Agentic AI in banking

Key information

Possible delivery options: The training can be delivered over one, two or three days, depending on the team’s preferences and how deeply participants want to go into the implementation aspect.

Tech stack: The programme is built on Microsoft Copilot Studio. A variant delivering the same objectives on the n8n platform, which can be run on the organisation’s own server, is also possible.

The training prepares IT teams to independently design, build and maintain AI agents in Microsoft Copilot Studio, integrated with business processes through Power Automate. Participants work on realistic scenarios from the banking sector: from handling enquiries and requests, through automating forms and documents, to classifying and analysing data.

During the training, participants will find out, among other things:

• How generative AI works and what its limitations are
• What agentic AI (Agentic AI) is and what benefits it brings to the banking sector
• How to create effective prompts using the KREDO method
• How to design an AI solution tailored to a specific process at a bank
• How to build a knowledge agent (RAG) fed with technical documentation and procedures
• How to configure conversation logic, clarifying questions and answer quality control
• How to build an autonomous agent with a trigger and a multi-step workflow
• How to integrate an agent with business processes through Power Automate
• How to use AI responsibly and safely in a regulated environment and what obligations the AI Act imposes

The training works with Microsoft Copilot Studio and Power Automate. All exercises and workshops refer to realistic processes found in the banking sector.

Objective

The objective of the training is to introduce participants to designing, building and maintaining AI solutions based on Microsoft Copilot Studio and Power Automate, set in realistic banking sector processes. The training covers working with knowledge agents (RAG), autonomous AI agents and automating workflows related to handling enquiries, forms and documents. The emphasis is on practical workshops, building awareness of the possibilities of agentic AI and developing the IT team’s competence as the owner of AI solutions.

Programme

The programme comprises 3 training days in a workshop format.

Day 1: AI basics, security and the first knowledge agent (RAG)

Module 1: Fundamentals of generative and agentic AI

• Participants share their experience with AI
• Key concepts: prompt, token, context, hallucination
• How to verify AI-generated content
• Comparing the key AI models available on the market
• What agentic AI (Agentic AI) is and how it differs from a chatbot and an assistant
• Anatomy of an effective prompt: the KREDO method
• Workshop: writing prompts using examples from the work of a bank’s IT team

Module 2: Security, ethics and compliance (AI Act)

• AI security in the context of a financial institution
• What can and cannot be entered into AI models: the limits of confidentiality and banking secrecy
• Risks: incorrect data, hallucinations, bias, information leaks
• The AI Act: the EU regulatory framework and its significance for the banking sector (risk categories, obligations of the provider and the deployer of an AI system)
• The 4D model of AI user responsibility
• A checklist for safe and compliant use of AI in an organisation

Module 3: Designing agentic AI solutions in banking

• From problem to solution: identifying processes at the bank to automate
• AI assistant vs AI agent: differences and uses
• Overview of examples: AI agents in banking processes (customer enquiry handling, document analysis, IT team support, test automation, internal helpdesk)
• Workshop: mapping processes in the participants’ organisation for automation

Module 4: A knowledge agent (RAG): configuration and feeding in knowledge

• Introduction to Copilot Studio: interface and key features
• Building, describing and configuring a RAG agent (Knowledge Only)
• Adding knowledge: technical documentation, regulations, procedures, policies
• Adding knowledge: external sources (intranet, knowledge bases, documentation repositories)
• Integrating the agent with knowledge specific to the team or the banking domain
• Workshop: building an agent fed with the participants’ documentation

Day 2: Refinement, autonomous agents and integration with processes

Module 5: Conversation logic, quality control and testing

• Clarifying questions: how the agent gathers context from the user
• Controlling the scope of responses: restricting the agent to a specified domain
• Quality control: checking the correctness and consistency of answers
• Testing the agent and analysing errors
• Workshop: testing and refining the agent

Module 6: Agent conversation and orchestration

• Topics, triggers and conversation orchestration
• Customising the conversation’s starting topic
• Creating and configuring new conversation topics
• Workshop: designing conversation scenarios for banking processes

Module 7: Maintaining and developing AI solutions

• Updating the agent’s knowledge base: when and how
• Testing and improving the quality of agents over time
• The team’s responsibility as the owner of an AI solution
• Development strategy: from the first agent to an ecosystem of AI solutions at the bank

Module 8: An autonomous AI agent: building and configuration

• Introduction: from a knowledge agent to an autonomous AI agent
• Building, describing and configuring the instructions of an autonomous agent
• Launching an autonomous agent: demonstration and tests
• Workshop: configuring an autonomous agent for a banking process

Module 9: Integration with business processes through Power Automate

• Handling requests: form, trigger, automatic processing
• Creating agent tools and automating the flow of information
• Automatic processing of information and AI results
• Passing results on to the next steps of the process
• Workshop: building a request-handling workflow (form, processing, notification)

Day 3: Document automation and implementation workshops

Module 10: Working with documents and data in automations

• Analysing documents using AI (contracts, applications, documentation)
• Summarising and classifying data
• Automating the processing of forms and attachments
• Workshop: automating document processing

Module 11: Project workshops on participants’ processes

• Workshop-style work on real use cases from the participants’ organisation
• Designing AI solutions from needs analysis to a working prototype
• Individual and group consultations with the trainer
• Summary and directions for further development

Learning outcomes

After completing the training, participants:

• Understand what agentic AI is and what benefits it brings to the banking sector
• Create effective prompts using the KREDO method
• Identify processes at the bank suitable for automation with AI
• Build and configure knowledge agents (RAG) in Copilot Studio, fed with documentation
• Design conversation logic and clarifying questions, and control the quality of the agent’s answers
• Build autonomous AI agents with triggers and multi-step workflows
• Integrate agents with business processes through Power Automate (forms, documents, notifications)
• Independently maintain and develop AI solutions as business owners
• Understand the requirements of the AI Act and apply the principles of responsible and safe use of AI in a regulated environment

Trainer

Franciszek Kołpanowicz: entrepreneur and certified AI trainer who helps leaders and teams become more effective through intelligent technologies. He has been published in Rzeczpospolita, among other places. He has more than 2,000 hours of work with generative AI behind him and shares his knowledge at public events, including the “Manager in the Age of AI” training organised by Rzeczpospolita, as well as at the UKEN and PASE conferences.

His background in philosophy (Jagiellonian University) and his experience in IT quality control allow him to combine analytical thinking with asking the right questions. His own practical journey with AI showed him how this technology can become a real lever for growth.

Drawing on hands-on experience, he shows leaders and their teams how to use AI not only to work more efficiently but also to think more clearly and make faster business decisions. His AI training has, among other results, cut the time needed for routine analysis of multi-page legal documentation by 80%.

In his training sessions he pays particular attention to the participants themselves, drawing on their experience and needs as the work progresses. As a result, his sessions are not repetitive routine but an enriching and individual experience.

His mission is to make artificial intelligence a practical tool for development: business, intellectual and cultural.

Prerequisites

• Basic computer skills and the ability to work in the Microsoft 365 environment
• Knowledge of the Microsoft 365 environment at user level
• A Microsoft 365 Copilot licence (Copilot Pro) for every participant
• Access to Microsoft Copilot Studio (account activation before the training)
• Access to Power Automate as part of the Microsoft 365 subscription
• No prior experience in building AI solutions is required

Organisational requirements on the client’s side

• Before the training, every participant has a Microsoft 365 account with a Copilot licence that has been set up and tested
• Login checked in advance on the device the participant will work on
• An active Microsoft Copilot Studio account (full activation can take up to 24 hours)
• Access to Power Automate verified before the training
• One workstation per person: a laptop for each participant, not shared

The training works with one tool ecosystem: Microsoft Copilot Studio and Power Automate, so that participants have time to genuinely master it in the context of their own work.

To be confirmed before the training

Licences and access:

• Does every participant hold a Microsoft 365 Copilot licence (Copilot Pro)? A Microsoft 365 account alone is not the same as Copilot Pro.
• Does every participant have their own account, or do participants share one? (a shared account makes it impossible to complete the exercises)
• Does every participant have access to Microsoft Copilot Studio?
• Is access to Power Automate active under the subscription?

Participants’ technical readiness (critical):

• Every participant arrives with a logged-in account on their own laptop: we do not set up accounts in the room
• Test before the training: every participant opens Copilot Studio and checks that they can create an agent and ask a question
• If any participant does not have access: advance notice will allow us to prepare an alternative route

Infrastructure:

• Stable Wi-Fi tested under load (all participants working simultaneously in the browser and in applications)

Verifying access to AI agents in Microsoft Copilot (critical):

Before the training it is necessary to check with participants (or with one representative participant, if identical permissions are assumed across the group):

• Can knowledge from SharePoint or company documents (internal sources) be connected to the agent?
• Can a participant create their own agent, and does that agent answer questions?
• Do all participants have the same level of permissions?

Each of these elements is controlled by a separate switch in the Microsoft 365 admin centre: in practice, a user may see an agent on the list but lack the permissions to talk to it. For this reason, confirming that agents are merely present in the interface is not enough.

Verification can be carried out in three ways: (1) confirmation from the client’s IT administrator, (2) a short online session with the trainer before the training (a separately priced service), (3) commissioning a full technical verification.

Without confirmation of these points, the agent-building workshops may turn out to be impossible to deliver in full: in that case the hands-on part will be replaced with a presentation.