A boutique AI consultancy in Madrid. Real innovation, tangible results.
We focus on value — AI is our tool. Cognitiva Labs builds cutting-edge AI and machine learning solutions, untainted by vague promises. Our approach is simple: assess, suggest and refine alongside your business until we land on a solution that adds value.
Companies across pharmaceuticals, aviation, telecom, e-commerce, logistics and the public sector trust us.
Services
From the birth of a concept to its realization, we make sure your company harnesses the true power of today’s technology.
AI strategy
To truly harness the power of AI, a robust strategy is paramount. We start by deeply understanding your business needs, then make sure your data is primed, your infrastructure is ready and your AI team is in place.
Big data
We help your business train AI models built for big data, and design, build and maintain the data pipelines behind them. Successful machine learning starts with the quality of the data, and that is where our expertise truly shines.
Data science
Dynamic pricing for airlines, resource optimization for logistics, causal evaluation of an incrementality campaign: we work across the spectrum, designing, tailoring and validating AI models that drive growth.
MLOps
We deploy and monitor machine learning models so they keep running at peak performance, and your AI initiatives stay efficient and focused on value.
Process
From a two-hour conversation to a solution in production.
01
Value determination
2-hour meeting
Dive deep into your challenges and find where AI can turn them into opportunities.
Make sure the objectives deliver tangible improvements to your operations.
Recommend AI solutions that yield results in the short term.
02
Technology assessment
Before we build
Review your data pipelines to confirm we have all the information the solution needs.
Run a preliminary analysis to gauge how achievable the goal is.
Summarize feasibility, with the goals and objectives rewritten accordingly.
03
Implementation
8+ weeks
Develop with a strong feedback loop, so you validate it with your domain expertise.
Put users first, so the solution is intuitive and relevant to the people who use it.
Deploy a solution tailored to your requirements and goals.
04
Delivery
And beyond
Explain and benchmark the results, with transparency at every step.
Beyond the slide deck: complete deployment setups, so the solution works and scales.
Plan the product’s roadmap, so it keeps delivering value.
Work
Selected projects. Our clients stay anonymous; the problems and results are real.
Forecasting disease incidence for a global pharmaceutical company
A global pharmaceutical company needed to know where and when cases of an infectious disease would rise. We built a system that forecasts cases one and two months ahead in every region, with calibrated uncertainty and early-warning alerts.
Health authorities report cases weekly, but the signals that come before a surge (temperature, rainfall, humidity, vegetation, land use, the El Niño cycle, even online searches) are spread across a dozen sources that publish on different calendars. The company wanted forecasts for every region of several countries, with honest uncertainty, refreshed as soon as new data appears.
What we built
Automated collectors gather surveillance reports, satellite climate and vegetation data, deforestation, El Niño indices, search trends and population, keeping a versioned monthly snapshot of each source. Gradient-boosted quantile models learn from a year of history per region and forecast the full range of likely cases; conformal calibration keeps those ranges honest. Every model is tested walk-forward and on an untouched holdout year before release, and forecasts reach the company through an API on AWS.
Built with Python, CatBoost, Google Earth Engine, MLflow, PostgreSQL and AWS.
One of Europe’s largest airlines priced thousands of flights by hand. We moved them to real-time pricing with our own custom-built algorithm, increasing revenue, enabling pricing by route and freeing the time of dozens of employees.
Manual, static pricing leaves a lot of revenue on the table. The challenge was clear: automate pricing to capture that untapped potential. The airline’s reputation depended heavily on its prices, so any solution had to be transparent and accountable to stakeholders, and react to changes in demand.
What we built
A transparent algorithm tailored to the airline, offering real-time pricing suggestions that balance revenue and demand. It beat the legacy model and kept stakeholders informed with explainable recommendations. A serverless design on Amazon Web Services, with a carefully built Python library, ensured scalability and accountability. The airline now adjusts prices confidently and autonomously, without compromising customer trust.
A leading Spanish telecom was losing many customers every month: marketing kept growing sales, but retention stayed low. We built a predictive system that flags when clients may leave, enabling personalized engagement.
A large customer base reaching saturation: a sign of market strength, but also a challenge. The company wanted to make marketing more cost-effective by focusing effort and budget on the customers most likely to convert, and to boost cross-selling with sharper, more tailored initiatives.
What we built
A two-step AI model. First, we pre-processed the available data, filtering out noise so the model could discern genuine customer interest. Then we predicted cross-selling opportunities with a striking degree of accuracy. Because data alone isn’t actionable, the results went into an intuitive dashboard: sales representatives see at a glance which products fit each customer.
Built with Python, scikit-learn, LightGBM and Streamlit.
A unified AI dashboard for a Spanish telecom that maps antenna performance issues and anticipates quality problems before customers notice them, on a data pipeline handling hundreds of gigabytes.
Day-to-day operations relied on combining Excel sheets and individual queries for antenna analytics. With more than 30,000 antennas to monitor, this was slow and made it hard to evaluate antenna connectivity and its impact on customers.
What we built
A data ingestion system for loads of up to 800 GB, with a backend on Airflow and Docker that kept processing efficient, fast and transparent. It feeds a dashboard with a clear, geographically detailed daily view of antenna performance. The team now responds faster to connectivity issues, links antenna problems to the clients affected and improves the quality of service.
Built with Python, Airflow, gradient boosting and Power BI.
An e-commerce company spent a large share of its budget on ads without knowing whether bidding more would win a better position. We built an algorithm that calculates optimal bids in real time, taking seasonality into account.
The company invested millions in ad placement. One site ran an intricate auction where the highest bidder gets the best position, yet in some periods a higher bid did not lead to a higher placement. Without knowing how the ad algorithm worked, spending felt like a shot in the dark. They wanted a smarter strategy, adjusted to the data rather than flat across all periods.
What we built
A predictive model that suggests bid prices, accounting for seasonality and daily fluctuations in demand. A lean design deployed on Google Cloud was ready for testing very quickly. The impact was twofold: better-informed ad spend in more than 50 markets, and people freed from the tedious bidding process. An internal dashboard compared the model with the old manual method.
Built with Python, Spark, Google Cloud Storage, serverless computing and PID control.
A major international organization wanted every statistical office in the world to generate its own population forecasts. We built resilient dashboards that handle their inputs and computations, with security measures for incoming data.
The organization needed to democratize its demographic projections. Most were produced in-house, yet most countries hold the vital statistics of their own populations in their national statistical offices. They needed a scalable, responsive and resilient application that let thousands of statisticians run their own simulations on the web, without technical hurdles.
What we built
Two web applications where users upload their data and receive population and vital-statistics projections, each with a personalized report, their data kept secure and the simulations running fast on the server. Behind them, the infrastructure runs hundreds of projections at once without compromising usability.
A logistics company allocated staff hours with fixed rules while demand for its products came and went. Our product-level demand forecasts made staff planning efficient and data-driven.
With many products facing intermittent demand, allocating the workforce was a constant challenge. Established procedures coped with daily fluctuations, but the company wanted more efficiency: the right amount of labor on the right days, without waste or shortfalls.
What we built
A forecasting model tailored to their needs. With sparse data, we used imputation methods and, critically, combined the client’s domain knowledge with iterative adjustments. We went beyond forecasting, adding optimization methods to guide workforce scheduling, so they anticipate and respond to changing demand.
Built with a time-series ensemble, NeuralProphet, Airflow and Streamlit.
Team
A small senior team. You work with the people who build your solution.
JC
Jorge Cimentada
Chief Data Scientist · PhD, Computational Sociology
Over eight years of experience with data and algorithms. Passionate about distilling complex ideas into actionable insights, valuing efficient solutions over prolonged prototyping, and keeping feedback cycles short so projects adapt as they go.
