White Mirror - Escaping the efficiency trap
In the "White Mirror" series I talk about positive ideas for using digital technologies in a way that can improve our lives. The title is a reaction to the popular "Black Mirror" series of often dystopic visions of our digital future. While highlighting potential dangers is important, shining a light on the positive potential of digital technologies seems to me at least as important.
AI is everywhere and in more or less subtle ways it is shaping the way we work. But a lot of organisations eagerly pushing for AI use are learning an old lesson the hard way: generating value is hard. More often than not, organisations step into an efficiency trap with AI, not only risking money but also the motivation of their staff. But there is another way.
The issue
AI companies and managers in other industries are thrilled by the promised productivity gains of generative AI services: Marc Benioff of Salesforce claims that AI could be doing upwards of 30% of the work, an IBM study among executives believes in productivity gains through AI of up to 42% by 2030, and JPMorgan reports about realized productivity gains with more to come. The excitement is palpable.
And, at first glance, that excitement is understandable: generating text at the touch of a button, rewriting codebases from one framework to the next, building whole applications or creating research reports in a fraction of the time, those are all things that generative AI can do. However, AI companies obviously have a self-interest in claiming massive productivity gains to keep their investors happy and many executives probably buy into these promises out of fear of missing the next big thing or being seen as outdated. Yet, there are objective and subjective objections that should give executives pause.
First, a number of studies finds a gap between perceived and actual productivity gains. From the famous MIT NANDA piece on how 95% of companies fail to generate value with AI, to outlandish costs of AI utilization with little to show for it: Company spending on AI is projected to be upwards of USD 650 billion (yes, with a b) and a single company blew through tokens worth USD 500 million in one month - a useful, if costly reminder of the importance of AI governance, on the supplier side and the customer side.
Second, subjectively we understand that being able to move faster doesn't necessarily get you to your goal faster. We don't have to go back all the way to ancient fables - are you more of a hare person or a tortoise person? - to feel skeptical about outlandish productivity gains promised by AI. Yes, I can generate an e-mail in the right tone for the recipient in a language I don't understand at the touch of a button, but did this really make me more productive? And are we generating value just because I can now blurt out reports that would have taken weeks in minutes? The emergence of "AI slop" certainly suggests otherwise.
The solution
Executives looking to really generate value with AI need to move away from telling their organization "use AI for the sake of using AI". They tend to ask the wrong question. Instead of asking "how can you do more of what you are doing with AI" they should be asking "how can you generate value with AI"? Then the interesting discussion starts!
Just as with digital transformation, AI is not about doing the same thing but faster. New technologies might lower the cost of doing a certain task but that doesn't tell you anything about the value of that task. While your employees are out there burning tokens for things that they already did before AI you are missing an opportunity to really transform your organization, starting with transforming your processes and outputs.
What is already real is that AI has shifted expectations on the client and management side. You previously used three days for report X, I know that AI can help you do it, so I now expect you to do it in one day (and I also only want to pay for one day of work). But this simplified view doesn't ask the value question. Sure, I can give you a document in a day, but will it really help you?
But if we transform processes to actually free up time to properly think about the content of a report, to stick with the example, we can suddenly see how AI can help to generate value because the discussion is no longer about quantity but instead about quality.
An analyst used to spend three days on a client report: one day pulling data and sources, one day drafting, one day formatting and checking. Management hears AI can help, so the new expectation is: do it in one day, bill one day. AI drafts the prose, formats the tables, tidies the citations. The output is the same report — same structure, same depth — just produced faster and cheaper. Everyone's happy until the client realizes the one-day report tells them what happened but not what to do about it. The analyst was never the bottleneck on typing; they were the bottleneck on thinking, and that didn't change. You've made the cheap part cheaper and called it transformation.
Same three days, but the process is redesigned around where the human actually adds value. AI now owns the mechanical layer end to end: it gathers and reconciles the data, flags anomalies, produces a structured first draft, and — this is the part that changes the output, not just the speed — runs the analysis three different ways, stress-tests the assumptions, and surfaces the two findings that contradict last quarter's narrative. The analyst no longer spends two of three days assembling; they spend those two days interrogating. They sit with the contradictions, pressure-test the recommendation with a colleague, and turn the report from a description into a decision. The client doesn't get the same report a day sooner. They get a report that answers a question they couldn't previously afford to ask.
Not until we focus on the value discussion and move beyond simply delegating our current working steps to generative AI, will we be able to escape the efficiency trap and leave the hamster wheel to actually move forward.
Subscribe to this newsletter and follow me on social media for more publications and thoughts:
LinkedIn
Mastodon