Somewhere in your organisation, your AI initiative has already filed its most accurate status report. It is not in the dashboard. It is in the workload.
Expected AI to raise their company’s productivity.
Said it had added to their workload instead.
Same tools. Same companies. The leadership expected lift. The workforce reported weight.
The standard reading of that gap is patient and flattering: adoption takes time, training will close it, the workforce will catch up. Give it a quarter. Here is my reading, after thirty years of watching organisations meet new technology: the 77% is not an adoption lag. It is a gauge. And the gauge is reporting a condition that existed before the technology arrived.
When the fuel gauge reads empty, a sane pilot does not argue with the instrument. He does not tap the glass and suggest the gauge needs more training. He accepts that it is reporting something real — and that the condition was there before he looked down.
The condition is the gap I have spent the past month measuring for you. Sixty-seven percent of supervisors in our client data had never made their expectations clear at all. Seventy percent of what leaders expect never becomes visible to the people expected to deliver it. And last week: the account most leaders draw commitment from was never funded in the first place.
Deploy AI into that gap and watch what the gauge does.
The AI system you just deployed arrives aimed at the visible fraction of the work — the part leadership can see from where it sits. But nobody ever established what the job actually involves, because nobody ever asked. So the invisible majority of the work keeps happening the way it always has, and the new work arrives on top: feeding the AI, checking its output, and repairing what it gets wrong. That extra workload is not a malfunction. It is an accurate reading of the gap AI was deployed into. The instrument is functioning perfectly.
When your people tell you the new system is slowing them down, they are not being obstructive. They are reading the gauge out loud. The 77% are not the friction in your AI rollout. They are its measurement system — the only honest one you have.
And the research community keeps confirming the reading. RAND and Gartner both put AI project failure in the 70 to 85 percent range, and every serious postmortem prescribes a version of the same remedy: align expectations before you deploy — starting with the most basic alignment there is, understanding what the work actually involves — and build the trust that makes honest reporting possible. That is not a technological methodology. Every item on that list is leadership. The researchers just stop short of using the word.
Before the vendor demo, before the license count, before the rollout plan — did anyone sit with the people who do the work and establish what the work actually is?
Not the org-chart version. The real version, with the workarounds and the undocumented steps and the twenty-minute fix that has kept the process alive for years. If the answer is no, then the system was aimed by leaders who never saw the whole job — so it hit the part they could see and missed the part that matters. Seventy-seven percent is exactly the number you should expect.
The fix is not another training module. Training teaches people to operate the system. It cannot make the system aimed at the right work. The fix is slower and cheaper and harder: go find out what the job is. Ask. Watch. Let the people who do the work tell you where the system belongs — and where it is a tourist. Then deploy, into work you can finally see.
The gauge will keep reading what it reads.
What happens next depends entirely on you — whether you treat that reading as information, as a call to action, or as insubordination.
test