There is a pattern I see in nearly every business I work with. It starts with excitement. Someone in the leadership team reads about a tool, the company buys a handful of subscriptions, a few people try them for a fortnight, and the energy then drains away. 6 months later the tools are still being paid for, barely anyone is using them, and the word people reach for is “fatigue.”
I run a B2B marketing agency in Manchester where we deliver content, strategy and campaigns across many client accounts, in sectors ranging from food safety to children’s footwear to recruitment software. The honest reason I can carry that volume is AI.
So when I hear that a team is tired of AI, my first question is never about the tools. It is about how they decided whether the tools were working. Most AI fatigue is not a technology problem. It is a measurement problem, and in my opinion it is very fixable.
3 things wearing people down
When you look closely, AI fatigue is usually 3 separate problems stacked on top of each other.
- The first is tool sprawl. A business ends up with 8 overlapping subscriptions because every department bought its own, and nobody went deep enough in any of them to get past the novelty.
- The second is expectation mismatch. People were sold autonomy and just handed an assistant. They expected something that would finish the work and got something that needs managing, checking and correcting.
- The third, and the one I feel most strongly about, is sameness. AI has dropped the cost of producing acceptable, average output to almost nothing, so the internet fills up with content that all reads the same way. Anyone can spot it now. The same 3 sentence rhythm, tidy conclusions, and a handful of words that give the game away.
Each of these happen, and each one quietly convinces people the technology was overhyped. None of them are a flaw in the technology, but they are flaws in how it was bought, framed and measured.
You are probably comparing it to the wrong thing
The biggest reason businesses fail to see a return on AI in my opinion is that they measure it against replacing a person. They ask whether the tool can do a whole job, decide it cannot… and ultimately conclude it has failed.
That is the complete wrong baseline. In my experience, AI raises your floor long before it raises your ceiling. It doesn’t make my best work dramatically better. What it does is make sure nothing leaves my desk at low quality, and it gets me to a usable first draft in minutes rather than hours. The value is not a finished article appearing from nowhere. The value is that the worst version of any piece of work is now far better, and the time between blank page and working draft has collapsed.
For a marketing agency like ourselves that matters enormously. A first draft of a campaign plan, a set of social captions, a client report or a piece of keyword analysis used to eat the day up. Now it takes the first few hours, and the rest of my time goes on the part clients actually pay for, which is judgement, strategy, actually tangibly understanding where the return is coming from and how to optimise that, and the bits that sound like a person wrote them.
If you measure AI by whether it replaced someone, it will always disappoint you, but if you measure it by how much low value work it removed, and where the reclaimed time went, the picture changes completely.
A simpler way to prove return
I keep my own measurement deliberately plain, because anything too clever tends to get abandoned within a month I’ve found.
First, I look at net time saved, not gross. The number that matters is not how long the tool took to produce a draft. It is how long the draft took minus the time I then spent editing, fact checking and rewriting it into something I would put my name to. A tool that writes a draft in 2 minutes but needs an hour of correction has not saved me an hour. Plenty of AI claims fall apart the moment you subtract the editing time, and that subtraction is the most crucial figure you can track.
Second, I watch time to first draft. For any repeatable task, how long does it take to get from nothing to a working version. If that number is falling, the tool is earning its place. If it is not, the tool is just purely there for decoration (something I see TONS of Founders fall victim to).
Third, I set a quality floor and check that the tool never drops below it. The floor is the standard beneath which nothing is allowed to reach a client. AI is very good at protecting a floor, because it never has an off day, never loses patience on the 10th version of the same caption, and never forgets a step in a checklist. Holding a consistent floor across a high volume of work is where I get my most reliable return, far more than any single impressive output.
Before adopting anything new, I ask 3 questions.
- What specific task is this for?
- What does that task currently cost me in time?
- What will it still cost once I include the checking?
If I cannot answer all 3, I am buying a subscription on the hope it will do more than it potentially can, which is how the sprawl starts in the first place.
The editing layer is where the return is protected
Here is the part I find tends to get skipped, and it is the part that decides whether you see a return at all in my head.
Marketing is differentiation. If every business in a sector uses the same tools in the same way, the output converges, and the thing that was meant to give you an edge quietly removes it. I spend a fair portion of my week stripping the tells of machine writing out of drafts, the words and rhythms that make a reader’s eyes glaze over, the neat phrases that say nothing. And I do it because the moment a client’s content sounds like everyone else’s, the return on that content completely disappears, however fast it was produced.
So the human layer is not a nice extra you add if there is time. I cannot stress this enough! It’s where the value is made and protected. The model gets me to a draft. The judgement, the voice, the specific example from a client meeting, the opinion that only comes from having done the work, that is what makes the output worth reading.
AI handles the floor. People handle the ceiling. The businesses that try to remove the people altogether are the ones that end up fatigued, because they are publishing the average and wondering why nothing lands.
Narrowing the stack
If a team near you is tired of AI, the fix is rarely a new tool. It is usually fewer tools, used more seriously.
I would rather a team learned 1 or 2 tools properly than dabbled across 10. Depth is where the compounding returns sit, because most of the value in these systems is in knowing how to ask, how to set them up, and how to build them into a process you repeat. The first week with any tool is the worst it will ever be for you. The teams that give up in that first week never reach the part where it starts to pay off.
Pick the small number of tasks that eat the most time. Choose one tool for each. Set the quality floor. Keep a person on the judgement. Measure net time saved and watch where it goes. That is the whole method, and I appreciate it is far less exciting than the marketing around AI would have you believe, but I feel actually that is rather the point.
The businesses getting a return on AI right now are not the ones with the most tools, or the loudest opinions about where the technology is heading. They are the ones who decided what they were measuring before they started, kept people on the parts that need taste, and treated these tools as a way to raise the floor across a large volume of work rather than a way to remove the people doing it.
AI fatigue is what happens when expectation outruns measurement. Fix the measurement first, protect the human layer involved, and the fatigue tends to lift on its own.
