How can government and public sector organisations operate with fewer resources and at lower cost while maintaining the same level of service?
Inspiration for one potential answer comes from the world of electricity production.
One of the major challenges for power generation is that demand for electricity is not steady. Instead it comes in peaks and troughs. Some of those fluctuations are long term and relatively predictable: demand is typically highest on winter evenings around 4pm-7pm, and lowest around midday in summer (see graph below). Others happen very suddenly, such as when everyone switches on their kettle during the commercial breaks of the X Factor or the World Cup.
In both scenarios, the UK needs to have enough power stations to meet that peak demand. The problem is that this is very expensive: a modern power station can cost anywhere between £350 million and £16 billion in capital costs alone. And since peaks are followed by troughs, for much of the time those power stations stand idle, representing surplus capacity.
This leads to an interesting thought. If the peaks and troughs of electricity consumption could be evened out (i.e. we could lower the peaks and raise the troughs), it would be possible to decommission some existing – or avoid having to build some future – power stations while still meeting the same demand.
Reducing peak demand could therefore save literally billions of pounds.
Policymakers are excited by the potential of smart meters to do just that. The UK plans to roll out around 50 million of them across the country by 2020. The idea is that by giving consumers real-time pricing information, smart meters can help reduce peak demand by encouraging households to switch off appliances during peak periods and instead use them when the price of energy is low. Initially that will require changing consumers’ behaviour. Longer term, internet of things-connected devices will be able to switch themselves on and off depending on the cheapest time.
Could it be that government and public sector organisations also suffer from this “peak demand problem”? Put differently, is a significant proportion of the cost of providing some public services associated with meeting peak demand that is well above average use? The answer appears to be yes.
Consider the following examples:
HMRC self assessment tax returns. According to official statistics, in 2016, 4.45 million online self-assessment tax returns were submitted in January, 43 per cent of all those received throughout the year. 29 January was the busiest day with 513,271 returns completed, more than 21,386 per hour. The busiest time was between 2pm and 3pm, with 50,358 customers (839.3 per minute; 13.9 per second) clicking submit, briefly making the HMRC website the third most visited in the world.
HMRC has to have sufficient call centre and back office staff, office space, IT equipment and server power (and all the auxiliary services those things entail) to handle this demand. Some of that cost remains for the rest of the year, representing surplus capacity.
Student Loans applications. The majority of applications to the Student Loans Company happen in February. With an average cost of £43.80 to handle each transaction, in 2015 the cost of handling all these applications in February alone was £5.16 million. (See the data.)
Applications for passports consistently peak in March and June. According to the Guardian, the UK Passport Office spent almost £1 million in a month on overtime in 2014 as it attempted to deal with 490,000 outstanding applications.
A&E Attendance. According to a House of Commons report, Monday is the busiest day for A&E departments, with attendance 10 per cent above the daily average and also 10 per cent above the next busiest day, Tuesday. 10am on Monday is the single busiest hour. The figures below show trends for days and times in a week, highlighting that the early hours of Saturday and Sunday are busier than other nights. The quietest time is 5am on Wednesday.
Two negative consequences are associated with these kinds of peak demand.
The first is that costs do not always rise in a linear fashion (i.e. double the service, double the cost). Indeed, in some cases they increase almost exponentially as additional resources have to be brought in at considerable extra expense. This is particularly true of staff costs. Paying in-house staff overtime, or worse – having to hire locum or agency staff – can be incredibly costly. English NHS providers spent an estimated £3.7 billion on locum doctors, nurses and other staff in 2015-16.
The second is that peaks can also cause dissatisfaction among citizens and public sector staff. In A&E they are often associated with longer waiting times. They can even cause a service to become inaccessible. This has happened on several occasions when new systems or policies have been launched, creating sudden surges in demand. For example, the DVLA website crashed on 30 September 2014 when 250,000 people rushed to pay their vehicle tax on the night paper tax discs were phased out, peaking at 6,000 online payments per minute. The site crashed again on 8 June 2015 due to the (one would assume, predictable) surge of demand when the paper driving licence was abolished and replaced by an online version.
Evidently, the peak demand problem does exist in some parts of the public sector.
But the parallels with the energy sector need not be all bad news. After all, if they share the same problem, might the potential solutions be the same, too? Could technology offer ways to influence citizen behaviour or provide other approaches that could spread out demand so that the same level of public services can be offered with fewer resources?
The answer may be: yes.
Below are eight initial thoughts on possible approaches that could be considered to help address the peak demand problem in public services.
1 – Accurately model the peaks. Sometimes the public sector may have surplus capacity because it doesn’t realise the peaks exist. The first step is therefore to identify where they occur. ASI Data Science, a data analytics and machine learning company, recount their experience of working for a major airline.
In some parts of the budget airline industry the profit per flight can be as little as £100. It is therefore imperative that airlines never cancel a flight due to reasons within their control, such as staff absences. To ensure this, the airline had long ago decided that whatever staff they had on shift at any one time, they should have an extra 21 per cent on standby.
ASI analysed the airline’s historical standby and crew data and used a variety of machine learning algorithms to observe the problem more closely. Their analysis revealed that there was a peak of staff absence from work in September each year, when the 21 per cent buffer certainly was needed. But staff absences were considerably lower over the rest of the year. When they looked into what was happening, they observed that temporary summer staff were informed in the week before September whether they would be offered a longer contract. The understandable dip in morale of those not offered such a position was thought to explain the increased absences. By understanding when and why their peaks occur, the airline has been able to improve its internal communications, and reduce its standby staff capacity to an average of just 16 per cent, saving £10 million annually.
2 – Remove the (fixed) deadline. Where peaks occur because of people rushing to meet a deadline (e.g. self assessment tax returns, student loan applications, etc.), the simplest solution may be to replace a fixed date deadline with one that is tied to a movable date, for example the applicants’ birthday. (Admittedly, few people would be best pleased to face their tax deadline on their birthday itself!) In the cases where services have crashed due to the launch of a new system, such as the DVLA website, the obvious advice would be to phase in the new system gradually.
Thinking about a greater role for technology, another method might be to enable the continuous and real-time exchange of data, such that a definitive ‘moment of submission’ is not needed. This, in part, is the idea behind personalised digital tax accounts, which would remove the need for submitting self-assessment tax returns by ensuring the government always has the latest figures on citizens’ earnings.
3 – Create real time transparency with open data. The energy sector hopes that by giving consumers live pricing information with smart meters, it will encourage households to use electricity-hungry appliances at off-peak times. A similar approach might help some parts of the public sector.
For example, perhaps hospitals could publish their A&E waiting times in real-time according to an open data standard on data.gov.uk. App developers could then be encouraged to create tools that show which hospital has the longest waiting times right now, so that citizens living within reach of two or more hospitals could make an informed decision on where they go. This could be especially powerful if the waiting times could be combined with open data on transport times so that citizens can make an informed decision about the total journey and wait time.
It is, of course, possible that this could actually end up increasing overall A&E demand. Seeing a short wait could make it more likely that people turn up with minor complaints. Some local experiments might help prove that either way. A potential remedy would be to see if data on waiting times could also be combined with comparable information for other instant-access medical services, highlighting the alternatives available to patients.
Providing real-time A&E data consistently would also allow machine learning tools to understand when peaks occur at much more granular level than is currently possible so that further interventions could be designed.
4 – Put in place financial incentives to avoid peaks. Though unlikely to be popular, the comparison with the energy sector offers the lesson that variable pricing can be used to help incentivise avoiding peaks. This is already seen with the peak time fares for public transport, but could be considered for wider use. Could passport applications, for example, cost more at certain times of year?
A better approach might be to put in place positive financial incentives. For example, rewarding people with a discount, or even entering them into a lottery to pay early. An interesting case study comes from the Behavioural Insights Team (BIT), which worked with the Yarra Ranges Council in Victoria, Australia, to develop a prize draw that acted as an incentive for ratepayers to pay their rates in full and on time. The intervention appeared to produce positive results. BIT report that there was an increase from 5.7 per cent (3683) of properties paying early in the previous year, to 9.9 per cent (6371) of properties paying early in 2017 – an increase of over 70 per cent.
5 – Use digital nudge. There are some types of peak where it may seem impossible to address the underlying cause, but where looking at secondary causes may help. A&E departments might be one such example. There are lots of complex reasons around A&E attendance. However people consistently cite difficulty in obtaining GP appointments as a reason for usage. One of the reasons that people cannot see their GP is that some patients fail to turn up to their appointments, needlessly blocking doctors’ time. Figures suggest that more than 10 million GP appointments are missed each year in the UK. Solving just a fraction of that number could potentially reduce the knock-on pressure on A&E departments.
How could that be done?
Some GP surgeries send a text message to remind patients of their appointment. Yet few pay much attention to the timing and wording of those messages; characteristics which have been shown to be very important by the Behavioural Insights Team. In one project, BIT ran a randomised control trial with the Department of Health and Imperial College London, to find a cheap and easy way to tackle some of the 5.5 million hospital outpatient appointments that are missed every year in England alone. (Read the full casestudy.)
BIT report that:
“Patients attending five different clinics at Barts NHS Trust were randomly allocated to receive one of seven different messages. We tried out a variety of concepts from the behavioural sciences, including fairness to others, social norms, and reducing friction costs. We didn’t know what would be most effective… The clear winner was a message that pointed out the approximate cost of the missed appointment to the NHS (£160).”
The project led to a 25 per cent reduction in missed appointments (from 11.1 per cent to 8.5 per cent) for no extra cost.
Percentage of outpatient appointments recorded as “did not attend”
In a similar way, if it were possible to a) roll out text messaging to all GP surgeries, and b) increase the effectiveness of the SMS wording through knowledge of behavioural insights (which I’ve previously described as digital nudge), achieving anything like a 25 per cent reduction could free up thousands of GP hours, creating a knock-on positive effect for A&E departments.
Even in cases where the peaks themselves cannot be reduced, there may be a role for technology and new methods to help limit the pressure on public services.
6 – Use technology to tap into volunteer / paid support network at peak times. GoodSAM is an app that helps augment the capacity of the 999 service. When somebody calls the emergency services to report that an individual has had a cardiac arrest, as well as dispatching an ambulance, many ambulance trusts are now able to send out an alert to GoodSAM. The GoodSAM app alerts qualified first aiders in the vicinity of the victim, highlighting their location and that of the nearest defibrillator so they can hurry to the scene. When every second counts, those volunteers can – and do – make the difference between life and death.
Building on this concept, public service managers and civil society organisations could look at areas when volunteers might be alerted to action via a digital platform to augment the capacity of public services specifically at peak times. For example, volunteers could be called to help during peak demand for hospital transfers. In a recent article, Nesta’s Vicki Sellick cited the example of The Mix, a charity providing support to the under 25s, that has already piloted tapping into a volunteer network to handle calls and web chats from home.
7 – Use scalable IT infrastructure. Thanks to cloud computing there is rarely a good excuse for large organisations to have their own internal server capacity. Products like AWS have been specifically designed so that services can almost instantly ramp up to meet peak demand, and then reduce their computing power during quieter times, ensuring they are not paying for redundant capacity during non-peak periods. Strangely, there is still a large number of central government and public sector bodies that maintain their own in-house servers.
8 – Using sharing economy principles to remove surplus capacity. In a small number of cases, public sector bodies may also be able to remove some of their surplus capacity by letting others use it during off-peak times. Share Somewhere is an organisation funded by Nesta’s ShareLab Fund that points to how this might be done. It enables small community groups to find or offer for hire their spaces when they are not being used, enabling them to monetise their surplus capacity. A similar approach is being piloted by some libraries who are developing a new source of income by providing some of their space for civil servants needing hot desks.
I am under no illusions that the ideas above represent a silver bullet to resolving the financial challenges currently facing the public sector. Furthermore, in some cases it is imperative that government and the public sector are equipped to deal with extreme peaks, for example in response to major disasters.
However, in cases where the peaks are self-imposed, or do not have to exist, thinking creatively about tackling the peak demand problem may well be one worthwhile approach to ensure that we can continue to deliver high quality public services in a tight financial climate.