A Step-by-Step Guide to AI in Business (with a real case on enhancing Customer Service quality)

AI Implementation Canvas: a practical tool

The Canvas acts as a bridge between vision and execution. It helps organisations turn general ideas about how AI could be applied into concrete actions while taking all the critical aspects of the project into account. With its support, companies can plan implementation comprehensively, engage stakeholders, analyse risks and resources, and measure outcomes.

The structure of the Canvas is built around three main areas that cover the full scope of the project:

1. Problem and solution vision – understanding exactly what we want to achieve and which challenges AI should address.
2. Risk and resource management – assessing potential threats, available competencies, budget, and infrastructure.
3. Project execution – translating plans into specific action steps, supported by timelines and success metrics.

This tool organises the process, improves communication across teams, enables progress tracking and ensures activities remain aligned with business objectives and organisational values.

AI in Customer Service

To illustrate what the implementation of AI looks like in practice, let’s consider a medium-sized service company that is struggling with common customer service challenges:

• long waiting times for customers trying to reach the call centre,
• low first-contact resolution rates,
• difficulties ensuring smooth communication between physical and digital channels.

• What are the project’s key milestones, and how should they be defined in precise terms?
• How can we realistically estimate the time required for each phase?
• Which activities depend on each other, and how are they linked?
• How should we factor in time for testing, adjustments and unforeseen issues?
• How can we strike the right balance between implementation speed and solution quality?
• How should we account for business seasonality and other external factors that could affect the project?

Answering these questions helps to create a realistic and flexible plan, which minimises the risk of delays and increases the likelihood of achieving the intended business objectives.

AI implementation schedule

Phase 0: Preparation and Project Definition

The preparation phase forms the foundation of the entire implementation process. If this stage is skipped or not given sufficient attention, the project risks failing and may remain nothing more than a declaration of intent. The key here is to develop a thorough understanding of the business problem, gather the right data and build a capable project team.

Before starting, it is worth asking a few key questions to properly guide the work:

• What specific business problem are we trying to solve?
• Who will be the end users of this solution?
• Do we have access to the necessary data?
• Is that data of sufficient quality?
• Who will make up our project team, and do we have the competencies required to deliver such a project?

This phase should result in the following: a clearly defined business problem; an initial vision of the solution; identification of key stakeholders; and formation of a core project team, comprising both internal and external members.

Unfortunately, many organisations try to skip this stage and jump straight to technological implementation. Yet solid preparation is precisely what serves as the foundation of success and the condition for achieving a real return on investment.

At this phase, we focus on several key areas of the AI Implementation Canvas:

1. Problems and opportunities in AI implementation

Preparing this section is a crucial step that helps us to identify exactly what we want to achieve and whether artificial intelligence is the right solution. In our customer service example, the team began by gathering hard data and talking to people directly involved in the process. Their activities included:

• analysing customer service metrics from the past 6–12 months,
• conducting interviews with agents, supervisors, and other team members,
• reviewing samples of recorded phone calls,
• transcribing and analysing email content,
• mapping the customer journey and evaluating service quality in key contact scenarios.

Through this, four main challenges were identified:

• long waiting times for contact,
• low first-contact resolution rates,
• high service costs,
• low customer satisfaction.

However, this stage is not just about listing problems ‘at first glance’. It is crucial to verify that we are identifying the root causes rather than just the symptoms. Often, what appears on the surface stems from deeper organisational or process-related issues.

It is important to involve frontline employees from the outset, as they are best placed to understand everyday challenges and identify the root causes of problems. Collaborating with external experts can also be valuable, as they can help to structure the collected data and validate priorities.

2. Users and stakeholders

This stage is often overlooked, yet it is crucial because implementing AI is not just about technology. First and foremost, it is about the people who will use and support it. This is why both end customers and internal project sponsors should be considered.

In our customer service example, we carried out the following activities:

• stakeholder mapping to identify who is involved in the project and what influence they have (e.g. decision-makers, observers),
• customer segmentation (if the end users are external clients) or employee segmentation (if the users are internal),
• creation of personas and user profiles, such as agents and customers,
• initial interviews to better understand the needs, concerns, and expectations of each group.

• defining roles such as Product Owner, Data Scientist, ML Engineers, Developers, Researchers, UX Designers, and Subject Matter Experts,
• assessing the competencies already available within the organisation,
• identifying gaps and planning to fill them with the help of external experts,
• addressing the team’s training needs so they can independently maintain and further develop the solution after implementation.

This approach enabled the formation of a team that not only deployed the technology, but also ensured that it remained aligned with business goals and user expectations.

Building the right team at the very beginning of the project increases the chances of success and helps avoid problems in later stages of implementation.

Phase I: Proof of Concept

Once the preparation stage is complete, we can verify the assumptions in practice. The Proof of Concept (POC) phase enables us to ascertain whether artificial intelligence can solve the problems identified earlier and deliver the anticipated benefits. This stage typically takes several months and requires significant team involvement.

In our customer service example, POC activities included:

• selecting a simple, measurable problem that could be solved relatively quickly,
• preparing and cleaning the data needed to build the model,
• reviewing available technologies and choosing the optimal solution,
• building a prototype and testing it on a selected data sample,
• analysing business processes to ensure the chosen solution fits the organisation’s context.

The goal of a POC is not only to confirm that the technology works but also to validate that the chosen approach makes business sense and is feasible in a real environment.

This is the stage at which the project starts to incur costs and requires specific technological decisions to be made. It is also an opportunity to identify potential risks and prepare for the next steps: piloting and full-scale implementation.

A reliable assessment of project feasibility can be achieved through the following actions:

• choosing the AI solution that is the simplest option and can deliver quick, visible business results,
• preparing the data by cleaning, organising, auditing and labelling it. This is often the most time-consuming but critical part of the proces,
• analysing business processes to select the right areas for POC,
• building a prototype and testing it on a data sample to verify whether the model works without involving users at this stage.

1. How to choose an AI solution proposal

After testing the assumptions within the proof of concept, the next step is to select a specific solution. At this stage, the team decides which technologies and approaches best address the identified problems and can realistically be implemented within the organisation.

In our customer service example, we carried out several activities to develop the optimal proposal:

• reviewing available market technologies dedicated to customer service,
• running Design Studio workshops with representatives from business, IT, and customer experience teams,
• jointly developing the concept of the solution,
• selecting the best-fit technology and preparing initial AI models.

As a result, the chosen solutions addressed specific challenges: chatbots and voicebots for communication, sentiment analysis systems, and recommendation engines for agents.

A well-chosen solution proposal meets current needs and fits into the organisation’s long-term strategy, enabling growth in line with its future direction.

2. How to define a value proposition for AI implementation

At this stage, the team should provide a clear description of how the technology will contribute to the organisation achieving its goals, and of the measurable results. In our customer service example, the activities included:

• stakeholder workshops to analyse problems and outline the initial value of the solution,
• verification of business hypotheses based on POC results,
• preparation of initial financial models showing potential savings and revenue,
• updating the value proposition after analysing test data.

This approach ensured that the value proposition was based on hard data and real business needs, rather than assumptions.

A well-prepared value proposition justifies the project to management or sponsors and clearly defines how success will be measured and evaluated. It provides a solid basis for deciding whether to proceed to the next stage of the project.

3. How to prepare data for AI implementation

Data is the foundation of every AI project. Without quality and availability, even the best technologies will not perform correctly. That’s why, at this stage, it is crucial to determine the source of the data, its condition, and the actions needed to prepare it for use in the models.

In our customer service example, the activities included:

• identifying data sources and formats (e.g. call recordings, emails, chats, CRM and helpdesk systems),
• mapping the data required for the selected models,
• analysing data quality and identifying gaps in available resources,
• checking compliance with legal requirements, including GDPR,
• preparing the data: transcription, anonymisation, categorisation, integration with other systems, and securing it in line with security standards.

Thanks to these actions, the team could be confident that the AI models would be based on reliable, up-to-date, and representative data.

Although preparing the data is often the most time-consuming part of the project, it is also the most rewarding. Well-executed work at this stage significantly increases the chances of the entire implementation being successful.

4. How should technology be approached in AI implementation

At this stage, the goal is to select the right tools that fit the organisation’s needs and technical capabilities.

In our customer service example, the activities included:

• analysing the existing infrastructure and systems—such as CRM, helpdesk, and communication platforms,
• reviewing available market solutions and comparing them with project requirements,
• assessing the possibilities for integrating new tools with existing systems,
• selecting specific technologies that met both functional and legal requirements.

As a result, the chosen technologies included NLP models for sentiment analysis, speech recognition and synthesis systems for voicebots, cloud-based solutions, and integrations with existing platforms.

Taking a thoughtful approach to technology helps to avoid complications further down the line and ensures that the chosen solutions solve current challenges and support the organisation in achieving its long-term goals.

Phase II: Pilot and Testing

After the POC phase, which involves validating the assumptions, the next step is the pilot, which involves implementing the solution in a controlled environment. This phase tests how the technology performs in a real-world setting and delivers results before it is rolled out on a large scale.

In our customer service example, the activities included:

• defining the scope of the pilot and selecting a representative group of users (e.g. a specific communication channel or customer segment),
• organisational preparation: training participants, planning how feedback will be collected and analysed,
• launching the solution on a limited scale,
• systematically collecting data and user feedback to evaluate effectiveness, efficiency, and satisfaction levels.

This approach enabled potential issues to be detected before the solution was implemented across the entire organisation, while also providing evidence to confirm its business value.

The pilot phase allows not only for testing the technology but also for assessing its impact on daily work, identifying potential resistance, and better preparing the organisation for full-scale implementation.

1. Identifying challenges and risks

During the pilot phase, it is important to take the time to thoroughly identify any potential challenges or risks that could affect the project’s success. Addressing these issues early on enables the team to prepare mitigation measures in advance, thereby increasing the likelihood of effective implementation.

In our customer service example, the activities included:

• analysing risks already identified during the POC and pilot phases,
• developing a mitigation plan,
• examining employee attitudes toward the new solutions to adjust communication and support,
• preparing mechanisms for quick responses to technical and organisational issues.

The most common challenges and risks were:

• Technical challenges, e.g. integration with existing systems and low speech recognition accuracy;
• Organisational challenges included resistance to change and shifts in team competency profiles.
• Legal and reputational risks included deterioration of the customer experience, GDPR non-compliance and lack of user acceptance of AI.

Thoroughly identifying risks and preparing a management plan means the organisation is better prepared for unexpected challenges and can move towards full implementation with greater confidence.

2. Ethics and security

This is not only a legal requirement, but also a way of building trust among customers and employees, while ensuring that technology operates in line with the company’s values.

In our customer service example, the activities included:

• analysing compliance with GDPR and other applicable regulations (e.g. the AI Act),
• discussing the potential ethical implications of AI implementation,
• consulting with legal advisors and data protection experts,
• preparing principles for customer transparency and human oversight of the system.

Fairness: continuously monitoring and eliminating bias in models and ensuring equal treatment of all customer groups.

Human oversight: defining when and how humans can intervene in the process.

Accountability: assigning clear responsibility for decisions made by the AI system.

Addressing these elements already during implementation minimises risks, builds trust, and ensures the company operates in compliance with both ethical standards and legal requirements.

Phase III: Full implementation

This is the final stage of the project, involving the full-scale rollout of the solution across the organisation. At this point, the technology will stop being a pilot experiment and will begin to genuinely support teams in their daily work, delivering measurable business benefits.

In our customer service example, the activities included:

• extending the implementation to all communication channels and relevant customer segments,
• preparing a detailed schedule and change management plan,
• conducting training sessions for all system users,
• monitoring key performance indicators (KPIs) in real time to continuously assess implementation results,
• ensuring compliance with ethical, legal, and internal procedures.

Full implementation requires careful consideration of communication and employee support, as it often represents the most significant organisational change in years. It is equally important to maintain stakeholder engagement and allocate sufficient resources for the system’s ongoing development.

This phase results in a fully operational solution that delivers the expected benefits and supports both employees and customers. It also provides a solid foundation for further development and optimisation.

Success factors

In summary, the most important factors that determine the success of an AI project should be highlighted. These ensure that the technology delivers tangible benefits and is positively received throughout the organisation.

In customer service, the key factors proved to be:

A strategic approach to data: data is the foundation of the entire project. Before choosing a technology, it is essential to audit the data, ensuring its quality and integration, as well as planning for its continuous improvement.

Iterative implementation and proof of concept: instead of rolling out across the entire organisation at once, it is better to start with a specific use case and conduct a proof of concept (POC). Quick, measurable successes build trust and help win stakeholder buy-in.

Balancing AI and human roles: AI should support people, not replace them. Clearly defining which processes are worth automating and which should remain in human hands helps to maintain a positive organisational culture while ensuring high efficiency.

Change management: Preparing the organisation for change from the outset increases the likelihood of acceptance and full adoption of the new solution.

Ongoing monitoring and development: these are essential after implementation. Systematically gathering feedback, monitoring KPIs and planning regular updates ensures the solution evolves alongside business needs.

An ethical approach: transparency, accountability and data privacy protection minimise legal risks and build trust among customers and employees.

Together, these elements create a solid foundation for implementation that delivers real results while staying aligned with the organisation’s values. Companies can use them to unlock the full potential of artificial intelligence in a responsible and profitable way.