.
Visible deliverable: Target architecture, migration plan, lineage map and cloud-optimization baseline.
Decision or outcome enabled: The organization could move the platform through a verifiable path and reduce the structural cost of its data operation.

Customer & Digital Intelligence

Telecommunications operator · LATAM

Problem: Marketing needed to prioritize audiences through a consistent reading of customer behavior.
Intervention: Statistical user segmentation and translation of the groups into criteria campaign teams could use.
Visible deliverable: Documented segments, assignment rules and an actionable reading for marketing.
Decision or outcome enabled: The segments moved into campaign management instead of remaining a descriptive view.

Data Strategy & GovernanceIn progress

Healthcare organization · Chile

Problem: The new data-protection law requires knowing which personal information exists, why it is used and who is accountable.
Intervention: Design of the governance model, personal-data inventory and accountability for its processing.
Visible deliverable: Personal-data map, ownership, policy and an evidence matrix for operating compliance.
Decision or outcome enabled: An operating basis for privacy decisions with accountable owners and traceable evidence.

TECHNICAL PROOF

Types of evidence: explanatory, reproducible and implemented

The five practice diagrams explain the method. These pieces go further: they allow inspection of code and tests, or define the boundary of a capability already implemented.

Reproducible artifact

Advanced Analytics & Machine Learning

A synthetic fintech suite covering out-of-time churn validation, actionable segmentation, next-best-offer, customer value and campaign incrementality.

Seeded causal data, versioned outputs and 110 tests covering leakage, integrity and reproducibility.

Inspect the lab

Reproducible artifact

Customer & Digital Intelligence

An origination funnel from product visit to first transaction, with loss at each step and a sensitivity scenario around the bottleneck.

ES/EN SVGs from one run; tests bind the step catalog to every synthetic event.

See the funnel case

Implemented capability

Agentic AI & Data Products

A conversational agent over data catalogs with MCP read and write operations, tool traces and confirmation before assets are changed.

Verifiable scope: 27 catalog operations, dry-run mode, human confirmation and an audit log.

See the operating scope

Present a case or take the self-assessment

Case submission

The response states whether the case fits the portfolio, the proposed approach and the requirements on the client side. No prior questionnaire is required.

Present a case

Data maturity self-assessment

Questionnaire on the state of the organization's data. The result shows the maturity level and the initial priorities.

Start the self-assessment