A Faster Horse: Your Team Is Faster with AI. Is the Work?

I run my advisory and training practice on my own. I attend the meetings, research and prepare the content the need calls for, produce the presentation and the executive summary, and deliver the work. I used to do each of these steps one by one. Today I use two, sometimes three, different AI models and prepare the work far faster. I also have them check for gaps, so what I send is more thorough and has been reviewed from several angles.

But the time organizations take to respond has not shortened in proportion. Inside the organization, the content still moves step by step: from email to email, meeting to meeting, approval to approval.

The problem is not the people on the other side. In my own work I can handle research, preparation and review together, on my own. In an organization, the same work has to be coordinated across different teams and decision-makers. I spent years inside those organizations and ran the same cycle myself.

What this experience tells me is this: AI makes the person doing the work faster. But if the decision-making structure stays as it was, individual speed does not carry through to the time it takes to finish the work.

The Fear of Missing the Train

Many senior executives are asking themselves the same question today: are we falling behind our competitors in using AI?

The concern is neither baseless nor rare. In IBM's 2025 study of 2,000 CEOs, 64% acknowledged that the risk of falling behind drives them to invest in some technologies before they have a clear understanding of the value they bring.1

Anxiety looks for fast, visible action. In my observation, one of the first steps is to send teams to prompt training. That is a valuable start. A well-designed course helps people describe what they need clearly, give the necessary context, and question and verify the output. These skills are the foundation of working well with AI.

But the question anxiety asks is not the real question. One asks whether we are using AI; the other asks whether that use turns into results. Making that happen is a management task, to be handled alongside the training. In which processes will the new skills be used? How will the time saved be used? How will authority and success measures be adjusted? Next to the training plan, these questions need an owner too.

The training calendar has a slot for employees. Which calendar holds what management will change?

We have seen this before. CRM, ERP, robotic process automation: in every wave, some organizations jumped on the bandwagon for fear of missing the train. Those who adopted the technology as a fashion were often disappointed. Those who rethought how work gets done alongside it made lasting gains. AI is no exception.

A Faster Horse

There is a line attributed to Henry Ford: If I had asked my customers what they wanted, they would have said a faster horse. There is no reliable evidence that Ford ever said it.2 But the line keeps circulating because it captures something true: when we meet a new possibility, we first use it to speed up the work we already know.

History offers an example. When electricity reached factories, many firms replaced the steam engine with an electric motor but left the layout of the factory as it was. According to the economic historian Paul David, the big leap in productivity took decades, and reorganizing factories around the new technology played an important part in it.3

Today AI resembles that electric motor; prompt training is learning to use the motor. Better motor, same factory. Developing an employee's individual skills does not, by itself, change how authority is distributed, how success is measured or how the organization works.

I Have Seen This Film Before

I lived through the same pattern before AI, but at a different layer.

At a payments company where I worked, we knew which customers were doing business with us and which were not. But we had a problem accessing the data, and the definitions of active, passive and dormant customers were not clear. Management meetings were spent less on what a report said than on which number to go by. For the same customer base, the figures from finance, marketing and operations did not match. However quickly we prepared the reports, the meeting stalled at the same point.

We had to take the raw data and run the analysis ourselves. First we clarified the definitions: who counts as an active customer, who has become passive, who has gone dormant? The analysis showed that an open account did not mean a working relationship. There were customers who had applied for a POS terminal and had an account opened, but had not put through a single transaction because we could not agree on price.

That distinction mattered. We could not approach a customer who had transacted and later stopped in the same way as one who had yet to make a first transaction. With one, we needed to understand why they had stopped; with the other, why they had never started.

After we presented the analysis to management, decisions and actions followed. The sales team's priority became activating existing customers, and current pricing information was shared again. In marketing, too, priority went to existing customers, especially those who were not transacting.

This experience is the other side of the observation I opened with. There, preparation had sped up but the approval cycle had stayed the same. Here, the reports were fast, but the data was not usable and we could not agree on what a number meant. The common lesson: preparing an output faster does not guarantee that the decision will also be made faster or better.

I read both experiences in three layers:

By decision architecture I mean designing these three layers together. The aim is not to speed up every decision. It is to make decisions at the right speed, with sufficient evidence and with clear ownership.

AI can help at each of these layers. It can speed up analysis, surface inconsistencies, and propose different definitions of an “active customer” and compare their consequences. But which definition fits the purpose of the work, where it will be used and who is accountable for the result are for the organization to decide. AI generates proposals; the organization adopts the binding measure.

The Research Points the Same Way

McKinsey's 2026 survey on AI looks separately at the companies that get the most value from AI, which it defines as “high performers”. About three-quarters of these companies are redesigning workflows for AI; among other companies, the share is around one-quarter.4

On its own, this finding does not establish cause and effect, nor does it rule out the importance of tools. But it shows that redesign is one of the features that set value creators apart.

Where Does the Time Saved Go?

There is one more question: where does the time AI saves actually go?

In BCG's research published in June 2026, 66% of frontline employees who use AI regularly say they receive limited or no guidance on what to do with the time they save.5

How we use the time saved is part of designing the work. Will it go to better customer preparation, to learning, to clearing the backlog? The manager and the team should settle these priorities together.

The measure matters as much as the priority. We are teaching new skills; will we keep judging people by the old success measures? Measuring output volume alone can put quantity ahead of quality and risks rewarding a copy-paste-send habit. It can also use up the time that could go to thinking and verification.

Monday Morning

When you plan the training, address how the new skills will be put to use as well. The question to ask: what work will we do differently with these new skills?

You do not need a large transformation programme to start:

1. Choose one workflow. Make it a process whose result touches customers or profitability and that can be tracked end to end.

2. Find the knot. Where does the work wait? Who reaches the data needed for the decision, and how? Which definition is not the same for everyone?

3. Decide what will change. Which decision will sit with whom, which data and definitions will be the reference, and when will an additional check be needed?

4. Complement training with practice. Build a common language through foundational training; then work on real tasks in the workflow you have chosen.

5. Track three things together. Waiting time, error rate and business result. If you measure speed alone, you will reward fast but wrong decisions.

6. Make what works visible and spread it. Recognize successful practices and the people behind them, and take them across the organization.

The real lesson of the line attributed to Ford is not about customers; it is about us. Customers ask for a faster horse because they look out from the world they know. We managers often look at AI from the same place.

Training expands what people can do. Management's job is to open up the space where those new skills will be used.

I see advisory work and training as two pieces of work that complete each other. In one, we develop people's skills. In the other, we address the definitions, authority and ways of working that help those skills turn into business results. A single process is enough to start: while developing the skills the team needs in that process, we can look together at where the work waits and which decisions need to change.

Once your team's AI training is over, what will change in the way you manage?