This project wanted to find out if we could reduce the demand for electricity at peak times. We partnered with energy advice app Loop to apply behavioural science to customer communications to see if we could increase opt ins to energy-saving events and further reduce energy use during the events.

Our study design involved different types of communication around encouraging users to opt in.

Below are some of the interventions participants were shown.

When we analysed our primary measurable outcome – electricity consumption during events – we found that there was no significant difference between the control group and treatment 1 (environmental framing) or treatment 2 (environmental framing + text message). We did find that treatment 3 (health framing) significantly increased electricity consumption compared to the control group [p < .010]. The effect was an increase of 15 Wh, which is roughly a 5% increase in electricity consumption.

When we analysed our secondary outcome – the proportion of each experimental arm that opted into each of the three events – we found that there was no significant difference between the control group and treatment 2 (environmental framing + text message). However, we found that treatment group 1 (environmental framing) significantly decreased opt-ins compared to the control group (a backfire) [p < .050]. This reflected a three percentage point decrease (50% in control group to 47% in this arm).

We also found that there was a weakly significant decrease in opt-ins for treatment 3 (health framing) compared to the control group (also a backfire) [p < .100], which was a two percentage point decrease.

We didn’t find evidence that the treatment effects that were significant differed across different events, indicating that the treatment effects were persistent.

Our results indicated that none of our interventions favourably changed opt-ins or electricity consumption during events – instead, we found some evidence of small backfires.

We think that this is because the customers who were participating in the DFS scheme were already sufficiently motivated - they had to specifically opt-in to be part of the scheme to be sent communications. This means that adding the environmental or health framings might not increase their motivation to participate in events or to reduce electricity consumption. Instead, it might be that adding text to emails adds slight frictions to opting in, reducing the number who opted in.

The health framing specifically may have been distracting or confusing for participants who are used to Loop providing financial or environmental reasons to engage with energy-saving behaviours.

Our findings suggest that focussing on minimising frictions to opting into events may be optimal to increasing savings from those participating in the DFS scheme. It may also be that 'stronger' interventions are needed – such as sending more notifications or defaulting everyone into text messages (we only sent around 500 text messages per event – around 15% of participants in treatment 2).

More broadly, it may be that trying to encourage more people to enter into the DFS scheme may be more beneficial than trying to further motivate individuals who have already opted in (around 25% of Loop's customers have opted in to the DFS scheme).

Overall, the trial shows why testing is important – we designed interventions that we thought would work, that ended up going against our expectations. Although our interventions were based upon previous research, our trial shows that results from different contexts are not always generalisable. Human behaviour is complex and context-specific, and the best way of finding out the impact of interventions is to test the intervention with the desired audience in the context of the behaviour you want to influence.