Artificial intelligence case studies: Good practice by local authorities

Background

As with many public bodies, Camden Council needs to collect, process and share residents’ personal data, including equality data, to better plan and deliver its services.

In order to collect the best data possible, the council was clear that it needed to work with others to build trust among its residents in sharing their personal data so they understand how and why their data is used, including when using AI-based technologies.

Actions taken

To do this, Camden Council conducted research in 2021 to understand residents’ attitudes to its use of data, including in relation to new technologies that allow greater automation.

The research included a literature review, a survey and interviews with a sample of residents of different sexes, ages and ethnicities. During the interviews, residents were presented with examples on how the council was already using data and how it would like to use data in the future, including in the context of AI.

A report of findings was published. It led to the setting up of a panel made up of residents representative of the borough, including people  with protected characteristics under the Equality Act 2010.

Camden Council engaged with this residents’ panel to co-create the Camden Data Charter. This provides a vision, principles, success criteria and governance to guide how the council should collect and use personal data, including when using AI-based technologies.

The principles of the charter are: 

1. Build trust through transparency.
2. Provide accountability and oversight.
3. Make sure data is secure, safe and ethical.
4. Make sure data is used for public good and be mindful of residents' data.
5. Be beneficial for all by using an outcomes-based approach.
6. Camden’s external partners should sign up to the Data Charter principles.
7. Protect individuals' rights and privacy.

In January 2023, a second residents’ panel reviewed the charter and decided to add another principle:

8. Ensure that the information Camden shares about data use is clear and accessible.

Outcomes and next steps

The Data Charter provides a clear framework on how the council should process residents’ personal data, including when using AI-based technologies. It also provides principles on how the council should communicate its use of data to residents so they understand that it can be done ethically and for their benefit.

This has helped the council to build more trust with residents which was evaluated by an independent evaluator over the course of both resident panels. It has done this by involving people with different protected characteristics under the Equality Act in the creation of the Data Charter. These representatives continue to be involved in reviewing how the council processes data. As explained in this short video, it has also enabled residents who took part to see that ‘the outcomes of good data collection can be very positive for the community’.

Camden Council also committed to ‘communicate with its residents through an appropriate channel, either Resident Panel or through Camden Talks, if a new technology or methodology for data use is proposed which significantly alters the delivery of Camden’s frontline service’. Where relevant, and in line with requirements under the Public Sector Equality Duty, this will include considerations of the equality impact of adopting a new technology.

Background

Camden Council’s Data Charter was co-created with residents in 2021 and reviewed in 2023. One of its key principles is to ensure that the information Camden shares about data use is clear and accessible.  

In 2021, the council produced case studies on its website to show how it was using data, including in the context of AI-based technologies. Although case studies were created with a feedback function for residents to use, few comments were left.

In 2023, Camden Council acknowledged that it needed to take a different approach to ensure information on its use of data reached all its residents, particularly disabled and older people as well as people who do not speak English as their first language.  

Actions taken

In response, and to better comply with the duty to make reasonable adjustments and the Public Sector Equality Duty (PSED) under the Equality Act 2010, Camden Council committed to:

• provide information about the Data Charter and data use in various languages and formats
• develop communication materials that break down particular aspects of data use, including when making changes to council services
• improve how the Open Data Camden platform displays information, and to make its contents more visual where possible 

Outcomes and next steps

Camden Council is making reasonable adjustments to the way it provides information on its use of data and AI-based technologies so people with different types of impairment can access information on its use of data and AI-based technologies. In particular, the council produced:

• accessible communications around the data charter in large print, as well as Easy Read.
• Animated explainers in collaboration with the Alan Turing Institute on two AI-based technologies the council uses. One is on machine learning and the other is about data matching. Camden has put these on their YouTube channel as well as on X (formerly Twitter) and Facebook and is considering producing more explainers.  

Camden Council is now looking at how it can provide information about the Data Charter in British Sign Language (BSL).

In addition, and in line with the PSED, the council is taking positive action measures to ensure that disabled and older people who cannot use the internet or who do not speak English as their first language can access information on its use of data and AI-based technologies. In particular, the council:

• printed information about the Data Charter that residents can take away and read
• started holding less formal and face-to-face events called ‘Camden Talks Data’ in libraries and community-run spaces

The council is now considering providing information about the Data Charter in languages spoken across the borough.

Background

To comply with the Public Sector Equality Duty, public authorities must embed equality in everything they do. This includes when they decide to buy a new product like an AI-based technology or to commission another organisation to use such technology on its behalf. It is critical that decisions to buy goods or commission services are informed by an appropriate consideration of equality. This helps those public bodies to make sure that they are buying goods and services that will work for all the people they serve. It may also help them to guard against mistakenly buying products and services that lead to unlawful discrimination or that worsen inequality.

As explained in our procurement guidance, when a public authority decides to contract goods and services such as an AI-based technology or service, it should consider whether doing so is relevant to equality. If so, it should consider the potential impact (positive or negative) of commissioning such technology on people with protected characteristics. Such considerations should help public authorities determine whether and to what extent equality should form part of its tender specification and contract arrangements.

Since 2021, Camden Council has developed a model to embed equality considerations when commissioning and contracting AI-based technologies and services.

Actions taken

1. In its procurement strategy, Camden Council requires staff wishing to procure AI-based technologies to confirm that:

• equality issues have been considered in terms of the subject matter, and
• an Equality Impact Assessment (EIA) has been completed and attached – if not, to explain why one was not necessary, detailing the consideration that has nonetheless been given to equality

2. Camden Council’s Goods and Services Contract template requires that the contractor shall:

• follow all relevant codes of good practice including those produced by the Equality and Human Rights Commission, such as our public procurement guidance, Buying Better Outcomes, our guidance on AI and the PSED and the Technical Guidance on the PSED, and
• at its own cost, produce Equality Impact Assessments as and when required by the council

3. Given the possibility of unintended bias when using AI-based technologies, particularly those that involve machine learning, Camden Council recently added the following additional clause to its contract template:

• ensure that the underlying product, methodology and / or model of delivery has not been developed in a way that may lead to the discrimination or unequal profiling of citizens

4. In its Contract Monitoring - post contract award, Camden Council requires that staff:

• advise on reporting requirements and Key Performance Indicators such as periodic equalities monitoring, and
• highlight the winning supplier’s social value offer. For AI-based technology, this would include the necessity to upskill Camden staff in its use and to help present the potential use of this technology to Camden residents

Outcomes and next steps

Camden Council has made significant progress in embedding equality considerations in its AI commissioning and contracting processes. We are looking forward to receiving examples of how the council has used its model in practice. We acknowledge, however, that this is an approach other public authorities can learn from.

Moving forward, Camden Council is planning to request that all AI contractors adhere to the Data Charter principles it has co-created with its residents’ panels, and for the principles to be built into contracts by default. For more information on the Data Charter and Camden residents’ panels, read this case study.

In addition, the council has set up a Data Governance Board and is looking to set up methodology panels and ethical panels made up of external experts with both technical and non-technical backgrounds to assess the need, feasibility and ethical considerations behind any AI project.

The council is clear that any AI-based technology it contracts out will be discontinued if:

• the monitoring and evaluation show that it demonstrates bias
• it is not identifying people in need of support where it should do (including, potentially, people who share protected characteristics)
• or more generally does not produce additional useful information

Background

188 languages are spoken in Hounslow. 72% of the population (aged 3+ years) speak English as their main language. A further 23%, whose main language is not English, speak English well or very well. However, in Hounslow 5% do not speak English or do not speak English well, this is higher than the average for England (2%) and London (4%). To ensure all residents can access the information on the Hounslow Council website in the languages and formats they need, automatic language tool Google Translate has been utilised. This makes pages immediately accessible in the user’s language of choice and  significantly reduces the volume of documents requiring costly translation. 

Actions taken

Hounslow Council’s content strategy encourages key information to be made available on the website in plain English and HTML. 

HTML is the preferred format because:

• it is much more accessible to all users, including people with visual impairments who use screen reading technology
• it enables users to immediately translate content into the language of their choice using Google Translate, which appears on all the council’s webpages
• content can be edited quickly and easily

As such, when possible council officers convert PDFs into HTML pages. 

Outcomes

By taking this approach, Hounslow Council aims to make it easier for residents who do not speak English as their first language to access its services. It should also help residents to participate fully in their communities. 

The Public Sector Equality Duty is a continuing duty. As such, the council must monitor the actual impact of implementing its content strategy on people with protected characteristics under the Equality Act 2010. This should also include the impact of using Google Translate on improving access to information for residents who do not speak English as their first language.

Background

The London Borough of Barking and Dagenham (LBBD) aspires to be a ‘digital council’.

In April 2021, in collaboration with technical specialists and behavioural scientists, LBBD decided to develop an AI-powered website chatbot. This would automatically respond to residents’ general enquiries about things like bin collection or council tax. It aimed to enable more customers to self-serve and reduce the number of costly phone enquiries, without compromising the quality of the service.

Actions taken

The AI-powered chatbot was piloted between April and June 2021. Following lessons learned, it launched in September 2021.

Before implementing the chatbot, LBBD considered its potential equality impact. They reviewed relevant evidence and identified that residents who could not access the internet, or who struggle to access digital services, could be put at a disadvantage. This includes people with protected characteristic(s) under the Equality Act 2010, such as older people, people with certain disabilities or people who do not speak English as their first language.

To mitigate this, LBBD decided to retain capacity in its telephone and face-to-face services. They also introduced 45 dedicated face-to-face digital champions who can support residents who need help:

• getting online
• navigating through content
• completing transactions

LBBD also tested the chatbot with a customer community panel, and with visually impaired people who use screen readers.

Outcomes

The chatbot has enabled more customers to self-serve, reducing the number of phone calls to the council’s customer contact team by 1,000 a month on average with tangible financial savings for the council.

The chatbot has a customer satisfaction score of over 85%. LBBD has started putting systems in place to monitor the actual equality impact of the chatbot. Task completion rates from the chatbot can also be monitored by language use. The council will know if users of a particular language are having difficulties in using it.  

The use of the AI-powered chatbot has also helped the council make its traditional methods of contact more available. As more residents use the chatbot, the telephone service focuses on supporting vulnerable customers. This includes people with limited digital literacy or access to the internet, and people who may experience language barriers.

The Public Sector Equality Duty also applies to how the service will run. As such, the council will need to continue monitoring the actual impact of the chatbot on people with different protected characteristics. This should also include monitoring the impact of the mitigating actions the council has put in place, and identifying new actions if needs be.

LBBD is committed to keep engaging with its customer community panel and residents to inform the development and use of potential other AI-based technologies. For example:

• The chatbot can currently respond in English to messages typed in many different community languages. Recent developments in generative AI mean that it may now be possible for the chatbot to give answers in the language the questions were written in. LBBD is currently considering this as 64% of LBBD residents have a first language other than English.
• The council is also planning to pilot voice recognition AI technology in its telephone contact centre to respond to low-priority, non-urgent calls. This technology will be tested by residents from different backgrounds to ensure it works well with the range of accents and languages spoken in the borough.

Background

In 2019, Maidstone Council faced a significant increase in social housing needs and homelessness. The council approached a technology company to trial an AI-based technology that uses data to predict an individual’s risk of homelessness. The council hoped that using this AI model would support housing officers to take more effective action to prevent people from becoming homeless.

Action taken

The technology company analysed academic research and historical data from across the UK. It found that many risk factors contribute to the likelihood of homelessness including:

• someone’s financial situation
• certain protected characteristics under the Equality Act 2010, such as age, disability and pregnancy and maternity

The company included these characteristics to its AI model to help predict the risk of homelessness.

Outcomes and next steps

The council’s data on the effectiveness of this technology covers a period from January 2021 to November 2022. This shows that:

• where the alerts from the AI model were actioned with an offer of support, 68% of residents engaged with the intervention
• in 98% of these cases, homelessness was prevented

Conversely, where the alerts were not actioned, 76% of at-risk residents became homeless.

The alerts provided by the AI model were reported to be accurate in 84% of cases. Overall, the use of these alerts appears to have helped the council prevent homelessness in 27% more cases than the standard non-technology approach, and 74% more cases than no approach at all.

The council collected feedback from the individuals they have helped. This included information about their protected characteristics.

Building on the success of Maidstone, Kent County Council has established a countywide pilot of the AI model with all district and borough councils in Kent.

As part of its equality impact assessment, Kent Council is keen to understand how various protected characteristics may play a role alongside other variables, such as socio-economic factors, in leading residents to becoming homeless. The AI model can then be refined to make the council more efficient in identifying and responding to risks of homelessness across the county.

The Public Sector Equality Duty (PSED) also applies to how services are run. As such, the council will need to continue monitoring the actual equality impact of using the AI model. This should include reviewing who is and is not benefitting from its use and taking steps to understand and tackle any discrepancies that may come to light.  For example, an AI-based technology that uses national data sets may fail to pick up on local demographics, particularly if there is limited data on some groups who live locally. This may be particularly problematic for new and migrant communities or for Gypsy, Roma and Traveller groups.