Skip to content
  • Report
  • 19th Oct 2017

The Behavioural Insights Team Update Report 2016-17

Over the past year, the work of the Behavioural Insights Team (BIT) has spread into new countries and issues: we have conducted 163 trials in 25 countries, opened offices in Singapore and New Zealand, and expanded our work in Australia and North America.

Our focus remains on social impact, but with a gradual shift to more complex behavioural challenges. Our ongoing interventions are making longer-term and deeper impacts, such as the 27 per cent increase in maths or English pass rates of further education students experiencing our ‘study supporter’ intervention, at a cost of less than £10 per student.

In order to address an increasing range of challenges, we have expanded the range of techniques we are using. For example, this year has seen the creation of a new Data Science team within BIT which has begun to explore the use of machine learning to improve and tailor interventions.

Results from across our UK team’s work this year include:

  • A 34 per cent increase in acceptances by students from under-represented schools to top universities following a letter from a current top-tier university student from a similar background.
  • A ten fold increase in the proportion of savers visiting the Pension Wise website when sent a one-side ‘Pension Passport’ rather than the standard pension ‘wake-up’ pack.
  • An 8% reduction in annual household gas consumption following installation of smart heating controls. This significantly outperforms loft and cavity-wall insulation pound-for-pound, and could cut millions from consumers’ energy bills.
  • A 38 per cent reduction in patient referrals to over-booked hospitals, resulting from a pop-up prompt in the GP referral system. The intervention is now being scaled across the NHS.
  • A 25 per cent increase in ‘social trust’ among young people with low levels of social trust, measured four weeks after the introduction of a ten-minute ice-breaker discussion about the similarities between participants at the beginning of National Citizen Service.
  • A 28 per cent increase in comprehension of privacy notices required for a back-to-work programme, following a rewrite which also dramatically reduced their length.
  • Successful demonstration of a feedback platform for government purchasing, which we believe will promote improved purchasing and the more rapid growth of high-performing small and medium-sized enterprises.

Our work with partners outside the UK includes:

  • Latin America: A text trial to around 750,000 businesses in Mexico – one of our largest ever trials – leading to a 37 per cent increase in tax declaration rates.
  • North America: With the support of Bloomberg Philanthropies, we conducted 60 trials in 36 states over the past year. These were designed to be rapid interventions to build city capability in the running of behaviourally based trials, but also to bring immediate benefits to citizens. These included increasing by 75 per cent the uptake of a free pension advice session by city staff in Scottsdale, Arizona, and increasing by 31 per cent the sign-ups to automatic water bill payments by changing a lottery into a ‘regret lottery’.
  • Asia-Pacific: Reflecting the wider trend across BIT, we took on increasingly complex issues, such as addressing domestic violence in Aboriginal communities.


This report references a trial we ran in 2016 with the Dawes Trust and West Midlands Police aimed at reducing reoffending by drivers caught speeding. To confirm this result, we then ran two replications of the trial with the Metropolitan and Sussex police forces. When the first replication returned statistically null results, we re-analysed the original trial to see if we could understand the discrepancy.

In doing so, we discovered that a sudden dip in prosecutions during the trial had not been properly accounted for, and that there were some errors in the original code used for our analysis, which, together, led to a miscalculation of the original result. As a result, we find that intervention had no statistically significant impact on prosecutions or reoffending.

We have informed our partners of these errors and thank them for their ongoing assistance. Although an unusual combination of factors led to errors in this case, we nevertheless also went back over a number of trials to check for related errors and are satisfied that this was an isolated case. We have since then bolstered our internal processes to prevent similar problems in the future.

Since then, we have analysed the results of the replication with East Sussex County Council and Sussex Police. In this case the intervention did lead to a statistically significant impact on reoffending within 6 months of receiving both the updated Notice of Intended Prosecution and the informational leaflet. Given our findings in both the WMP and Sussex trials, these results were surprising and suggest that more work is needed to better understand why the intervention works in some contexts and not others. We will continue working with our partners to explore this in the coming months. 

This is an important reminder of the importance of replicating results before they are scaled.

Additional downloads