Case Study

An AI Knowledge Assistant for a Municipal Contact Center - Brokers and Residents

Softwise.AI partnered with the City of Poznań to build an AI-powered knowledge assistant for the Poznań Contact Office - the city's municipal call and contact center. The system gives brokers a natural-language search interface across roughly 600 documents drawn from many administrative units, returns answers with cited sources, and supports residents directly via a 24/7 public chatbot first launched during the COVID-19 pandemic. Brokers spend less time hunting through documents and redirecting calls, residents get answers any hour of the day, and the bot keeps learning - automatically expanding the topics it can handle.

CLIENT

Poznań Contact Office

YEAR

2024

INDUSTIRES

Public Sector - Municipal Administration & Citizen Services

DELIVERABLES

AI Knowledge Assistant for Brokers, 24/7 Public Chatbot for Residents, RAG Across ~600 Documents, "Teacher Mode" for In-House Knowledge Curation, Usage Analytics

The problem

The Poznań Contact Office acts as the City of Poznań's municipal call and contact center. It informs residents about where, when, and how to settle matters at the City Hall - which documents and forms are required, what fees apply, which office to visit. It also books appointments, runs telephone surveys, and supports outgoing information campaigns. In effect, the Office is the front door to the city for residents, and a routine-matters layer that lets substantive departments focus on the work only they can do.

That is a wide remit handled by a single team. Legal regulations, benefits, fees, and required documents change constantly. Information sits across many administrative units. Most queries used to end with a transfer to a specific office because brokers could only safely provide general information. And during the COVID-19 pandemic, demand for accurate, accessible answers spiked exactly when in-person service collapsed.

The City asked Softwise.AI for a system that would give brokers fast, sourced answers across the entire municipal knowledge estate, expand what they could resolve at first contact, and serve residents directly outside office hours - while staying simple enough that every member of the team, including employees with certified disabilities, could use it day to day.

{Key Challenges}

Why a single municipal hotline covering hundreds of topics couldn't scale answer quality?

Discovery with the Office surfaced an interlocking set of operational constraints:

Constantly changing regulations
Legal regulations, benefits, fees, and required documents change frequently. Keeping every broker current across the full landscape - manually - was effectively impossible.
A very wide topic range on one hotline
A single team handles questions across the entire scope of city administration, from civic registry to fees to bookings - a span that no individual broker can keep in their head.
Heavy reliance on call redirection
Because brokers could safely provide only general information, a large share of calls had to be redirected to specific offices - increasing wait times and pushing routine load onto substantive departments.
Knowledge spread across many administrative units
Source documents lived in many different city offices, each with their own publication and update rhythms.
Accessibility for all employees
The Office employs people with certified disabilities. Whatever was deployed had to be unambiguously user-friendly and intuitive - not a layered enterprise tool.

{Process}

How we built an AI knowledge assistant for both brokers and residents in 5 stages?

Softwise.AI delivered the engagement in clear stages, working closely with the Office to consolidate scattered municipal content into a single, usable knowledge base before any AI was deployed on top of it.

1. Topic mapping with the Office

The team worked with the Contact Office to map the actual flow of citizen queries, identifying the topics that drove the highest call volume, the highest redirection rate, and the highest broker frustration.

2. Document aggregation across administrative units

Roughly 600 documents were collected from the many City Hall units that own them, normalized, and consolidated into a single knowledge base - giving brokers, for the first time, one place to look.

3. Building the broker-facing AI knowledge assistant

A natural-language search layer was built on top of the knowledge base, returning answers grounded in source documents, with citations and the ability to display graphical elements and links to external resources for clarity.

4. Adding "Teacher Mode" for in-house knowledge curation

Brokers with special "Teacher" permissions were given the ability to train the bot on additional issues and correct its answers using a simple text editor - so the system improves continuously, driven by the people closest to the citizens, not by a vendor change-request queue.

5. Public-facing chatbot during COVID-19

During the pandemic, the Office launched an external version of the bot to serve residents directly, 24/7, helping them find information on administrative matters when offices were closed. The same knowledge base now powers both internal brokers and external citizens.

Solution

The system is a unified knowledge platform serving two distinct user groups - internal brokers handling phone, email, and walk-in queries, and external residents using the public chatbot - on top of a single, continuously curated knowledge base.

AI knowledge assistant for brokers

Brokers query the system in natural language and receive answers drawn from the underlying ~600 documents. Each response includes the source document it came from, plus - where relevant - graphical elements and external links. For general questions outside the bot's direct scope, the system routes the broker to the correct answer or office.

"Teacher Mode" - continuous improvement by the Office itself

Brokers with "Teacher" permissions can extend and refine the bot's knowledge through a simple text editor - adding new issues, correcting answers, and capturing institutional knowledge as it accumulates. The Office is not dependent on Softwise.AI for ongoing content updates.

24/7 public chatbot for residents

A public-facing chatbot, available on the City Hall website, lets residents ask administrative questions any time. The bot is testable live on the City of Poznań's Contact Office website.

Automatic analytics and learning

The system collects analytical data on the questions being asked. Office staff use those insights to systematically expand the scope of issues the bot can answer automatically - so the share of queries fully resolved by the bot grows over time.

Accessibility by design

The interface was designed to be simple and intuitive for every member of the Office team - including employees with certified disabilities - because the value of the system depends on every broker actually using it.

Roadmap: CRM and call-center integration

Planned next steps include integrating the system with the city's CRM and call-center platform, so brokers can access both the knowledge base and the service panel in one place - and so some CRM fields can be filled automatically based on the queries brokers enter.

{Results}

First-contact resolution across ~600 documents and 24/7 public service

The system is live and supporting the Office every day:

~600 documents consolidated into a single, searchable knowledge base across many administrative units.
First-contact resolution for most cases - significantly fewer call redirections to substantive departments.
More queries handled per broker - measurable efficiency gain on the same headcount.
A 24/7 public chatbot for residents, live on the Poznań City Hall website.
Automated knowledge-base updates drawing from various external sources.
Automation of the most popular queries, with the scope expanding continuously through Teacher Mode and usage analytics.

{Business impact}

Better citizen service and broker efficiency without growing headcount

Higher first-contact resolution. Most cases now get resolved at the first contact rather than escalating to a specific office - reducing wait times and freeing substantive departments.
More throughput per broker. Same headcount handles more queries, with less time spent searching across scattered documents.
Citizens served outside office hours. The 24/7 public chatbot gives residents access to municipal information any time, an outcome that became critical during the pandemic and has only grown in value since.
Institutional knowledge that doesn't leak. Teacher Mode means the Office captures its own expertise into the system as new topics arise - rather than losing it to staff turnover.
A foundation for deeper integration. Planned CRM and call-center integration will let brokers work from a single interface and reduce manual data entry further.

{Key recommendations}

5 principles for AI knowledge assistants in public-sector contact centers

A few principles emerged from this engagement that apply to any municipal or government contact center deploying AI:

Aggregate the knowledge base before deploying the AI. Half the value of this project came from getting 600 documents from many city units into one searchable place. The AI is what makes that estate usable, not what creates it.
Cite the source on every answer. In public administration, every piece of information given to a citizen must be defensible. A bot that can point to the underlying document is one a city can actually let people use.
Give the staff the keys. Teacher Mode turns the contact center team into the long-term owners of the knowledge base - which is the only sustainable way for a public-sector AI system to stay accurate.
Design the interface for everyone on the team. A tool that some employees can't use isn't a productivity gain; it is a new silo. Simplicity and accessibility are not nice-to-haves in this context.
Use analytics to grow the bot's scope, not just measure its usage. The most valuable analytics output is the list of unanswered or poorly answered questions - that is the city's next content backlog.

SUMMARY

Softwise.AI delivered a unified AI knowledge platform for the City of Poznań's Contact Office - consolidating ~600 documents from across municipal administration into a single source for brokers, layering natural-language search with cited answers on top, giving Office staff Teacher Mode to extend the bot themselves, and powering a 24/7 public chatbot serving residents directly. The system raises first-contact resolution, lets the same broker headcount handle more queries, and gives the City a foundation it can extend - through planned CRM and call-center integration - well beyond the original brief.

TESTIMONIAL