AI and the future of service delivery

A hot topic in the charity sector right now is how artificial intelligence (AI) can help streamline processes, increase efficiency, and help with creativity. But how can it help with service delivery? And how can it be used to diversify income streams?

 

 

AI and the sector – what’s changed?

 

The term ‘artificial intelligence’ was coined almost 70 years ago in 1955 in a proposal for a research project on AI. While there have been milestones every decade, AI has only really been talked about in the charity sector in the last few years.

 

Thanks to the Charity Digital Skills report, we can track charities’ interest in AI and how they are (or aren’t) using AI tools. In the 2024 report, using AI tools is a priority for 53% of large charities compared to 26% of small charities.

 

Currently 61% of charities are using AI in their day-to-day work or operations, which has increased from 27% in 2023.

 

But when it comes to AI and service delivery, only 5% of charities are offering services built on AI tools, whilst 12% say they are using AI behind the scenes. Overall, 15% of charities are using AI in service delivery.

 

So how are charities using AI in service delivery and what does the future look like?

 

 

Caddy, the AI-powered assistant for Citizens Advice advisors

 

In 2022, when ChatGPT launched, Stuart Pearson, Head of Innovation at Citizens Advice Stockport, Oldham, Rochdale and Trafford (CASORT), and his colleagues started to look at how AI could be used in service delivery.

 

As a charity that provides trusted information and advice to people, using AI can present a reputational risk as generative AI can be prone to hallucinations. This is when a large language model generates a response that is false, nonsensical, or inaccurate.

 

Stuart and his team started looking at how they could benefit from the technology, while mitigating the risks. They very quickly came to the agreement that whatever they built would involve a “human in the loop” validation system to ensure that the advice given by the advisor when speaking with the client via phone or in-person was accurate and reliable. 

 

Post-pandemic, an opportunity presented itself when there were a number of trainee advisors recruited to the charity, which was putting a strain on the supervisors who had to check and validate their responses before they could be sent to clients.

 

To help reduce the strain on supervisors as well as improve client response times, the team built Caddy – an AI-powered assistant that only draws information from GOV.UK and Citizens Advice’s own website and their propriety knowledge base.

 

It helps advisors find information quickly, but responses still need to be approved by a supervisor. With support from National Citizens Advice, Caddy has been trialled with six local Citizens Advice offices and feedback has been overwhelmingly positive.

 

Initial analysis of response times shows that Caddy generates a response in four minutes (including the supervisor’s approval) versus nine minutes by an advisor searching on their own.

 

Advisors have received training in how to give better prompts, but Stuart and his team are now building a function where, before providing an answer, Caddy will prompt the advisor to provide more information about the client’s circumstances so that it can generate a more tailored response.

 

The charity has just received funding to build another AI solution for case note automation. Citizens Advice is required to write up every interaction with clients into case notes – for insurance purposes but also for other advisors to see what advice has already been given to the client.

 

They are piloting an AI solution that will transcribe the call and then generative AI, using their templates, will generate a case note. So far, it’s saving up to 50% of write-up time, leaving more time to help more people.

 

Following its success, CASORT is looking at how they could potentially monetise the model they have built, to provide an income stream for the charity.

 

 

How RNIB is using AI to scale its accessible document service

 

One of the services offered by RNIB is converting documents into accessible formats. It’s used by businesses, government agencies and the NHS so that they can communicate in people’s preferred format. This could be braille, large print, or audio formats. The types of documents that need converting include prescriptions, bank statements, mortgage applications, critical welfare information and more.

 

There are workstreams for each of these different types of documents, but it requires a huge amount of staff time to write custom code and fix software errors.

 

Aidan Forman, Director of Technology and Digital Transformation at RNIB, and his team researched how AI could not only help in streamlining the process but scale the service too.

 

They developed an AI-based solution called Mailings, using Microsoft Azure AI services, to automatically convert documents into an internal markup tool. That document is then checked by an RNIB employee for accuracy and then sent to their publishing and braille embossing facilities.

 

Using AI has drastically reduced the time needed to convert communications into the different formats needed for accessible communications. It’s also improved the quality of their audio formats. Previously they used a robotic sounding voice, but Azure AI Neural Voice generates a natural, conversational audio recording which makes for a much better user experience.

 

In terms of efficiency, RNIB estimates that conversion time has reduced from up to two weeks to around three hours through using an AI solution. This frees up capacity to take on more clients and generate more income.

 

 

The future and AI

 

AI is not a trend or a fad. It’s here to stay. And charities that don’t invest in AI capabilities are going to be left behind. As Caddy and Mailings demonstrate, AI not only saves time and increases efficiency, but it can also increase revenue too.

 

Aidan recommends that charities invest time and headspace in looking into AI, researching the various models, getting the fundamentals in place, and testing and learning. “You don’t have to go and do everything. You can start small. At RNIB we experimented with proofs of concept and then built that up.”

 

Once you can prove a solution works, put together a business case to scale up the service.