Homework in an Online School: Building a Review Process That Does Not Burn Out Your Team
A course without assignments is a video course people watch at 1.5x and abandon by lesson three. A course with assignments but no real review process is guaranteed chaos: submissions get lost in Telegram, the curator replies on day five, and the student has already cooled off and left.
Review practice is what separates "we sold access to videos" from "we got a person to a result." It is also the most expensive operation you run.
Calculate what one review costs you
A cohort of 120 students. 8 modules, one assignment each. That is 960 submissions. At 12 minutes per submission (read, check against criteria, write feedback), you get 192 hours. At $5/hour that is $960 per cohort, or $8 per student.
Most schools have never run this number, which is exactly why it eats them. Every minute you cut from the review cycle gets multiplied by 960.
Three review models and how to mix them
The core mistake is assuming all assignments are equal and all must reach a curator.
1. Auto-check. Quizzes, checklists, assignments with a reference answer unlocked after submission. Costs nothing, runs instantly. Good for testing theory comprehension and step sequences.
2. Peer review. Students review each other against a strict checklist. Double benefit: people learn by analysing someone else's work. Only works with explicit criteria — otherwise it degrades into "nice job!"
3. Curator. Human feedback. The most expensive and the most valuable — spend it on assignments involving judgment, strategy or creativity.
A working split: 60% auto-check, 25% peer review, 15% curator, with the curator handling the 2–3 pivotal assignments that determine the outcome. Those 960 submissions become 144. Those 192 hours become 29.
A vague prompt costs you hours
Half of a curator's time goes not into feedback but into decoding what the student meant.
Bad: "Describe your target audience."
Good: "Describe one audience segment using this template: who they are / what job they are solving / where they look for solutions / three phrases from real reviews. Max 200 words. A filled example is attached."
Every reviewed assignment needs four things:
- A response template
- A checklist of 3–5 pass criteria
- A reference example of completed work
- A deadline with a concrete date
This is one-time work at course creation that saves dozens of hours per cohort.
SLA and queue: review as a process, not a mood
- 48-hour SLA on business days — the threshold where a student still remembers what they did.
- A queue with statuses: new → in review → needs revision → passed. Work is never "somewhere in a chat."
- A revision cap of 2 rounds. After that the curator either passes the work or jumps on a short call. Endless ping-pong kills motivation on both sides.
- Escalation: anything sitting over 72 hours is flagged to the team lead.
Three weekly metrics: average response time, share of submissions passed on the first try, and submissions per curator. If first-try pass rate is under 50%, the problem is your assignment wording, not your students.
The 3–1–1 feedback template
- 3 — what specifically worked (quote the submission)
- 1 — the single most important fix
- 1 — the next step or a question to think about
Add a bank of stock comments: 20–30 snippets cover roughly 80% of all remarks, because students make the same mistakes. The curator pastes a snippet and adds two personalised sentences — time per submission drops from 12 minutes to 5.
Where AI fits
AI should not assign the grade — that destroys trust in feedback. But it removes the mechanics:
- checks formal requirements (length, all template sections present, links included)
- extracts the structure so the curator does not read three pages for two points
- drafts feedback against the checklist for the curator to edit rather than write from scratch
- flags submissions that look weak or unusually strong so curators start there
The rule: AI prepares, a human decides and signs off.
Handling students who never submit
Use a cascade, not a single blast:
- +24 hours past deadline — a short bot reminder
- +72 hours — a personal message asking "what blocked you?" instead of a reprimand
- +7 days — an offer to move to the next cohort or lock in a new deadline
Step one automates; step three does not. Step three is what brings people back.
Pre-launch checklist
- Every assignment has a template, checklist and reference example
- You have decided what goes to auto-check, peer review and curators
- The SLA is written down and visible to students
- There is a status queue, not a chat
- The revision cap is fixed
- A bank of stock comments is ready
- Deadline reminder cascade is configured
- A weekly 30-minute curator sync for hard cases
At CREO we built the curator workspace on exactly this logic: a status queue of submissions, feedback templates, deadlines and reminders in one place instead of five tabs. Alongside it: gamification that keeps course completion at 70–80%, Telegram funnels, a sales page builder and a CRM. Our support replies in about 15 minutes, and the platform already serves 100,000+ students.
Beta testing starts in August 2026. Join the early list at platform.creo.ua — beta testers get first access to running the school through MCP.
FAQ
How long should reviewing one assignment take?
Aim for 5–7 minutes when you have a checklist, a feedback template and a bank of stock comments. Without them curators spend 12–15 minutes, most of it decoding the submission and inventing a response structure.
Can AI fully replace a curator?
No. AI is good at checking formal requirements, compressing a submission into its structure and drafting feedback, but the final call must be human — otherwise students stop trusting the assessment.
What review SLA is considered normal?
48 hours on business days — the threshold where a student still remembers the context of their work. Anything over 72 hours should escalate automatically to the team lead.
