For more than five years, I led technology and process transformation at AT&T. I was involved in dozens of initiatives meant to connect systems, simplify processes and remove manual work. In a large corporation, this was extremely difficult. Even when everyone agreed that a process should be improved, the change required coordination across departments, technologies, budgets and approval bodies.
We had meaningful successes. A strong initiative would shorten delivery by ten percent, remove a tenth of the manual work, reduce the number of errors and escalations. Those improvements created value and should not be underestimated. Yet looking back, I am not sure I would describe many of them as truly transformational. We were improving individual steps of an existing process. Its basic structure remained unchanged: the same work passed through the same teams, waited in the same queues, and the same context had to be explained again and again.
That is exactly why I believe today's opportunity is different. Bringing artificial intelligence into a company does not have to mean only that people perform the same tasks faster. For the first time, we have a technology that can also change how work flows through the company.
Where companies actually lose time
Most thinking about AI starts with a reasonable question: how can it help a person complete a task faster? A salesperson prepares a proposal in half the time. An analyst summarizes a report in minutes. An accountant processes documents without retyping them. These are real gains, and often the easiest place to begin.
There is a catch, however. If we simply insert AI into an existing process, we speed up individual activities, yet the total time to an outcome may barely change. We accelerate the work and leave the waiting untouched. And in most companies, waiting is the bigger problem.
Picture an ordinary customer request. Sales reviews it. It is passed to operations. Operations asks finance for a check. Finance waits for supporting documents. The approved request goes to delivery, quality reviews the output, and the result finally returns to the customer. The actual work at each step takes a few minutes. The total elapsed time is measured in days.
Most of that time, the request is not waiting for work. It waits in an inbox. It waits for someone to have capacity. It waits for a missing piece of data. It waits for an approval. It waits because one system does not know what another system knows. It waits because the person who took it over lacks the context of the person before them.
The difference between working time and waiting time is central to bringing AI into a company. A task that requires four minutes of focused work travels through the organization for four days. Improving those four minutes is useful. Removing those four days is transformational.
Making work faster is useful. Removing the waiting is transformational.
What bringing AI into a company really means
Bringing AI into a company does not mean buying a license or launching a company chatbot. It means changing at least one working process so that a measurable result moves. In my experience, the value of AI in a company appears on three levels.
The first level is faster individual work. AI helps write a text, summarize a document, prepare an analysis. It is the easiest place to start, and for many companies it delivers the first measurable savings.
The second level is coordination of a whole process. This is where the so-called AI agent enters the picture. There is nothing mysterious about it: an AI agent is a program that can carry out a series of steps on its own while holding on to the context. It reads the customer's history in the CRM, adds data from the ERP, checks a rule in an internal policy, prepares a response and moves the case forward. A person no longer has to carry information from one window to another; the agent holds the steps of the process together instead.
The third and most valuable level is a new arrangement of the process itself. Once information stops being locked in separate systems, many steps lose their reason to exist. Which approvals were created only because data used to be hard to access? Which handovers between teams exist only because each team sees into a different system? Which queues can disappear entirely? True transformation does not come from accelerating the existing process. It comes from redrawing it.
I am not claiming that every process should run on its own. Judgment, relationships, exceptions and accountability remain with people. But the range of what can be rearranged has expanded dramatically over the past two years.
The biggest value does not appear where AI writes faster. It appears where the waiting between people and systems disappears.
Where to start: one problem with a clear before and after
The most common mistake I see is starting with a strategy. A working group is formed, a document gets written, a budget is sought, and sometimes the company waits for a grant call. A year later it has a presentation and a pilot in a test environment, yet no process has changed.
It is more useful to start with one operational problem that has a clear before and after. Two places are worth the first look.
The first is the work your team does by hand. Manual work is rarely the real problem. It tends to be a symptom of a stopgap that hardened into a process because nobody had time to fix it properly. It is also the cleanest first assignment, because the difference is immediately visible: a day of senior time each week currently spent doing work meant for a machine. This is why manual work is usually the first thing to go when AI arrives.
The second place is the data you already have and nobody looks at. When a retail network lined up six months of its own stock transfer requests, it turned out that more than half of the items the stores were asking for could not be shipped from the warehouse. The data had been there the whole time. What was missing was a view that put it together.
You can tell a good assignment by a single sentence. The sentence "we want to use AI in customer service" says nothing about what will change. The sentence "we want to shorten the first response to a customer from two days to two hours" is an assignment: it has a before, a target after, and a team that will feel the difference.
How to choose an AI solution for your company
When choosing a solution, it does not pay to decide by the brand of the model. Models today are essentially a commodity: available, cheap and interchangeable. Four more practical questions decide.
First: does the solution change your process, or is it just another tool? A chatbot license is a tool. By itself it will not change how work passes through the company.
Second: where does your data stay? For many Slovak and European companies this is the deciding question. When we replaced a mid-sized IT integrator's timesheet with a voice note, all processing ran on the client's servers and no data left the company. That was not a technical detail but a condition without which the solution could not have been deployed. The result: 98 percent of the team used the system daily, and 1.95 million euros of billable work surfaced that had never been recorded anywhere.
Third: can you trace back why the system decided the way it did? People trust a system when they can see its decisions and correct them. When we built a case-handling overview for a European metropolis, every allocation decision had to be traceable, otherwise the city could not have used it.
Fourth: what remains when the supplier leaves? A solution that lives only on the supplier's side disappears the day the cooperation ends. What should remain is code, documentation and a person on your team who can run the system. Dependence on a supplier is not a technical question but a business one.
You can tell a good solution by what remains when the supplier leaves.
What bringing AI into a company actually costs
The exact amount depends on the problem you are solving, and I will not pretend a price list exists. It is more useful to understand how the economics of this work have changed.
For twenty years, a serious transformation engagement had a predictable shape: a six-figure budget before the first line of code, months of planning and a horizon measured in quarters. With numbers like that, waiting was rational. That math has changed. A small experienced team that uses AI from day one now delivers in weeks what once took a year. The cost of moving has dropped by an order of magnitude. The cost of waiting has not: a day of senior time each week spent on a manual report, margin quietly leaking from unrecorded work, a customer standing in front of an empty shelf. These items never appear in any statement, yet you pay them every month.
That is why I consider one sprint the smallest meaningful unit: a project of a few weeks with a fixed scope and a fixed price. If it delivers, you have a result you can build on. If it does not, you have spent a sum you knew in advance, and you know more about your company than any workshop would have told you. The loss is bounded, and you decide about continuing only once you have seen the work. A grant can add to the result. It should not be the condition for starting at all.
I go into the economics of such a project in more detail, from what makes up the price through computing the return to reading a vendor's offer, in a separate text on what it costs to bring AI into a company.
The questions to ask before you start
For years, transformation programs asked: how do we automate this step, how do we connect these systems, how do we reduce the manual work. These are still good questions. AI, however, makes it possible to ask more fundamental ones.
Why does this handover of work exist at all? Does another team genuinely need to take ownership, or did the previous person simply lack access to the information?
Does this approval protect the company from a real risk, or was it introduced years ago as insurance against bad data?
Does this task require human judgment, or mostly collecting data from three systems and applying a written rule?
Who would notice if this waiting disappeared tomorrow?
Many process steps are not a natural part of the work. They are remnants of technological limitations that no longer apply.
Work will get faster. The question is whether it will move
Today I see opportunities for genuine process redesign almost everywhere I look. That does not mean transformation has suddenly become easy. Companies still need clear goals, reliable data, a process owner and the trust of their people. But the range of what is possible has expanded dramatically. The biggest competitive advantage of the coming years will probably not come from faster employees, but from fewer handovers, shorter queues and a better flow of work from request to outcome.
So the question for your company is not whether to bring AI in. It is: will you use it so that people do the old tasks faster, or to redraw how value flows to you?
If you want to find out where your first sprint would make sense, sketch it out in writing in our diagnostic. A partner reads your draft, marks it up and sends it back. Only then do you decide whether it is worth a call.