ECCO2 reaches 3 million m² and launches NARA AI V5

Givisiez, October 2026

More than 1,000 properties demonstrate that Building Intelligence works from individual buildings to large real estate portfolios.

ECCO2 optimises heating systems during ongoing operation – without replacing the heating system and without structural intervention. Sensors capture the building’s thermal behaviour. NARA AI combines this data with weather information and the digital twin to control the heating system proactively according to actual demand.

Learning from more than 1,000 buildings: with NARA AI V5

The extensive data base from three million square metres continuously feeds into the further development of NARA AI.

NARA AI V5 improves the quality of input data, detects temporary changes in individual apartments and processes local weather information more precisely. This makes heating control more robust and even more closely aligned with actual demand.

NARA AI is built on more than 20 terabytes of data. Around 10 million new data points are added every day.

NARA AI V5: AI with a clear purpose

NARA AI V5 heats buildings only as much as actually necessary, while maintaining the same level of comfort for tenants.

This reduces energy consumption, operating and ancillary costs, as well as CO₂ emissions. At the same time, measurable operational data provides a reliable basis for ESG and portfolio management.

“3 million m² is an important milestone for us. At the same time, it is only the beginning. What matters is that we learn from every building and apply this experience to the optimisation of the entire portfolio.”

Antoine Eddé
Founder and Chairman of the Board, ECCO2 Solutions AG

ECCO2 announces NARA AI V5

AI goes green.

ECCO2 is pleased to announce NARA AI V5, the latest evolution of its AI-powered heating optimisation platform.

Since its introduction in 2021, NARA AI has continuously learned from the real-world behaviour of buildings, weather and heating systems. V5 takes this approach a step further — not by making the AI more complicated, but by making the information it receives cleaner, more precise and more useful.

The result is a more resilient heating optimisation system, improved comfort in challenging weather conditions and a remarkably efficient approach to AI.

Better information. Better decisions

NARA AI ultimately has two fundamental questions to answer:

How warm is the building?

and

How will the weather affect it?

The quality of the answers matters enormously.

NARA V5 therefore focuses on two areas where even very sophisticated predictive control can encounter difficulties: the calculation of average indoor temperature and the accuracy of local weather information.

AI-improved indoor temperature.

ECCO2's High Resolution Sensing network now provides real-time indoor temperature data from 26,500 sensors across the buildings monitored by NARA AI.

This wealth of information is a strength — but it can also contain temporary patterns that do not represent the actual thermal state of the building.

For example, during a holiday period, a number of residents may leave their apartments at the same time and turn down their thermostats.

For the residents, this is perfectly sensible behaviour.

For a heating optimisation system, however, the resulting temperature drop could look like a signal that the building is not receiving enough heat.

NARA AI V5 recognises such patterns and avoids the related "hallucination".

Using real-time pattern recognition, V5 identifies misleading inputs that could otherwise distort the building’s digital twin and interfere with its continuous fine-tuning and excludes them from the calculation of average indoor temperature.

Importantly, the exclusion is precise in both time and scope. Once the anomalous pattern has passed, the data stream is brought back into the calculation.

The result is more stable predictive control — and therefore more consistent comfort for residents.

The example below illustrates the effect on the calculation of average indoor temperature (AIT).

The grey curve shows the average calculated from all available sensor data, including sensors located in durably uninhabited units.

The blue curve shows the AIT calculated by NARA AI V4 using static exclusion. Data from permanently uninhabited units is excluded, but temporary absences are not accounted for.

The green curve shows NARA AI V5 using dynamic exclusion. V5 detects temporary absence and the resulting temperature anomaly, and temporarily excludes the affected data stream from the AIT calculation.

During the highlighted episode, the individual sensors tell a misleading story: the temperature drops in a number of apartments because their occupants are away. NARA AI V5 recognises the pattern and keeps the building-level AIT stable.

All processing is based on anonymised data, in accordance with applicable Swiss data-protection requirements and the GDPR.

AI-improved fog management

Weather forecasting presents a different challenge.

Predictive heating works well when the system knows precisely what weather is coming. Temperature and solar radiation forecasts are therefore among NARA AI's most important inputs.

But Switzerland has a particular talent for making weather forecasting difficult.

Along a river, dense fog can linger unexpectedly throughout the day while the neighboring hillside enjoys bright sunshine.

Two buildings only a few hundred meters apart can therefore experience markedly different outdoor temperatures and solar gains — and consequently require very different amounts of heating.

Even the most advanced weather forecasts struggle to capture these local variations at building level. They may predict “sunny everywhere” when fog persists along the river, or “fog everywhere” when the surrounding areas are already clear. And even when fog is correctly forecast, predicting precisely when it will lift is surprisingly difficult.

For predictive heating control, these seemingly small differences have a significant impact.

NARA AI V5’s forecast-enhancement technology is specifically designed to address this challenge.

It combines high-resolution, high-frequency forecast information with dedicated fog-management logic, enabling NARA AI V5 to adapt intelligently as fog develops, persists or lifts unexpectedly.

The underlying method is currently the subject of a patent application, so we will keep the technical details to ourselves for the time being.

What we can say is that backtesting against ECCO2’s HyperCube database demonstrated more robust and comfortable heating control under foggy conditions, while maintaining minimal impact on energy efficiency.

Made in Switzerland. Running in Switzerland

NARA AI V5 continues ECCO2's commitment to keeping customer data in Switzerland.

Customer data is stored on Swiss servers and transmitted using the infrastructure of our long-standing technology partner Swisscom, providing a Swiss-based foundation for the ECCO2 Building Intelligence platform.

The NARA AI V5 engine itself is an ECCO2-specific implementation built on a powerful open-source technology foundation.

Rather than relying on a general-purpose, text-based large language model, the ECCO2 R&D team has developed a highly specialised decision-making system designed specifically for the task at hand: optimising heating systems.

This distinction matters.

NARA AI does not need to write an essay about a building. It needs to make a very good heating decision.

By removing unnecessary layers between data and decision, ECCO2 has observed 400% improvement in processing efficiency in its internal benchmarks.

This is a particularly Swiss approach to AI: use exactly as much technology as the job requires — and no more.

20 terabytes of experience. 10 million new data points every day.

NARA AI V5's initial training was supported by ECCO2's 20+ terabyte HyperCube database, built from years of experience in heating optimisation

Today, ECCO2's sensor network contributes approximately 10 million additional data points every day.

This creates an unusually rich feedback loop:

Because NARA AI is trained and refined around the highly specific task of heating optimisation, it does not need to compete with general-purpose AI systems at everything.

It only needs to be very good at one thing:

making buildings use less energy while keeping people comfortable.

And that is precisely what it is designed to do.

AI with a smaller footprint

Artificial intelligence is often associated with very large computing infrastructures and equally large energy requirements.

NARA AI V5 takes a different path.

By using a highly specialised architecture, ECCO2 runs its AI locally on dedicated infrastructure with an energy requirement comparable to that of a single desktop PC.

If AI is being used to reduce the energy consumption of buildings, it is rather nice if the AI itself does not need very much energy to do so.

Less computing. More useful intelligence.

We call this approach Green AI.

It reflects a principle that has been part of ECCO2 from the beginning: help create a better balance between comfort, energy consumption, economics and the environment.