Make public services more accessible using AI

Here are some steps the ALLIANCE followed, you could consider following them if you also wanted to explore integrating a conversational artificial intelligence (AI) into the support or services you offer:

Step 1. Choose the right system

The ConnectingYouNow (CYN) system was highlighted to the ALLIANCE by Scottish Government as a potential future for accessing public services in Scotland, with health and wellbeing being the specific use case. Subsequently, the ALISS team was approached by Scottish Government as there was a desire to run a public beta test of CYN. For various reasons, other public sector websites (e.g. gov.scot, NHS Inform) had been ruled out.

The request from Scottish Government was considered by the ALISS team, who set out a number of pre-conditions for the implementation of CYN on aliss.org. There were:

• Responses need to be of a sufficient quality. E.g. service recommendations should be within a reasonable distance of the location identified.
• Look and feel needs to be in line with that of aliss.org.
• However, it should be obvious that it is (was) a beta test.
• A specific privacy policy for CYN is required to be visible on the site.
• Language used should follow the ALLIANCE’s communications guidelines (e.g. using words like ‘conditions’ instead of ‘diseases’).
• Accessibility should be verified, with a specific focus on screen reader compatibility.

Step 2. Consider how data will be used and managed

• A Data Protection Impact Assessment was undertaken and documented, in line with ALLIANCE data protection policies and procedures.
• We supplied details of the ALISS API end points to the CYN technical team. This allowed CYN to be trained on this data.  It was also agreed that CYN would scan the third-party websites, linked from ALISS service entries to obtain further information which could inform the recommendations produced by CYN.

Step 3. Test internally

• CYN was first installed on the ALISS test site (not publicly accessible), to enable the ALISS team to thoroughly test its functionality, prior to installing on the live site.

Step 4. Promote and test externally

• Following ALISS team testing, CYN was made live as a public beta test. To promote uptake during this phase, it was agreed that the CYN side bar would open automatically when users visited the aliss.org.
• We also promoted the CYN beta test via social media, and the ALLIANCE website. We spoke at various online and in-person events about ALISS and CYN.

Step 5. Design around your user’s needs

• Guardrails were put in place to cover situations such as when an emergency response to particular searches (e.g. suicide) was required.
• We decided that users should not be overwhelmed by the number of recommendations. We decided that three recommendations would be ideal in the first instance. Users can request more recommendations, if they wish.
• Users also have the option to see why a service has been recommended and they can give a thumbs up/down for feedback.

Step 6. Look for continous improvements

• Following the beta phase, it was agreed that we would continue with the CYN AI assistant on ALISS. However, the sidebar would not be automatically open (in line with user feedback).
• Continual liaison with the CYN team has ensured improvements are made to enhance the user experience.
• Tighter integration with ALISS and enhanced design has been scoped but not implemented as yet.