AI workslop eats up to 38% of your reported time savings

Your dashboard says the team saved nine hours last week. The team says it felt like none. Both are telling the truth, because the dashboard counts what the agent produced and nobody counts what it took to fix. Forrester's Predictions 2026 puts unmeasured rework at 22 to 38 percent of self-reported time savings in mature AI programs, and past 50 percent in early-stage ones. That is not a rounding error. That is up to a third of your reported ROI, gone before anyone opens a spreadsheet.

What AI workslop actually costs

The phenomenon has a name and a research base. A September 2025 study by BetterUp Labs and the Stanford Social Media Lab, published in Harvard Business Review, surveyed 1,150 full-time desk workers and found that 40 percent had received AI workslop in the previous month: output polished enough to pass a glance and hollow enough to push the real work downstream. Each incident took an average of two hours to resolve. That prices out at roughly 186 dollars per employee per month, and over 9 million dollars a year for a 10,000-person organization.

Two hours per incident is the number to sit with. An agent that drafts a report in four minutes did not save the three hours a human writer would have spent, not if a colleague then spends two hours rewriting it. It saved one hour. Your dashboard recorded three.

Why the rework never shows up

Rework is invisible by design. The agent run is instrumented: it has a trace, a token count, a duration, a completion event. The fix is not. It happens twenty minutes later, in a different tool, by a different person, logged nowhere. Workday's survey of 3,200 business leaders found that 85 percent of employees now save between one and seven hours a week with AI, and nearly 40 percent of that saved time is immediately swallowed by correcting and reworking AI output.

Forrester's prescription is blunt: treat edits to agent output as a first-class telemetry event, not a side effect. If your measurement stops at "task completed," you are reporting gross savings and calling it net. The gap between those two numbers is exactly where the 22 to 38 percent lives.

The honest formula is net, not gross

The correction is one subtraction:

net time saved = (human baseline - agent handling time) - rework minutes

Take a concrete case. An agent handles invoice extraction with a human baseline of 12 minutes per document and runs 3,000 times a month, a normal volume for a production n8n or LangSmith workload. Gross savings: 600 hours a month. Apply Forrester's mature-program rework band of 22 to 38 percent and the honest range is 372 to 468 hours. The 132 to 228 hours in between are the fiction, and finance will find them, because finance eventually reconciles every claimed saving against headcount and output.

The gross number feels better in a QBR deck. The net number is the one that survives the follow-up question.

Record the fix as part of the task

The fix is cheap: attach the rework to the same task record that claimed the saving. One event per completed task, with the correction time on it:

curl -X POST https://humanhours.dev/api/v1/track \
  -H "Authorization: Bearer hh_live_..." \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "invoice-processor",
    "task_type": "invoice_data_extraction",
    "outcome": "success",
    "human_baseline_minutes": 12,
    "metadata": { "rework_minutes": 4, "edited_by_human": true }
  }'

Three rules make the number defensible:

• Sample the runs you did not hear about. Complaints surface the worst 5 percent. Audit a random sample of completed tasks each week and ask the receiving human one question: did you edit this, and for how long?
• Price the reviewer's hour, not the median hour. Workslop flows downstream to the people senior enough to catch it. Their time costs more than the blended rate in your ROI model.
• Report the band, not the point. "372 to 468 hours net" reads as measurement. "600 hours saved" reads as marketing, and gets treated accordingly.

Smaller and true beats bigger and hollow

The teams that keep their agent budgets through the 2027 planning cycle will not be the ones with the biggest gross number. They will be the ones whose number nobody can puncture. A time savings claim with the rework already subtracted is smaller, and it is the only kind that holds up in the room where budgets get cut.