The AI Tax No One's Measuring
HBR found that deploying AI without redesigning the work around it adds $186 per worker per month in labor costs. Not reduces. Adds. The AI is running. The outputs are being generated. And humans are absorbing the gap.
- AI implementation
- workflow design
- hidden costs
- AI adoption
- botsitting
HBR put a number on something that has been costing organizations money without appearing anywhere in their accounting. In a 2026 piece on what they termed "workslop," the researchers found that deploying AI without redesigning the work around it adds, on average, $186 per worker per month in labor costs. Not reduces. Adds. The AI is running. Outputs are being generated. And humans are absorbing the quality gap between what the AI produces and what the workflow actually requires, correcting errors, reviewing outputs before they ship, rerunning failed processes, and catching downstream problems that propagated before anyone noticed. This isn't an AI failure state. These are functional, deployed systems producing the tax as a structural feature of how they were implemented.
Glean's Work AI Index research gave the same phenomenon a different name: botsitting. Employees spending measurable time supervising AI outputs, not because the AI is broken, but because the quality standard was never defined at deployment and the default is human review of everything. The AI drafted it; the human checks it. The human's checking time approaches the time drafting would have taken; on net, the person now has two tasks where they had one. The AI deployment added a task rather than removing one.
What the tax adds up to
The scale of this matters. In a two-hundred-person organization where the workslop effect touches the hundred employees actively running AI-augmented workflows, $186 per month produces roughly $220,000 per year in added labor cost. Add the software licenses ($30 to $100 per seat per month depending on the platform), the integration work, the project budget for the initial deployment, and the opportunity cost of the workflows that didn't get redesigned because resources went to the AI implementation instead. The AI initiative that was supposed to produce cost savings is, in this accounting, a significant cost center.
This is likely the modal state for the large middle of organizations that have deployed AI without scaling it. The statistics that bracket this group are by now familiar: somewhere around 88% of organizations are using AI in at least one function; roughly 7% have scaled it with measurable EBIT impact. The 81% in between didn't fail to deploy. They deployed successfully. The workslop tax is probably what's running in most of those deployments right now, silently, without appearing on anyone's dashboard.
The cause is the sequence, not the tool
The cause isn't the tool. It's the sequence. AI was deployed alongside existing workflows rather than into redesigned ones. The workflow still requires human judgment at every decision point; the AI just produces a starting artifact. The human who used to draft the contract section, write the intake summary, or categorize the support ticket now reviews the AI-generated version and corrects it to meet the standard they know is required. The review takes less time than the original drafting, sometimes. It takes more time when the output requires significant correction. Nobody measured the delta before deployment, so nobody knows whether the net labor impact is positive.
The quality standard question is the one that doesn't get asked at deployment time. "What does acceptable output look like for this specific workflow?" isn't a typical item on an AI implementation checklist. The typical checklist covers integration, testing, user training, and go-live. What comes after go-live, which is the continuous monitoring of whether the AI is producing outputs that actually meet the standard the workflow requires, is typically left to the humans running the process. When those humans correct silently rather than escalating, and when the escalation path doesn't exist anyway, the correction work becomes invisible overhead. Botsitting.
The quality standard belongs in the brief
There's a governance implication here that runs upstream of deployment. An AI agency that proposes an approach without a post-deployment quality standard is configuring the botsitting condition into the engagement structure. The quality gate should be a named deliverable: what does correct output look like, measured how, checked by whom, on what schedule. An agency that treats this as the client's problem to figure out after go-live has delivered software and transferred risk.
The deeper implication is about where the measurement obligation sits. Organizations that measure AI deployment by completion (the agent is live, the integration is running, the training is done) are measuring the wrong thing. The relevant metric is labor impact at ninety days and six months post-deployment. Is the workflow taking less time? Are outputs meeting the quality standard without correction rates that erode the time savings? These are questions a buyer should build into the engagement scope before signing, because the agency's incentive is to complete the build, not to own the quality of what runs after it.
The $186 per worker per month isn't a fixed cost of AI adoption. It's a predictable consequence of deploying AI without defining what good output looks like and who owns it ongoing. That's a quality standard question, and it belongs in the engagement brief before day one.
If you want that question answered for your specific situation, the Forge Playbook does it. Answer a few questions about your business and we'll put together a tailored outline of which workflows are worth automating and what a realistic budget looks like for each. Free, no obligation, takes about three minutes.
