If you've been wondering why your AI investment isn't showing up in business performance, a new Glean report puts numbers on the problem for the first time. Six thousand full-time knowledge workers, co-authored with researchers at Stanford, UC Berkeley, and Harvard.
The finding: organizations seeing real business results from AI don't outperform everyone else on technology selection. They out-manage them.
That's the distinction most leadership teams are missing. And understanding it is the difference between the 13% who are moving the needle and the 87% who feel productive but can't point to company-level results. Here's what the data says, what it means for your team, and what to do about it.
Most executives are measuring AI by usage: who's using it, how often, how many prompts per week. That's the wrong metric.
AI didn't give your team a tool. It gave them a direct report.
A year ago, most of your team did the work themselves. Today they hand it to something else, wait, and decide if what comes back is good enough to ship. That's managing. And managing well requires skills most people have never been taught.
The root cause: your team got promoted overnight. No announcement, no training, no playbook for how to delegate to something that's fast, tireless, and occasionally confidently wrong. So most of them are winging it.
The result is what the report calls botsitting: 6.4 hours a week lost to re-running prompts, correcting outputs, and providing context the AI should have had from the start. That's not a model problem. That's what happens when you skip the briefing.
Here's the data point that reframes the whole conversation.
The top-performing organizations in the report don't win on model selection or tool spend. The same models are available to everyone. What separates them is management infrastructure:
The "aha" moment: the companies in the 13% treated AI adoption for what it actually is — a management challenge, not a software rollout. Everyone else ran a software rollout.
And one more number that stings: 53% of workers say the information their AI actually needs to do good work isn't accessible to it. That's not the model being limited. That's poor setup. Context-rich teams cut quality failures nearly in half compared to context-poor ones. The botsitting tax is almost entirely self-inflicted.
Here's the approach that separates the 13% from everyone else.
Tier 1: Build Context Once, Not Every Session
The Concept: The biggest driver of botsitting is workers re-explaining the same background to AI at the start of every task. That's dead weight, and it adds up to hours a week.
The Application: Build durable context, briefs, examples, quality standards, company background, that your AI tools can draw on automatically. The goal is that "what good looks like" exists before the work starts. Tools like Cowork are built for exactly this. One setup removes the re-briefing overhead permanently.
Tier 2: Teach the Brief, Not the Prompt
The Concept: Prompting gets all the attention. Briefing is the actual skill. The quality of what comes out of AI is almost entirely set by the clarity of what goes in.
The Application: A good brief is learnable, and it looks the same whether your direct report is a person or an AI: clear objective, relevant context, an example of a strong output, and a quality bar to evaluate against. Every experienced manager knows that under-briefing a human means spending the week rewriting their work. The same equation runs on AI.
Tier 3: Make Reviewing Part of the Job
The Concept: The report found that 69% of workers admit to "botshitting" — shipping AI output they haven't verified or can't fully explain. That's not a character flaw. It's a missing checkpoint.
The Application: Build review into the workflow as an expected step, not a suggestion. The companies winning aren't those with the heaviest AI usage. They're the ones where people review output before it ships, catch the errors, and hold the standard. Reviewing is the job now.
Measuring AI by usage instead of outcomes. Prompts per week is a vanity metric. The person generating 200 prompts and shipping unreviewed output is not your AI leader. The person generating 40 and getting clean, trustworthy work every time is. Most organizations are rewarding the wrong behavior.
The fix: Change what you're measuring. Track quality and rework rates. Recognize the people who produce the least corrected work, not the highest volume.
Blaming the model when the real issue is the brief. When AI output is consistently bad, most teams assume the tool isn't good enough and either switch products or give up. Rarely is the model the problem. The report is clear that the only real variable between strong and weak results is whether someone set the work up properly.
The fix: Before changing tools, audit the brief. Add context. Add an example. Set the bar before the work starts. Then evaluate.
Skipping training because "people will figure it out." This is the most expensive assumption in the building. The gap between 52% and 90% on training and support doesn't happen by accident. It's a deliberate choice the winning companies made. The companies that skipped it are the ones staring at a 6.4-hour botsitting tab every week.
The fix: Treat AI skills the same way you'd treat any other role-critical capability. Formal training, clear standards, and recognition for the people who do it well. That investment is exactly what separates the 13% from everyone else.
Before this report, it was easy to read slow AI ROI as a tool problem. After, it's clear it's a management problem. And that's actually good news.
Better models take years to develop. Better managers can be built this quarter.
The 13% who are moving the needle aren't waiting on the next model release. They're building the management infrastructure that lets their teams actually get clean work out the other side. That's the opportunity in front of every leadership team right now. It's also exactly the gap Biggest Goal exists to close: not teaching people to prompt, but teaching them to manage.
If you want to stay ahead of the research that actually matters for your business, Micah curates the AI News Brief every day. No hype, no fluff, just the developments that affect how you run your team.
Subscribe at your.biggestgoal.ai
If you've been wondering why your AI investment isn't showing up in business performance, a new Glean report puts numbers on the problem for the first time. Six thousand full-time knowledge workers, co-authored with researchers at Stanford, UC Berkeley, and Harvard.
The finding: organizations seeing real business results from AI don't outperform everyone else on technology selection. They out-manage them.
That's the distinction most leadership teams are missing. And understanding it is the difference between the 13% who are moving the needle and the 87% who feel productive but can't point to company-level results. Here's what the data says, what it means for your team, and what to do about it.
Most executives are measuring AI by usage: who's using it, how often, how many prompts per week. That's the wrong metric.
AI didn't give your team a tool. It gave them a direct report.
A year ago, most of your team did the work themselves. Today they hand it to something else, wait, and decide if what comes back is good enough to ship. That's managing. And managing well requires skills most people have never been taught.
The root cause: your team got promoted overnight. No announcement, no training, no playbook for how to delegate to something that's fast, tireless, and occasionally confidently wrong. So most of them are winging it.
The result is what the report calls botsitting: 6.4 hours a week lost to re-running prompts, correcting outputs, and providing context the AI should have had from the start. That's not a model problem. That's what happens when you skip the briefing.
Here's the data point that reframes the whole conversation.
The top-performing organizations in the report don't win on model selection or tool spend. The same models are available to everyone. What separates them is management infrastructure:
The "aha" moment: the companies in the 13% treated AI adoption for what it actually is — a management challenge, not a software rollout. Everyone else ran a software rollout.
And one more number that stings: 53% of workers say the information their AI actually needs to do good work isn't accessible to it. That's not the model being limited. That's poor setup. Context-rich teams cut quality failures nearly in half compared to context-poor ones. The botsitting tax is almost entirely self-inflicted.
Here's the approach that separates the 13% from everyone else.
Tier 1: Build Context Once, Not Every Session
The Concept: The biggest driver of botsitting is workers re-explaining the same background to AI at the start of every task. That's dead weight, and it adds up to hours a week.
The Application: Build durable context, briefs, examples, quality standards, company background, that your AI tools can draw on automatically. The goal is that "what good looks like" exists before the work starts. Tools like Cowork are built for exactly this. One setup removes the re-briefing overhead permanently.
Tier 2: Teach the Brief, Not the Prompt
The Concept: Prompting gets all the attention. Briefing is the actual skill. The quality of what comes out of AI is almost entirely set by the clarity of what goes in.
The Application: A good brief is learnable, and it looks the same whether your direct report is a person or an AI: clear objective, relevant context, an example of a strong output, and a quality bar to evaluate against. Every experienced manager knows that under-briefing a human means spending the week rewriting their work. The same equation runs on AI.
Tier 3: Make Reviewing Part of the Job
The Concept: The report found that 69% of workers admit to "botshitting" — shipping AI output they haven't verified or can't fully explain. That's not a character flaw. It's a missing checkpoint.
The Application: Build review into the workflow as an expected step, not a suggestion. The companies winning aren't those with the heaviest AI usage. They're the ones where people review output before it ships, catch the errors, and hold the standard. Reviewing is the job now.
Measuring AI by usage instead of outcomes. Prompts per week is a vanity metric. The person generating 200 prompts and shipping unreviewed output is not your AI leader. The person generating 40 and getting clean, trustworthy work every time is. Most organizations are rewarding the wrong behavior.
The fix: Change what you're measuring. Track quality and rework rates. Recognize the people who produce the least corrected work, not the highest volume.
Blaming the model when the real issue is the brief. When AI output is consistently bad, most teams assume the tool isn't good enough and either switch products or give up. Rarely is the model the problem. The report is clear that the only real variable between strong and weak results is whether someone set the work up properly.
The fix: Before changing tools, audit the brief. Add context. Add an example. Set the bar before the work starts. Then evaluate.
Skipping training because "people will figure it out." This is the most expensive assumption in the building. The gap between 52% and 90% on training and support doesn't happen by accident. It's a deliberate choice the winning companies made. The companies that skipped it are the ones staring at a 6.4-hour botsitting tab every week.
The fix: Treat AI skills the same way you'd treat any other role-critical capability. Formal training, clear standards, and recognition for the people who do it well. That investment is exactly what separates the 13% from everyone else.
Before this report, it was easy to read slow AI ROI as a tool problem. After, it's clear it's a management problem. And that's actually good news.
Better models take years to develop. Better managers can be built this quarter.
The 13% who are moving the needle aren't waiting on the next model release. They're building the management infrastructure that lets their teams actually get clean work out the other side. That's the opportunity in front of every leadership team right now. It's also exactly the gap Biggest Goal exists to close: not teaching people to prompt, but teaching them to manage.
If you want to stay ahead of the research that actually matters for your business, Micah curates the AI News Brief every day. No hype, no fluff, just the developments that affect how you run your team.
Subscribe at your.biggestgoal.ai