Can Eliq’s Energy Insights reduce electricity by 25%? ‘Science court’ says Yes.

The results of a recent paper shocked us at Eliq so positively to the point where we were reluctant to share them because they sound too good to be true. But true they are, share them we will, and you can take them to “court” because mighty academia is backing them up: Eliq’s energy insights can lead to 25% reduced electricity consumption on a YoY basis. But back to the court, the defendant calls as a witness Cosimo Faeti, MDataSc, author of the paper Causal Impact Analysis, 20241, who used Eliq’s data to research and quantify the causal relationship between energy usage and Eliq’s energy insights.

In October 2023, Cosimo Faeti joined Eliq as a data science intern on a mission, with the following research proposal: Do Eliq’s energy insights impact energy usage? And is it a cause-and-effect relationship – meaning can we be sure that it’s Eliq’s insights that cause any behavioural changes? Fast forward to early 2024, when Cosimo swore by science to tell the truth and nothing but the truth and presented his results. The defendants claim:

Eliq’s insights do, in fact, impact electricity user behaviour – in the overwhelming majority, for the better.

And this is not a generic, navel-gazing claim. More specifically,

In over 80% of the cases studied in a focused area, electricity users reduced their electricity consumption between 24-26%. This translates to:

– Daily savings between 8.2-8.9 kWh

– Annual savings between 3,000-3,200 kWh which translates to annual savings ranging between £630-1,000

But you don’t have to take these numbers at face value. We call Cosimo Faeti to the stand to defend these claims.

Your Honour, association does not imply causation

The first thing Cosimo poignantly remarks is that association does not imply causation. Cosimo explains that proving a causal relationship between the results and Eliq’s energy insights was central to his paper. In fact, proving if and how Eliq’s energy insights affect electricity user behaviour steered Cosimo’s methodology of choice: Causal inference2 – the science of inferring causation from association, to help us understand when and why association and causation differ. He explains:

Causal inference is useful when we want to understand the cause-and-effect relationship between variables so that we can intervene in the cause to bring a desired effect. It is imperative to not confuse association, or correlation, with causality. Association occurs when two variables move together, while causality occurs when a change in one variable causes a change in another. Often, people assume a cause-and-effect relationship when two variables show a high correlation but in most cases, this is wrong.

Cosimo continues to present some highly entertaining evidence on just how wrong confusing association with causation can be. Exhibit A, your Honour.

Respectfully included from Tyler Vigen’s website, Spurious Correlation. Visit for more wonderfully absurd examples.

– Objection: Argumentative!

Causal inference is something Artificial Intelligence does not excel in. A machine learning algorithm can identify associations en masse and make predictions based on them but it cannot effectively identify causal relationships.

– Sustained. Moving on

Presenting evidence: Comparing pre- and post-treatment periods

So how did Cosimo Faeti design his research to reveal a causal relationship between Eliq’s platform and electricity usage? The first assurance comes from the group under examination. The context of analysis led Cosimo to study only the electricity users with access to Eliq’s platform, to ensure that the correlation is controlled and can safely reveal causation. Cosimo comments on this choice:

The context of the analysis led us to select only those electricity users who got access to Eliq’s platform. Our context of analysis isn’t the most suitable for applying causal inference techniques, but we still made it and got promising results.

The second assurance comes from the studied group in relation to the time studied. The paper examines electricity usage between two periods.

– A time period before the access to Eliq’s energy insights

– A time period of equivalent length after access to Eliq’s energy insights.

All electricity usage data comes from smart meter readings of Eliq users who consented to store this data in Eliq. The new variable introduced in between is the launch of Eliq’s platform and subsequent access to energy insights powered by Eliq. The second period, referred to as “post-treatment” in the paper, is split and studied under two possible projections: Actual electricity usage with access to Eliq’s platform and the counterfactual.

Cosimo explains it in a much more entertaining and tangible way:

Think of it as a parallel universe study. For each location, each representing a household, we have data covering several months before and after the treatment. In other words, we have electricity consumption data for the pre- and post-treatment periods. We can call this our current universe.

Since the goal is to measure Eliq causal impact on electricity consumption, we build a model on the pre-treatment period data to approximate a counterfactual scenario in which the users did not engage with the “treatment”, Eliq’s platform. In other words, what would electricity consumption be if the treatment had not happened. This is the parallel universe.

To measure the causal impact, we use the difference between the two scenarios, our current universe and the parallel universe. And the difference was, let’s say noticeable.

Case Closed: Eliq’s Insights effectively impact electricity usage

The results from Cosimo’s paper were astounding. As we mentioned in the beginning, in more than 80% of the cases, users of Eliq’s insights noted 24-26% reduction in electricity consumption. Users saved up to 8.9 kWh daily in consumption – the equivalent of a month’s long washing machine use1, and up to £1,000 annually from energy bills – the equivalent of upgrading to an A+ TV and tumble dryer. 

Cosimo explains the results further:

Further analysis demonstrated that most locations showed a downward shift in baseline electricity consumption. The remaining locations reported a long-term trend of behavioural changes, leading to a steady but constant reduction in electricity consumption.

The reasons include the efficiency of Eliq’s energy management solutions and its energy insights platform within just one year after adoption, the flexibility and adaptation to user context and lifestyles, and the high accuracy of customisation and personalisation, translating into user behavioural changes in the way they use energy.

The defendant calls Vaidas Armonas, Head of Data Science at Eliq, to the stand. Vaidas was in close contact with the study, supporting Cosimo Faeti with the dataset navigation and result evaluation. Vaidas testifies:

After years of closely following up on Eliq’s original design and experiencing the impact of Eliq’s Energy Insights, we had qualitative evidence that the platform is changing energy users’ behaviour for the better. However, frankly, we did not see that impressive a result coming, so when it did, we ourselves scrutinised everything throughout the research project. Indeed, according to the analysis, the amount is average for a significant number of locations that were analysed.

No further questions, your Honour. In the case of Eliq’s energy insights vs overachieving marketing claims, the jury finds the defendant guilty of effectively impacting electricity usage and user behaviour!

1 Cosimo Faeti. Causal Impact Analysis. 2024. [Master’s thesis, Università di Pisa]. Contact the author directly here to consult the original research.
2 Matheus Facure. Causal Inference in Python. Applying Causal Inference in the Tech Industry. 2023. O’Reilly Media, Inc.