How to implement AI in business in 9 steps – change can be beneficial

While artificial intelligence will not save us from all the ills of running a business, its smart implementation into an organisation can deliver more than we expect.

Implementing AI models can help streamline processes, automate routine tasks, and solve problems faster. This in turn minimises the risk of errors and increases productivity.

It is important to understand and accept the limitations of AI to avoid inflated expectations and focus on the real benefits.

The implementation process should include a goal-setting phase, followed by a prototyping phase and the creation of an MVP. This allows for incremental testing and adaptation of solutions. In this way, organisations can better tailor AI technology to their specific needs.

Equally important to the stable functioning of an organisation is reducing stress and creating a conducive working environment. In a highly unstable business environment, it is a benefit to be able to minimise the obstacles.

Based on our experience, we have created a description of AI implementation in 9 steps.

1. What motivates you?

First, think about what made you consider implementing AI in business.

FOMO is one of the well-studied cognitive biases, or the mechanisms our brains use to act with less effort (implicitly, more efficiently). Fear of missing out occurs when you check your emails, instant messages, and news sites 50 times a day to make sure you don’t miss any ‘important’ updates about what is happening. We find that FOMO also applies to business and technology. For some leaders, the fear of ‘falling behind’ is driving technological change and, more recently, artificial intelligence-based solutions.

Optimisation is about doing things faster, not wasting time on boring, repetitive tasks. Focusing on the things that matter to grow the business. Finding time to experiment and innovate.

These are two common reasons why companies choose to implement AI and process automation.

Each of these can lead to operational optimisation, freeing up time for creative and organisational development tasks.

If you often ask yourself, “Why do I spend 30 minutes taking notes after every meeting?”, “Do I need to send another contract to an interpreter?” or “Do my staff need to pick up the phone to answer repeated customer questions?”, then you are looking for optimisation in the right place. You can probably achieve these optimisations without spending a lot of money and as part of a package from one of your technology providers.

Then there’s the third and most forward-looking strategy: we look at the problems and needs of the company and its customers, how we can solve them, and whether AI can help us do that. We start with high-level goals and identify where we want to make a difference. This use of AI, as opposed to mundane, simple applications, can provide a competitive advantage, position the company as an innovation leader, and gain the trust of customers.

Assuming there is a problem that can be solved by artificial intelligence is different from looking for a problem to justify implementing advanced tools like Claude or Co-pilot.

We carefully analyse processes, such as observing people working on a production line or closely following the company’s workflow, to diagnose problems and consider which ones can be solved using artificial intelligence.

At Edisonda, we monitor employees as they perform their tasks as part of our AI Discovery research.

2. What do you want to achieve?

“We don’t want to be left behind” is not a concrete goal. It is easier to make changes when we know exactly what we want to achieve.

An example of a well-formulated objective might be to optimise a transport service for large building materials. In detailing this objective, we will focus on organising orders and deliveries so that they arrive seamlessly, on time and in optimum condition.

At Edisonda, we have supported several such projects, ranging from systems that support the delivery of liquid concrete, to ordering large quantities of structural steel, to sharing transport for small, complementary orders from depots to construction sites.

In each case, the use of algorithms and systems to handle large and diverse data sets, and interfaces that can be understood by customers with different levels of expertise, has enabled a quality of service that was previously unavailable to customers in certain markets.

3. Where do you start?

Answer the question: What is the current state of your organisation? Focus your attention on the areas we have identified in this step.

  • What kind of business do you have? What kind of business would you like to have? What do you offer and what would you like to offer?
  • How many clients do you have?
  • Is this number stable? Are they satisfied with what you offer? Do they have needs that you are not meeting? Do they look forward to change or are they likely to be put off by it? What are your opportunities for new business?

If you do not have this knowledge, it is a good idea to start by checking the data you do have, including the number of orders, complaints, opinions about the company and the offer. It is also a good investment to commission research. An independent, external ‘eye’ and ‘ear’ can often help redefine the needs and solutions that will most benefit the business.

For example, we carried out an analysis of their apprenticeship certification services for a large organisation in the UK. This helped the organisation to assess the case for investing in a modern examination and certification system, identifying the key aspects of the service provided and the areas where the investment would bring the most benefit.

Eventually, the organisation decided against building a large, dedicated platform and based the new iteration of the service on simpler, remote solutions that were more accessible to users.

  • What technology do you have? What would you like to have? How could it help you? Can you afford to change?

If the technology is ‘free’ (as AI is often advertised) – what will you pay for it? Can you afford such an expenditure (of money, time, labour, data)?

  • Will the new technology be compatible with your technology stack?

If your company is part of an ecosystem like Microsoft, implementing artificial intelligence into your processes may be easier than you think. Running a Co-pilot costs relatively small amount of dollars a month and can streamline mundane activities like organising emails, summarising meetings or scheduling tasks. Things are different if you work in a distributed technology environment. Then you have the option of customising and developing a solution based on well-known Large Language Models (LLMs) such as Open AI or Lama.

  • Competition – do you have any? Where are they based? What do they do that you don’t? What do they do that you could do better? Where are they already using AI solutions?

This is another area where an outside view can give you new perspectives. Consider outsourcing your benchmarking or competitive analysis to a company that specialises in such services.

The point is not to spend six months documenting the current situation in detail. However, if you believe that customer order processing should be faster and more efficient, you should assess the current situation before making any changes.

Firstly, this will help you to check whether the effort and money invested in implementation is likely to pay off. Secondly, in step 9 you will have a baseline against which to compare when you reach your goal, and in the earlier stages it will be easier to check that you are heading in the right direction.

It is not true that implementing AI is about replacing human employees with a machine that responds to a single sentence with a ready-made solution. This is just another marketing promise with the details in the small print.

4. Where do you start to “keep the boat from capsizing”?

A major change involving significant resources weakens the company if it fails. If it succeeds – it sometimes leaves bitterness, a feeling that the result is less than expected in relation to the commitment.

A big change is often very stressful, even for those not directly involved. In the case of AI solutions, a common fear is that employees replaced by technology will lose their jobs.

It’s easier to believe in success and the value of your efforts when you’ve had the success of a smaller project and the lessons learned from implementing one that didn’t necessarily succeed.

Dare to experiment, not on the assumption that it will succeed, but just to learn.

As AI technologies are still in the development stage and their potential changes from month to month, it may be best to opt for a ‘proof of concept’ solution. This approach involves building a prototype model, usually with less functionality and working on a limited set of data. Such a prototype allows us to confirm that our assumption is having the intended effect and to test the solution in a safe environment. The PoC phase in AI projects is an essential part of the process that we should not bypass.

Experimenting and iterating is also a good idea if you don’t know exactly what you want to achieve, what you’re starting with, what’s missing, what inputs you need, what might slow you down, what might go wrong, or finally: how to convince people (employees and customers) of the new solution.

In short, if you are trying out a relatively small project, this experiment will give you the information you need to plan a bigger change, and you can learn from it to implement the bigger project.

5. Expenses – Conduct a solution review

When comparing costs, don’t stop at the licence price. Look at the cost of maintenance, support, training and implementation of other changes needed to make the solution you are considering work.

For example, SAP implementations require a lot of configuration and only work effectively if they are operated by people who are used to the solution and keep their knowledge up to date.

On the other hand, if you expect data to be entered by people who are unfamiliar with the logic of these systems, they may use an interface that is unintuitive to them. In the best case scenario, teams organise themselves by delegating the person who is most familiar with these systems to complete the applications on behalf of the team.

Similar problems can arise when taking the first steps to work with solutions based on AI models. These tools still require the use of appropriately worded commands to produce a quality result, which may be beyond the skills of employees who have previously relied on the support of professionals to create marketing content, for example.

Careful analysis of processes for potential automation, checking the cost of implementation and then maintaining the solution will be key.

It may turn out that Co-pilot positions only make sense for some jobs, and that process optimisation can be achieved with simple automation that does not require computing power.

Edisonda is already advising companies on how to make the right technology choices.

6. What might your organisation lack (people, skills, technology or ‘something else’)

People and skills

Introducing new solutions and tools always means changing the way we work. For example, the way we are used to asking questions in Google Search is inefficient when it comes to ChatGPT, Co-pilot or Claude. This means that people have to learn a different way of doing things, using these new tools and organising their work differently. Initially, they may take longer to complete tasks than before, when they acted almost reflexively. This time for changing habits and learning should be built into the plan for implementing the change.

You should also not assume that every employee will complete the training on their own. This can sometimes lead to unfortunate results.

Think about how you want people to use AI and provide them with that information. Think about how you can help them learn new skills. You may have people in your organisation who already have the skills needed to implement, operate and share knowledge with others. If you don’t have these skills in your organisation, and you don’t want the specialists you hire to make the change to spend their valuable time training others, look at the training courses available.

Train at least some of the employees who will be able to pass on their acquired knowledge and skills to the rest.

Well-prepared data

If you are thinking about using generative AI in your business, you need to know that the key to success is data structure. The data on which the language model will work should be structured and stripped of unnecessary information. The better the data, the better the model will perform.

Often, structuring the data generates ideas for using artificial intelligence. For example, the introduction of electronic processing of structured invoices is now prompting many accounting firms to use AI to automatically post invoices. This is possible because the data on which the AI model is trained is structured.


And it’s not just the digital solution you want to implement, but also other technologies – the ones you already have in your organisation. Including those without which the implementation will not achieve the intended results.

Technology vendors often co-fund innovative projects involving the use of AI. Microsoft, for example, will fund both the creation of a prototype and the final implementation, often to the tune of $10-20,000. Such support can be a good solution, but we must bear in mind that it ties us to a particular vendor for the long term.

Risk assessments

First and foremost are data security issues. Artificial intelligence requires access to vast amounts of data, exposing companies to the risk of privacy and data security breaches.

The second major challenge is ethical, especially in the context of task automation, which can lead to significant changes in the labour market, reducing employee engagement or leading to layoffs.

A third, but no less important, attribute is the transparency and explainability of algorithms, which is key to building trust and understanding among users.

Much of this is already described and regulated by the AI ACT Declaration, signed last month, which commits developers of AI-based software to be fully transparent and to design solutions that respect human rights.

7. Implementation plan and change management

In addition to the implementation of the digital solution itself, the ground for change must be properly prepared for success. That foundation is people and processes. Lack of understanding of the reasons for the change, the benefits of the change, the skills to use the change, unwillingness to take the time and attention to learn what is required – these are the most common reasons for low implementation ratings.

Others include acting routinely, trying to use the new technology in the same way as the previous solution, or continuing to use the ‘old’ solution.

To counter this, changes need to be properly communicated to both managers and customers so that their concerns do not outweigh the benefits.

8. Impact measurement

At the beginning, you set the goals you want to achieve with the implementation of AI. Depending on the goals you set (based on systematically collected data), you can assess whether you are getting closer to the goal or whether something is moving you away from the goal. Choose appropriate, i.e. measurable, indicators and use their readings to decide on next steps.

For example, if you’re using AI to improve the hiring process (and you’ve defined exactly what that means for your business), you’ll look at the time it takes to pre-select complex applications, the fit of the application to the needs of the job and the team, and perhaps the time it takes for the new hire to become independent in the tasks at hand. None of these indicators will tell you whether the implementation has resulted in changes for the better if you have not measured these indicators before implementing the change.

Bearing in mind that AI can make wrong decisions (in the absence of sufficient data or by relying on a distorted, biased algorithm), you are in a position to try to verify the algorithm’s decisions. Would you decide to monitor the career progression of a sample of rejected candidates to see if not offering them the job was the best decision?

By measuring metrics on an ongoing basis, rather than retrospectively after the assumed implementation date, you can ensure that you are moving in the right direction and focusing on what is important to your business and your employees.

9. What to keep in mind

1) A failed experiment is not a failure – it is a lesson and a method to avoid losses.

2) Are you missing something? People, skills, technology, data? Don’t give up or assume it will be found somehow. Look within your team – or outside it – for skills and ways to make the change as optimal as possible.

3) Calculate the risks. Don’t pass the buck. Share the responsibility (and success!).

Want to talk about when, where and how to implement AI? Let's talk!

Michał Madura
Senior Business Design Consultant

    Do you wish to receive the latest information related to the topics of business and innovation design, as well as information about Edisonda's activities, projects, and offers?

    Please select the channel through which we can contact you (consent is voluntary):

    Information provided in the form will be used only in order to reach back to you. Contents of the correspondence might be archived. More details could be found in our privacy policy.