About Accela
Accela is a global provider of a cloud-based civic platform for government jurisdictions and citizens to build and strengthen communities. Their platform seeks to enhance government services by boosting efficiency, and improving citizen engagement through the facilitation of operations ranging from building permits to code enforcement inspections.
One component of the Accela civic platform aids in the management of the 311 request line contact center.
Accela’s Challenge
311 request lines – also known as non-emergency complaint lines – face challenges in efficiently managing and accurately categorizing a high volume of public inquiries. This inefficiency leads to delayed responses and potential misallocation of resources.
“We felt really strongly that innovating our 311 request line process would not just improve citizen engagement, but also majorly boost efficiency,” said Cathy Grossi, who was VP of Product Management at the time.
311 requests can vary significantly. Anything from a barking dog to a newly-discovered pothole can fuel a submission. A common scenario, however, involves a citizen who is unclear on where to make their request, so they might search by department on their city’s website, make their best guess, then send a generic email and cross their fingers for a response. Oftentimes, requests end up – with too few details – in the wrong departments. The added effort of re-routing requests and collecting additional details bogs down the process resulting in slower responses and less satisfied citizens.
Accela surmised that applying GenAI to this challenge would eliminate the ambiguity around submitting 311 requests AND improve the efficiency in responding to them.
Why Tribe AI?
Accela’s internal data engineering team had an idea about the transformative potential GenAI could bring to their business but was already at capacity implementing ongoing product maintenance and innovations, so they sought the guidance of a third party GenAI expert they could lean on in this endeavor. Through a referral from their trusted private equity investor Francisco Partners, Accela began work to build a customized solution that would leverage GenAI and give their organization a hands-on proof-point that they could model future projects for.
Developing the Use Case
The two teams came together for a four-week project aimed to innovate the 311 request line, focusing on improving accuracy, efficiency, and resource utilization.
Streamlined Submission for Citizens
Submitting a 311 request is often viewed as a cumbersome and inefficient process for citizens. Whether there’s uncertainty regarding where to submit the 311 request, impatience with operator hold times, or frustration regarding delays related to misrouted requests, citizens have many reasons to forgo submitting a 311 request. Yet, 311 submissions can be tremendously helpful to municipalities and give citizens the ability to voice their needs and concerns.
More Efficient Routing & Reponses for Municipalities
Municipalities often have an operator on staff 24/7 to aid in receiving and routing 311 request line submissions. Because of the volume and variance in requests received, this process can be quite clunky and time-consuming. Call hold times and misrouted requests are key contributors to this disjointed process.
The notion of 311 line inefficiencies is not new. In an effort to help reduce call volumes, some cities have created mobile apps that they encourage citizens to use when they initially call in to 311. However, when you download the mobile app, there are dozens (sometimes over 100) different categories to choose from to classify the request, which can be challenging to navigate.
The Accela team was able to provide common 311 request examples, then categorize and map them to the appropriate department for response. This effort enabled the GenAI-powered chatbot to take natural language inputs for citizens and infer intent – or ask additional questions to uncover it – then capture all the necessary data before routing the request to the appropriate department for resolution.
“We wanted to create a single, user-friendly interface where citizens could easily report issues, which would then be automatically routed to the appropriate municipal department. The goal was to streamline the process and reduce the manual effort required to handle these requests”, says Grossi.
Proposed Solution
In collaboration with Accela, a team of experts from Tribe AI came together for a four-week engagement to conceptualize, design, and develop a POC solution. This work included deep discovery, scoping and building the prototype to prove value, and creating a product plan with the next steps to evolve and productionize the tool.
The proposed solution – a chatbot built leveraging Tribe’s internal platform tooling that has been designed to accelerate AI product delivery - would utilize GenAI, LLMs, and goal-oriented staging to perform a set of predetermined functionalities centered on the accurate categorization and submission of citizen 311 requests.
Tech Stack Details
Full-stack cloud-based application working alongside the existing Accela environment:
- Cloud: Amazon Web Services (AWS) - Bedrock
- Large language models: Anthropic Claude
- Languages used:
- Frontend: Typescript & React
- Backend: Django
How it works:
Citizens open the chatbot, use natural conversational language to share their issues or requests and then the chatbot utilizes multiple layers of large language models (LLMs) and goal-oriented staging to guide the conversation through a predetermined architecture. The chatbot may ask additional questions in order to categorize the request and uncover all the information needed to accurately route the request including all necessary details.
The solution leverages Tribe AI’s proprietary chatbot functionality which gives the teams a jump start in rapid prototyping a custom solution.
“I wasn’t expecting the chatbot to be multilingual, especially in the POC. That was definitely a bonus feature for us because some municipalities are actually required by law to accommodate a variety of languages,” said Grossi.
Tribe AI “Community Connect” Solution for Accela provides multiple out-of-the-box features:
The solution is multilingual
The solution is friendly & understanding
The solution understands nuance
Accela’s Experience Working with Tribe
“The Tribe AI team has this really smart process that helps them quickly and precisely move through the discovery phase. The technology piece is second nature for them so once they understood the challenge, they could very effectively apply technology and iterate quickly with us,” said Grossi.
Cathy applauds the Tribe AI team for their attentiveness. Because of the time they put into understanding Accela’s challenge – by asking thought-provoking questions and documenting key success metrics – the Tribe AI team was able to perform much of the work on their own. This allowed for more significant progress when the two teams came together to demo the prototype.
“Amazingly, they had a prototype up and running in less than two weeks. Seeing the solution live made it so much easier to share feedback and iterate. In the end, we were able to get the prototype routing with 95% accuracy, which is extremely impressive,” said Grossi.
Tribe Team Members
Patrick: Product Lead
Alex: AI Advisor
Nick: Technical Lead
Orges/Will: Engagement Manager
Impact
“A couple of our customers tried the prototype and predicted it could reduce the manual effort and costs associated with operating a traditional 311 contact center by 30-40%,” says Grossi.
Cost-savings predictions aside, two metrics: ‘categorization accuracy’ and ‘time to submit request’ were defined to help assess the impact of the project.
Accuracy of categorization:
Accela sought to have 311 request line submissions categorized at an accuracy level as close to 100% as possible. At the end of the four-week engagement, the prototype was delivering a 95% categorization accuracy level.
Time to submit request:
Accela wished to improve the current speed of 311 request line submissions. Some submission instances were recorded to take as long as 10-15 minutes, due to call hold times or missing request information. The POC demonstrated an average of just 70 seconds to submission using the prototype.
The solution in action
Categorization + Performance
The Future
There was a list of functionality the two teams wished to explore that were out of scope for the four-week POC engagement. Additionally, some functionality wasn’t possible without moving into the deployment and integration phases.
Future functionality considerations:
- Voice agent to also accept audio request submissions
- Location validation through the integration with a mapping/GIS tool
- App based option to enable on-the-go, request submission
- Land management tools to allow for seamless permitting
- Image upload & reading functionality that can pull required details from the image
