Generative AI was the big topic of 2024, and is likely to remain so for some time to come. However, companies that are starting to deploy it internally are taking too superficial an approach, which is preventing them from realizing its full potential.
Companies have not missed the AI train, and are pulling out all the stops to tame this technology from which they have high expectations. Use cases have been identified by the thousands, and pilots launched by the hundreds of thousands, yet they are still not fully satisfied with the results.
Managers or employees not ready or not confident? Perhaps, as many studies have shown. Governance problems? This is also a major problem, and the absence of governance sometimes helps to achieve results more quickly, at the risk of making big mistakes.
In fact, we don’t talk too much about adoption problems, but rather about problems of value creation.
Mass adoption doesn’t mean value
Just because a technology is massively adopted by users doesn’t mean it creates value, or enough of it. Collaboration tools are the best example of this (Collaborative tools in the workplace: a real waste?): everyone uses them, but so poorly that we lose efficiency while increasing mental charm.
There’s little risk of this with generative AI, where it’s hard to imagine that individual misuse will have an effect on the collective , but it still has to be used well (in a controlled and advanced way) and for the right things (where the value created or costs reduced are greatest).
The pitfall of so-called augmentation tools
Generative AI is presented as an augmentation tool insofar as its promise is to increase the employee’s capabilities rather than replace them.
Augmenting the employee should therefore enable him or her to do things faster and/or better, and it is therefore at the individual level that things will be measured, with time saved being the easiest indicator to measure.
On the face of it, this seems logical, but it misunderstands the reality of work: we rarely work alone, and even when we do, we’re more often than not a link in a chain that makes it possible to achieve a final result.
The fact that one person saves time in the chain is undoubtedly a good thing, but..:
1°) If he’s the only one , there’s still a huge potential gain that hasn’t been realized. Instead of saving 1 hour on the line, we could be saving 2, 3 or 10 times more.
3°) If the rest of the chain doesn’t keep pace with those who are making productivity gains, nothing will be gained in the end, especially if a bottleneck remains at the end of the chain. Let’s take a simple example, even if it’s a bit of a caricature: if a document passes through the hands of several people, and one of them does his or her part of the work faster, but the next person isn’t available at that moment to take over, the gain is zero. Even more of a caricature, if a person or a team completes a deliverable in 3 days instead of a week, but we maintain the rhythm of a weekly final validation session, the final deliverable will still only be validated after a week and, worse, the people concerned may be twiddling their thumbs for two days while waiting for validation and new instructions.
To put it another way, an individual approach is a good start, but the greatest gains will be made with a more collective approach.
The need for a global, end-to-end approach
When it comes to improving the productivity of knowledge workers (because that’s what it’s all about), there are three levels.
1°) the individual
2°) The collective (when individuals work together synchronously or well).
3°) The process: the chain that leads to the finished “product”, whatever its nature, which is the raison d’être of everything else. Nothing that is produced and happens upstream has any value apart from its contribution to this deliverable. If it’s not delivered and validated as meeting expectations faster and/or with higher quality, all the gains made along the chain are useless.
What worries me is:
1°) that we’re measuring the time saved by one person on one task, and not the time saved on an overall process .)
2°) that we don’t prioritize use cases according to the potential savings on a given process , but only according to the type of task at an individual level (AI Success Depends on Tackling “Process Debt”).
This involves :
1°) Start with what needs to be produced
2°) Understand how it is produced
3°) Rethinking the way in which it is done, in the light of what technology allows
4°) Rethink the players involved: do we need so many people, so many new skills, should we outsource all or part of the tasks to partners or technology?
5°) Finally, establish the use case for AI at the individual level, with appropriate training if necessary.
In short, we need to apply AI to work and then to people, and not the other way round.
AI falls victim to a well-known managerial problem
The implementation of AI here falls victim to a well-known problem: nobody in the company cares to know and understand how people work (How to Manage Complexity without Getting Complicated).
We know what a person’s job is, what he or she has to produce, but we never go into the details of how. There are prescribed tasks, of course, but there are also many tasks that are not, because they are the result of poor tools, or of faulty collaboration, or, on the contrary, because they are made necessary by faulty tools or processes, or, quite naturally, by the large proportion of informal activities and tacit processes in knowledge work, where the employee has considerable power over the way in which he organizes his work.
A manager knows a person’s job, knows his or her activities as defined in the job description, but rarely knows all the hidden tasks behind them.
I once indulged in this little game, spending hours next to one of my colleagues. Of course I knew what he did, but I wanted to see how, i.e. everything he had to do to get there. I might as well tell you that there were things that horrified me, and that the person concerned was in no way responsible for them: he was adapting to a faulty context, wasting an inordinate amount of time and seeing his mental load increase unnecessarily.
There’s nothing like taking the time for empirical observation, but today we also have the data to objectify things (What data do we need to understand how people work?), even if the exercise has its limits (The quantified organization: Grail or Big Brother?).
Whether we’re talking about AI or anything else, we can’t significantly improve things without understanding exactly what people do on a day-to-day basis, and limiting ourselves to the very macro notion of activity.
Bottom line
AI is intended to improve people’s work in terms of speed and quality, but also insofar as it frees them from time-consuming tasks.
However, it can only be deployed in a truly productive way if the work in question is well understood: to what deliverable does it contribute, what is the chain of activities and stakeholders involved, and what tasks are performed by each individual.
An approach focused on the individual in no way guarantees gains at process level, or at least not to the same extent as an end-to-end approach.
Image : Teamwork by Jacob Lund via Shutterstock
