Column
AI Is Most Valuable for Work People Know They Should Do But Keep Putting Off
A randomized controlled trial with 11 TAs and 88 students found that AI draft assistance increased feedback provision by 10.8 percentage points — without quality loss or increased time. The mechanism wasn't efficiency. It was reduced initiation friction. Here's why that distinction matters for how you deploy AI at work.
Hi, I’m Keito Inoshita from Affectosphere Group.
“Not enough feedback” is a recurring complaint in most organizations.
Managers want to give meaningful 1:1 feedback. But the week fills up, the review period arrives, and the feedback that should have happened throughout the quarter gets compressed into a single rushed session.
Customer success teams know follow-up touchpoints drive retention. But writing them — specifically writing them — keeps getting pushed to later.
A paper published on arXiv in June 2026 (Romina Mahinpei, Victoria Dean, Ruth Fong, Lydia T. Liu, Manoel Horta Ribeiro, arXiv:2606.03095) ran a rigorous randomized controlled trial on this exact problem in an educational setting. 11 TAs and 88 students. AI-assisted feedback drafts versus standard conditions.
The result: AI draft assistance increased feedback provision rates by 10.8 percentage points, feedback length increased, and neither quality nor time-per-response deteriorated.
The mechanism behind it points to something useful beyond the specific finding.
3 Points for Today
- The highest-impact target for AI assistance is discretionary work — tasks that are valuable but not enforced, and therefore commonly deferred. Not efficiency gains on existing workflows; activation of valuable work that wasn’t happening.
- AI drafts work by reducing initiation friction, not by reducing effort: the blank page is the barrier, not the total work involved. AI drafts bypass the blank page.
- 1:1 feedback, performance reviews, CS follow-ups are direct analogues: wherever valuable feedback-type work is systematically deferred due to initiation cost, AI drafts are a deployable intervention.
① The Concept of Discretionary Work
The researchers use the term “discretionary work” to describe a category of tasks that is easy to overlook in typical AI productivity discussions.
Discretionary work: valuable, non-mandatory, and therefore frequently deferred when time is constrained.
In the study’s context: TAs are not required to write detailed personalized feedback on every student’s code submission (they grade, but granular feedback is optional). Everyone understands that detailed feedback improves learning outcomes. But the optional nature means it competes with everything else in a TA’s week — and often loses.
The RCT tested whether providing AI-generated draft feedback changed this calculation. The AI draft gave each TA a starting point based on the student’s submission — not a replacement, but a first version to edit and send.
The 10.8 percentage point increase in feedback provision rates is the primary finding. But the secondary finding is equally important: feedback length increased. The AI draft didn’t just make it easier to provide minimal feedback. It also enabled TAs to provide more substantive feedback than they would have otherwise.
② The Mechanism: Initiation Friction, Not Effort Reduction
The intuitive explanation for why AI drafts would increase feedback provision: less work to do, lower barrier to completion.
But the researchers’ interpretation is more specific, and more useful.
The intervention worked not because it reduced the total effort of writing feedback, but because it eliminated the initiation barrier — the blank page.
Starting from scratch activates a different kind of resistance than continuing from a first draft. Generating the first sentence of a piece of feedback requires making choices about framing, scope, tone, and emphasis. A draft resolves those choices provisionally and invites editing rather than creation.
For knowledge workers, this distinction is familiar. Tasks don’t get deferred because they’re too long. They get deferred because starting them feels like a larger commitment than the available time or energy supports.
AI drafts change the psychological structure of the task: it’s no longer “write feedback for this student” (open-ended, starting from zero) but “review and edit this draft” (bounded, starting from something).
This “reduced initiation friction” framing is a more precise targeting criterion for AI assistance than general productivity. The question becomes: which valuable-but-deferred tasks in your organization have high initiation friction as the primary barrier?
③ Applying This to Workplace Workflows
The TA-to-student feedback context maps directly onto several common organizational pain points.
1:1 feedback and performance reviews
Managers are expected to give meaningful, specific feedback in 1:1 conversations and performance review cycles. This is clearly valuable work. It’s also frequently under-delivered — not because managers don’t care but because writing specific, thoughtful feedback takes initiation energy that doesn’t always exist at the end of a full week.
An AI draft based on recent 1:1 notes, meeting logs, or project data changes the task from “write feedback about this person’s quarter” to “review and refine this draft.” The initiation barrier comes down.
Owned by: HR business partner and manager development function. A measurable KPI: track 1:1 feedback completion rates and feedback length (characters or specificity score) before and after introducing AI drafts into the HRIS workflow. The research gives you a benchmark to compare against: 10.8 percentage points is a meaningful lift.
Customer success follow-up
Customer success work is full of high-value, discretionary touchpoints: post-onboarding check-ins, usage-based health score follow-ups, proactive outreach before renewal conversations. All of these are “should do, easy to skip when busy.”
AI drafts generated from meeting notes and product usage data give CS reps a starting point that reduces the writing cost. The research finding — that draft assistance increases both quantity and quality of feedback without increasing time cost — suggests the same dynamic should hold here.
KPI: follow-up touchpoint completion rate, and the correlation between AI-assisted follow-up frequency and 90-day retention or expansion rates.
Mentor and senior employee onboarding comments
During employee onboarding, structured feedback from senior team members on early work product is valuable for accelerating skill development. It’s also easy to deprioritize. An AI draft generated from the new employee’s submitted work gives the senior employee a starting point rather than a blank page.
This requires an HRIS or onboarding platform integration — but it’s a configuration choice, not a new build, for most enterprise tools.
Rethinking Where AI Creates the Most Value
Most AI ROI discussions focus on efficiency: how much faster can existing work be completed? This research suggests a different frame: which valuable work is currently not happening due to initiation friction, and where can AI drafts change that?
“Discretionary work” as a targeting category reorients AI investment prioritization. The highest return may not be in tasks that are already being done well (where efficiency gains are incremental) but in tasks that are being systematically skipped — where AI assistance moves the needle from “not done” to “done.”
That’s a different kind of value, and a different kind of ROI calculation.
The question worth asking: in your organization, what valuable-but-deferred work is piling up because starting it is harder than it should be?
That’s where AI drafts might have the highest impact.
That’s all for today!
References
- Romina Mahinpei, Victoria Dean, Ruth Fong, Lydia T. Liu, Manoel Horta Ribeiro (2026). AI Assistance for Discretionary Work: Increasing Feedback Provision in Higher Education. arXiv preprint.
* This article was written in part with AI assistance and may contain inaccuracies.