History Repeating

"They say the next big thing is here."

Sunday morning, the radio is on, and this line from History Repeating brings back a question that has been on my mind for a while. Who in a company actually needs an expensive AI license? And if we do give it to someone, do they actually use it to create more value for the company, or do they mainly use it to make their own life a little easier?

The longer I think about this discussion, the more familiar it seems. History repeating?

---

Does everyone really need this?

I am old enough to have seen a few of these cycles, although the debate about typewriters arriving in offices is before even my time. With the PC, however, we know that there was a long discussion about where the promised productivity gains had actually gone. The productivity paradox was a common topic at the time, and what it described was fairly simple: computers were suddenly everywhere, but in the productivity statistics they were remarkably hard to find.

At some point, however, that discussion came to an end and nobody asked anymore whether every employee really needed a PC. Laptops, mobile phones, smartphones, a second screen and a whole range of other things followed. Of course, these tools have an economic benefit. Nevertheless, hardly anyone today calculates how many additional minutes of productivity a second monitor has to generate before the purchase pays off. Outside the office it is not much different. A tradesman who drives three screws a day may not need a cordless screwdriver. If he is driving screws all day, the situation looks rather different. And if he can choose between two employers, one of whom gives him proper tools and one of whom does not, that may eventually play a role as well.

We may well reach a similar point with AI. The question when choosing an employer would then no longer be only which laptop or which phone I get, but also which AI I can use, which model and what I am allowed to do with it.

---

It has not always worked out that way

That is the part of the story that speaks in favor of equipping everyone. There is another part, and anyone who has been in this industry long enough will recognize it.

Around 2013, a considerable number of companies equipped entire departments with tablets that were sitting in drawers two years later. Internal social networks were rolled out across whole organizations and then used by a small fraction of the workforce. Company phones for everyone worked well in some organizations and not at all in others.

Interestingly, the productivity paradox did resolve itself later on. The effects arrived, but with a considerable delay, and the usual explanation is that they only materialized once companies had actually changed the way they worked. That explanation is not undisputed, but it fits what most of us have observed in practice. The devices on the desks were clearly not enough on their own.

---

I do not produce five times as much

I can see this quite well in my own work. I did my job before AI and I do it today, and no, I am not suddenly producing five times as much. Nevertheless, quite a lot has changed. For a number of things I need less support than I used to. I can structure information or put it into a table, try out a formulation, have an idea challenged, create an illustration or develop several variants of a slide relatively quickly. Instead of spending a long time on a single version, I can try four or five with manageable effort and then continue with the best one. The result is usually better, and in total it is also more than before, although not by a multiple.

Interestingly, this has not led to me finishing earlier. If anything, my own standards have gone up, because if I can try five variants, I will. At the same time, the time I save on formatting, research and other fairly mechanical tasks is time I can spend on the actual content: thinking, comparing, reflecting and deciding.

There is also an effect we talk about relatively little, which is waiting time. When I used to hand a small intermediate step to someone on the team, ten minutes of actual work did not mean I had the result ten minutes later. Someone had to be available, I had to explain what I needed and there might be questions before the result eventually came back. A small task could easily turn into a few hours or a day of elapsed time. Many of these steps now happen immediately, and that does not only save working time. It changes the speed at which I can continue my own work.If I tried to express all of this as productivity in the classic sense, as output per hour, the number would probably look rather unremarkable. The larger part of the gain sits in quality and in the time I no longer spend waiting for intermediate results, and neither of these appears in that calculation. Perhaps this is not so different from why the computers of the 1980s were so hard to find in the statistics.

Expectations have shifted as well and I expect them to keep shifting. If I ask someone today to consolidate two tables, rework a presentation, translate a document or fix the formatting, hardly anyone gets away with delivering the result hours or days later. That is already the case in my own environment. In an organization where almost nobody has such a tool, none of this is noticeable yet and that is precisely what makes the decision uncomfortable. It is not possible to wait until the pressure arrives, because the pressure only arrives once the decision has been made.

This changes the question, however. It is no longer whether someone with a license creates more value, but whether someone without one can still meet the expectations that now apply. People who do not get the tool are not working any worse than before. They are simply slower than everyone around them.

---

Two calculations that are easily confused

One assumption first. I am taking for granted here that quality holds up and that control works. If unchecked AI output finds its way into work that someone eventually puts their name on, the value quickly becomes negative. That is a separate topic, one I wrote about recently, and it is not an argument against the license.

When we talk about the business case, I think we are actually mixing up two very different calculations. The first one is simple. If I can use a license instead of hiring an intern, or if five people in a department with licenses can handle the work that would otherwise require an additional hire, the case makes itself. The difficulty is that almost nobody can demonstrate this in advance. It usually shows up later, and often not as a decision at all, but as a position that is quietly not refilled after someone leaves. The second case is the one we are actually discussing most of the time. It is not a question of either or, but of whether everyone simply gets something. And for that there is no return per employee that anyone could calculate. We never tried with the second monitor either.

What remains is a ratio. A license costs a few tens of euros a month, while what an employee costs the company in the same month is a multiple of that. We are currently discussing this amount with a degree of diligence that nobody applies to an office chair. Of course, this does not mean that we can simply buy every available license for everyone and then stop caring whether it is used at all. Whether a license costs 20 euros or 100 makes a difference, and someone who practically never needs a tool may well not need it. That was never any different with other software.

I do think, however, that we are making the calculation too narrow. More output is value, and so is better quality. So is needing less support from colleagues. Shorter turnaround times can be enormously valuable in a project. And if better tools mean that people spend less time on work that annoys them or wears them down, I would not put that at zero either.

---

Who actually gets the productivity gain?

Let us assume for a moment that AI actually delivers the productivity gains we expect from it. A task that used to take someone five days can be completed in four at the same quality. What do we do with the fifth day?

The obvious answer is more output. In my own case, however, what I mainly observe is a higher standard rather than shorter days and I do not think that is a coincidence. In work where the demand for quality and volume is practically unlimited, and that includes consulting and large parts of knowledge work, the day you gain goes into higher expectations more or less automatically, unless somebody actively decides otherwise. The shift in expectations I described above is precisely the mechanism behind this.

Whether part of the gain stays with the employee is therefore a question that somebody has to ask, because it does not ask itself. Less time pressure, fewer extra hours, at some point perhaps shorter working hours: all of this is possible, but none of it happens by itself. And the higher expectations reach the employee either way, whether or not they are given the tool. That alone makes it difficult for me to argue for withholding it.

We have, after all, been discussing models like the four day week for years. It would be a rather interesting turn if AI, of all things, eventually delivered some of the productivity that has so far been missing for it. This is an open thought rather than a forecast, but I find it at least as interesting as the question of how many additional PowerPoint slides per hour we can produce with AI.

---

History Repeating?

At MULTIPLAI, we decided after a short pilot that everyone gets the tool. Two things belong in that picture, because they made the decision considerably easier for us than it would be elsewhere. As an AI consultancy, we can hardly support clients in adopting these tools and then argue internally about who has earned a license. And we are growing fast, which means that we were partly looking at the simple calculation described above, tooling instead of an additional hire. As evidence for the harder case, our decision is therefore of limited value. What does transfer, I think, is the approach: we tried it briefly and then stopped asking the question per employee.

My thesis is that the companies which make good tools broadly available early will end up in a better position than those which weigh every single license against an expected return. I cannot prove this today. Then again, neither can anyone who claims the opposite.

It remains a bet and it can go wrong. If the licenses are unused by half the people a year from now, I was wrong. Above all, I was wrong if the way people work does not change, because that is exactly what it came down to with the PC.

With the PC, the mobile phone and the cordless screwdriver, nobody asks about the value per employee anymore. Whether we get to the same point with AI, I do not know. I would be interested in how you see this, and how you made the decision in your own company.