AI systems > Talbotti
Talpa Helsinki





chatbot talpa customer service financial management services

Talbotti is an electronic contact channel introduced by the City of Helsinki’s financial management service.

The purpose of Talbot is to improve and increase customer contact opportunities, also outside customer service hours.

Chatbot answers questions around the clock and offers, within the opening hours, the possibility to transfer to live chat to continue the conversation with a customer advisor. 

The new contact channel enables self service for customers and also advises, provides up to date information, links to websites and speeds up the search for information for residents and city employees. The chatbot learns from the data collected in customer service and it can answer most common questions related to  invoicing and financial management services.

Contact information

  • Department

  • Financial Management / Customer Service
  • Contact team for inquiries

  • Tuire Järvinen
  • External suppliers

  • IBM
  • Contact email

  • talpa.asiakaspalvelu@hel.fi
  • Contact phone

  • 09-31025300

More detailed information on the system

Here you can get acquainted with the information used by the system, the operating logic, and its governance in the areas that interest you.

DatasetsShow More Show Less

Key data sources utilised in the development and use of the system, their content and utilisation methods. The different data sources are separated by subheadings.

AI training material

The Ai training material includes sample questions from the content area of the service. The questions have been collected during the development project from the City of Helsinki service experts and they represent the questions citizens ask in customer service situations. The material is owned by the City of Helsinki, but it has not been licensed. The material does not include personal data.

The training material is uploaded to the AI, which constructs the model based on the training material. Some of the example questions are also used to test the model.

Conversation logs

The system stores the questions and answers given during the service during the service. Saved logs are regularly analyzed in the service development team and shows how the service has answered customer questions. Based on the analysis, the content areas that need to develop the training material and the responses from the service should be found. After the analysis, the training material will be refined and exemplary questions or, for example, new conversation loggers to improve the service experience of the service. Customer feedback can also be used to refine and add information on them.

The conversation logs will be removed from the system after 6 months, after the analysis and changes have been made, but usage-related reporting, such as usage rates, response rates and other parameters, will be stored for long-term development.

The conversation log is owned by the city of Helsinki, but it has not been licensed.

Data processingShow More Show Less

The operational logic of the automatic data processing and reasoning performed by the system and the models used.

Model architecture

The chatbot service is based on the IBM Watson Assistant software product, which is provided as a SaaS application from the IBM cloud from the Frankfurt Data Center in the EU. The service utilizes AI and machine learning models to process natural language questions. 

The customer asks questions through a digital interface that works across different devices. If the AI is unable to answer questions, or otherwise more suitable, the customer can contact the customer service representative. In this case, the service will connect the customer to a service advisor who can continue the conversation. The customer can also give feedback on the service, which is used for further develop of the service.

System architecture description



The development of the use and quality of the service is constantly evaluated. The user statistics and quality meters are regularly stored in reports. Quality meters include the number of correct answers compared to all questions and direct customer feedback collected on the service.

Non-discriminationShow More Show Less

Promotion and realisation of equality in the use of the service.

The service is currently only available in Finnish. 

The service does not automatically know how to correct spelling errors, which may limit its use by customers who are not native speakers.The service is not yet extended to new language versions.

Human oversightShow More Show Less

Human oversight during the use of the service.

The service automatically identifies and compiles topics that it cannot answer. These are submitted to the Financial management service experts to develop the conversation paths of the service further.

The service is monitored by reviewing the feedback and conversation logs and by analysing the statistics on the use of the service. Based on this data, experts identify the topics customers find interesting and then they develop discussion paths.

The monitoring experts are tasked to create new discussion paths based on the analysis and to improve the retrievability of recommendations, e.g. by using new keywords.


Risk managementShow More Show Less

Risks related to the system and its use and their management methods.

The service will never ask for personal information that allows a person’s identification. However, the service involves the risk of processing personal data if the customer enters unnecessary personal data into the service. To manage the risk, the service has an automated function that immediately removes social security numbers or email addresses from the conversation logs. The conversation logs are also regularly reviewed by experts, who delete any personal data that may have been overlooked by the automated system.

Was this information useful?