Detect Architecture


As described in the article Discovering Detect, Detect is considered the first step in the energy efficiency journey: a savings detection tool. It determines which buildings in your portfolio offer the greatest potential for savings and what technologies/energy savings measures can be adopted to be more competitive on a recurrent basis.

The logic behind this product is founded in a combination of energetic, simulation, and artificial intelligence (AI) models that can be performed massively and without the need to install meters or any type of hardware. Let’s look into its technical process:

Detect architecture overview

Below you can see an overall view of Detect's architecture:


  • ETL: The ETL (Extract, Transform and Load) is the process in charge of gathering all the data necessary to execute all calculations in Detect. The data retrieved includes weather data, the consumption profile, the location's metadata (such as the building use) and the local holiday calendar.
  • Clustering: The clustering is the process through which each building is grouped with up to 100 similar buildings.
  • Benchmarking: The benchmarking is the process through which each building is compared to the other buildings that have been included in their cluster in the previous step. From this benchmark, we gather the potential savings for each building.
  • Recommendations: This step estimates the impact of implementing the different available energy recommendations on the building. The impact is calculated from a financial and energy consumption point of view. 
  • Reports & User interface: In this step, we plot the results that have been obtained in all previous steps in the user interface and reports.
  • Added value models: This additional section is activated only in some specific cases that will be further discussed in the following sections.

ETL - Data processing

Once the account has been configured in the platform as explained in this article,  Detect gathers all the required inputs such as weather, electrical energy consumption and metadata (surface*, prices, tags, zip code, activity…) 

*If the surface is configured as <10m2, Detect will estimate the surface of the building through the surface imputer: a linear regression per activity that has been trained with real surfaces from our database. 

As part of this data processing, Detect validates the quality of the data (data gaps, outliers in energy consumption, constantly repeated values...) and discards those locations that do not match the required criteria. The following list shows some of the discarded reasons:

  • If the consumption for the last 12 months is lower than 1500 kWh, the location is discarded as the consumption is considered to be too low.
  • If there are less than 6 months of data, the location will be discarded owing to lack of data.
  • If a location does not have any reference device, Detect is not able to identify which device represents the total consumption for the building and needs to discard the location.
  • The location is discarded if it does not have electricity readings.
  • The location could be discarded if the specific consumption (kWh/m2) is too high or too low considering the use of the building.


The clustering module groups similar buildings, up to 100 buildings in each cluster. The criteria to find the most similar buildings are the following:

  • Similar building use: The building use majorly determines the consumption need and pattern. As an example, offices that usually consume from 9h-18h every day and have nearly zero consumption during weekends.
  • Similar need for heating and cooling: By taking into account the degree days, we will be comparing buildings that are part of the same climate. A similar need for heating and cooling is equivalent to the same weather.
  • Similar energy consumption pattern: For buildings that have the same building use and a similar need for heating and cooling, different energy consumption patterns will determine which type of equipment is available in every building. For example, an office with electrical heating will increase its consumption as the heating need increases, while an office with gas heating will behave very differently. 

In conclusion, a building will be considered similar to another if they have the same building use, the same climate and a similar consumption pattern. Under those conditions, the comparison will be considered valid.



Three-dimensional plot of clustering results for a given consumption pattern. The red dot represents a building, the orange colour represents similar buildings/neighbours to that building and the blue colour represents other buildings. 




The benchmarking module estimates the energy efficiency and savings potential for a given building, based on a comparison within the cluster of similar buildings obtained in the clustering module.

Each building is given an energy efficiency score that ranges from 0 to 100, with 100 denoting a building that is already operating at maximum efficiency. Additionally, for each building, we provide a monthly comparison chart of energy consumption values with the average and top 10% of buildings in each cluster. This comparison allows us to evaluate the savings potential for each building, expressed in three ways: energy (KWh), revenue, and as a percentage of the building’s current consumption.

With this information, the user can quickly identify buildings with a high margin for improvement as well as the most energy-efficient buildings, and proceed to implement and validate any further efficiency solutions.


Global energy performance/ Benchmarking outputs provided by DEXMA Detect



In this step, the available energy efficiency recommendations are evaluated for every building from a financial and energy consumption point of view. If a recommendation provides savings to the building, it is then shown in the user interface. There are 3 types of recommendations:

  • Recommendations that require investment, such as implementing a photovoltaic system in the building.
  • Recommendations that refer to the users' behaviour, such as transferring energy consumption from the most expensive tariff periods to cheaper ones.
  • Energy tips: Generic recommendations that apply to all buildings and have an informative goal, i.e. do not entail any financial calculations.

For some recommendations, it is necessary to know the installed heating and cooling equipment on the premises (for example, some heating recommendations can only be applied if the heating system works with gas). This can be estimated by Detect if there is a clear correlation between degree days and electricity consumption. This estimation will only be possible for some activities since some buildings might not be as affected as others by degree days' variations.


For further information please check the following article about Detect energy conservation measures.


Virtual audit recommendations in DEXMA Detect user interface


Report & User interface

All the outputs from Detect are shown through an interface inside the platform and can be exported in 3 ways:

  • Through a monthly generated report.
  • Through a csv file that shows the details of Detect calculations (available in the user interface).
  • Through a csv file that shows the details of Detect energy measures simulations (available in the user interface).


Example of Detect report 


Was this article helpful?