Where to Start AI Transformation: 5 Situations That Hold the First Project Back

The question of where to start AI transformation is one of the most important ones — and one that has no universal answer. The entry point matters: the first project sets the pace, shapes the team's attitude toward AI, and largely determines whether it makes sense to move forward at all.

At the same time, the decision about the first project is often made under conditions of uncertainty. The company sees that competitors are launching something. Leadership returns from a conference with a sense that action is needed. The IT department has already tested several tools and is ready to propose options. These are all understandable and normal triggers — the question is how not to let them become the only criterion for the choice.

Below are five situations we see most often, and practical guidance for those who want to take them into account when choosing their first AI project.

One of the most common scenarios: the IT department or an external consultant proposes launching a chatbot for internal employee support, or a dashboard with AI-powered analytics. The tool works technically — and the project gets launched.

The problem is that neither an HR chatbot nor a data visualisation on its own affects what the company ultimately exists for: revenue, operational efficiency, speed of decision-making. If an employee used to spend 10 minutes searching for an answer in the corporate knowledge base and now spends 2 minutes — that is convenient, but it is unlikely to change the business.

It is a different story if the same chatbot is embedded in the customer support process and reduces response time from 48 hours to 4 — that already affects client retention and team workload. The difference is not in the technology, but in which business problem it is connected to.

When a company decides to implement AI, there is often a desire to solve the biggest problem right away. Automate the end-to-end supply chain. Rebuild the demand forecasting system. Roll out AI assistants to the entire sales team at once.

These are ambitious goals — and they are typically the ones that fail as a first step. The reason is straightforward: complex processes have too many variables to control at once. Data from multiple systems, different teams with different requirements, long approval cycles. It can take six months to see the first visible result — and by that point, both the budget and leadership's attention have often shifted elsewhere.

The first AI project should be narrow enough to launch in 6–8 weeks, and significant enough that the result is clear to the whole team. A small but real result builds trust inside the company and creates the foundation for the next step — with a larger budget and less resistance.

AI runs on data. That sounds obvious — but most companies do not know the real state of their data until they start their first project. And this is precisely where many projects lose months.

A typical situation: data is stored in three different systems that are not connected to each other. One holds sales data, another logistics, a third finance. The formats differ, some data is entered manually and contains errors, and historical records for the relevant period either do not exist or are unreliable. Before AI can do anything useful with this data, it needs to be gathered, cleaned, and brought into a consistent format.

A good practical step before choosing the first AI project is a quick data audit: where the data lives, what condition it is in, who manages it, and how easily it can be accessed. This takes one to two weeks and can save several months of work.

The technology is in place, the system works — but employees continue using Excel or the old tool. The project may have been executed well technically, but the result is zero change in how work actually gets done.

This happens almost always for the same reason: the people who are supposed to work with AI every day were not part of the implementation process. They were informed about the new tool after the fact, given one training session, and expected to adapt. If someone does not understand why a new process is needed and does not feel their input was considered — they will work the way they always have.

The first AI project is both a technical and an organisational experiment. A team that is involved from the start — helping to define the problem, testing the prototype, giving feedback — ends up becoming the project's strongest advocate inside the company.

'Improve efficiency' is not a goal — it is a direction. Without a specific, measurable outcome, it is impossible to tell whether the project worked. And it is impossible to justify the next step — whether to the board or to the CFO.

In practice, it tends to look like this: a project launches with the stated goal of 'automating request processing', but without any understanding of how long that currently takes or what exactly will count as success in three months. The project gets delivered technically — and nobody can answer whether things actually got better, or by how much.

Before launching, it is worth locking in three things: exactly which process is changing, how it works today in concrete numbers — time, cost, error rate — and what result within 60-90 days will mean the project succeeded. It is also worth keeping in mind that ROI from AI projects is calculated differently than from standard IT implementations: direct savings on headcount are often not the main indicator. For a more detailed look at how to build that assessment, see our article on measuring AI ROI.

A good first AI project meets several criteria at once. Here is what we check together with clients when making the choice:

These are the criteria we use to find the right entry point together with clients — and this is how the first project becomes the start of a transformation, rather than another experiment.

Working out exactly where your company should start is one of the most common questions we help with.

We offer a free 60-minute consultation: we look at your specific processes, identify where AI is likely to deliver results fastest, and map out the first steps.

The methodology for choosing and launching first AI projects is part of a broader system I describe in the book “The AI Transformation”. If you want to go beyond the entry point and understand how to build AI transformation step by step inside your company, the book gives you the full structure.

The book is currently available for early pre-order access.