Artificial intelligence is changing the professional services market. Who will gain the advantage?

How AI really helps today

Just a few years ago, ChatGPT was better at text descriptions and analysis than at numbers. The scale of today's applications is incomparable – a whole layer of agents, connectors to company systems and specialized engines have grown up around the models. In professional services, AI is good at market analyses, building standard financial models, verifying data consistency in large spreadsheets, automatic completion of contracts, the first layer of research and translations. in a quality that until recently required a sworn translator.

Real effect? Companies that have implemented AI in core processes report performing the same work with a team twice as small – without any noticeable loss of quality. This is not marketing. It's just new work arithmetic.

Where AI still goes wrong – and why

However, it is necessary to distinguish between two things that are commonly lumped into one category of “hallucinations”. The first is the real limitations of the model – a poorly posed question, a lack of context, an attempt to obtain a clear answer where the doctrine itself is not consistent. This isn't an AI hallucination – it's a tool misused.

The second one – much more common in Polish legal and tax realities – is hallucination forced by lack of access to data. The simplest analogy: the difference between a student taking an exam by memory and a student with an open book. The first one, when he doesn't remember, guesses – it sounds convincing, but he confabulates. The second one opens the textbook and quotes.

The publicly available models – when asking about Polish individual interpretations – are the first student. Because EUREKA, the database of interpretations run by the Ministry of Finance, does not provide a public, structured interface from which AI could download data on an ongoing basis. The model has no way to “open a book” – it operates on what it has learned during training, and there access to interpretation is fragmentary. Effect? The model “guesses” the signatures issued by the Director of the National Tax Information, mixes theses and mixes the facts. This sounds very believable – which is the worst thing about this type of error.

On the other hand, they are commercial legal databaseswhich have been licensing and structuring case law for years. Their dedicated AI assistants do not respond from memory – they first search for a specific document in the database and only then formulate a response indicating the source. The hallucination is dramatically limited because the model has something to hold on to.

An expert who uses a general chatbot to research interpretations actually has another problem to solve. An expert who uses a specialized tool with access to the database gains real value. It's not the fault of “artificial intelligence as such.” It's a matter of what tool we use.

Implementations – this is where companies get lost the most

A typical picture today: the company has several licenses for different language models, some AI agent platform, a plug-in for the customer service system, its own chat integrated with documents, and marketing uses a different tool than finance. Each element works separately, but together they create chaos. The second misconception is that AI must be an expensive project. Not at all — sometimes a simpler, well-tailored agent will do more for the company than an expensive package, 80% of which no one will ever use the function. Writing prompts and arguing with chat is a very basic way to use AI today – although it's still better than doing nothing. The real value comes only when integrating with a specific company's systems and data, appointing leaders responsible not only for coordination, but above all for data security – and for ensuring that AI actually changes processes and is not just another icon on the taskbar.

Digital colleagues – no one predicted this in 2023

The thing that really What we didn't expect would happen so quickly is the ability to create your own agents – digital “colleagues” who independently perform multi-step tasks.

The simplest example – handling a cost invoice. Just a few years ago, the invoice was sent by e-mail in PDF format, the accountant manually entered the data into the system, decreed it and sent it for approval. Five to ten minutes per invoice. Today, after the introduction of the mandatory KSeF, invoices are structured from the source – which is a game changer for automation. The agent downloads the invoice from KSeF, verifies the contractor on the VAT white list, compares it with the order in ERP, orders it according to historical rules and sends it to the merchant. The role of man is limited only to exceptions. KSeF, perceived as an onerous chore, is actually becoming a catalyst for true automation.

This model – “the man for exceptions, the agent for the standard” – permeates subsequent processes: handling payments and bank arrangements, preliminary analysis of contracts, first versions of tax calculations, pre-selection of CVs. This is not science fiction. This is the current state, available to companies that want to implement it.

Juniors – what about them?

The answer is uncomfortable but honest: a significant part of junior work today is performed by AI. This affects employment levels not only in professional services, but in the entire white collar sector. Does this mean layoffs for juniors or is their work unnecessary?

My answer: not layoffs, but reassignment – provided the company can retrain. Junior, who understands the process from the substantive side, knows the company from the inside. This makes it a natural bridge between an external AI implementer and a specific company – knowledge of internal processes combined with practical knowledge of AI is a priceless competence today. Why fire someone who was involved in research when you can involve him in the development and supervision of dedicated tools? These are new roles that did not exist five years ago.

A trap the industry hasn't seen yet

If AI does not massively replace white collars in the near future – and everything indicates that it will not fully replace them, because many expert tasks require assessment, context, and responsibility – a seemingly prosaic question arises: where to find future seniors and management?

An expert is not born an expert. An experienced specialist with fifteen years of experience started with painstaking research and hours spent on analyses. These “boring” years built intuition and gave exposure to hundreds of different facts. A junior who never had to manually analyze dozens of cases because an agent did it for him will not develop the same intuition – not because he is worse, but because he didn't have the opportunity.

The pool of seniors will naturally begin to shrink. If the market's response is “we will buy from the competition”, then in 7-10 years it will turn out that there is no one to buy. Companies that today radically cut younger ranks will tomorrow fight for seniors at prices that will eat up savings from layoffs.

It's worth asking the question directly: what is simpler? Slow down, save money, show the shareholder a better quarter? Whether to fight for the employee and retrain himcreate a competence in the company that did not exist a year ago – and explain it to shareholders from Frankfurt or New York who expect a return here and now? The answer seems obvious. And that's why global corporations choose the first path.

And this may be a unique opportunity for Polish companies. Polish business – the one without a foreign shareholder pressing for quarterly results – has a structural advantage in this game: greater freedom to invest in people that global corporations are currently putting on the market. The most talented people, who until recently were impossible to buy, are now looking for a place where their substantive competences and newly acquired efficiency in working with AI will be truly valued. This window opened for maybe three, maybe five years. Whoever uses them will build an advantage for the next decade.

What does this mean?

Three years ago it was possible to observe. Today, observers have already lost – they just don't know it yet. One question left: cWhether your company is on the side of those who are building an advantage for the next decade – or on the side of those who are saving on layoffs and will show shareholders a good quarter. The answer to this question will not depend on the management. It will be decided on Monday morning, when someone decides whether to send the junior's CV to the trash or offer him a new role.

Author: Maciej Gierada, partner at Gekko Taxens