The Content Moderation Crisis: Scale, Cost, and Burnout
As your platform grows, user-generated content (UGC) explodes. A marketplace with 10,000 sellers generates thousands of product listings daily. A social network with 100,000 users sees thousands of comments, posts, and messages. Each piece of content needs review: Does it comply with your policies? Is it fraudulent? Is it abusive?
The traditional solution is hiring in-house moderators. UK-based content moderation staff cost £25,000–£35,000 annually per person. Add training, tooling, and workspace: you're at £30K–£40K per FTE. For a platform with 5,000 daily items needing review, you need 3–5 moderators. That's £90K–£200K/year in payroll alone.
But cost is only half the problem. The other half is human: content moderation is emotionally exhausting. Moderators see disturbing images, abusive messages, and fraud attempts daily. Burnout is high. Turnover averages 18–24 months. When your experienced moderators leave, you lose institutional knowledge and accuracy drops.
Here's where offshore moderation changes the game. Kenya has a 85%+ English-proficient workforce, strong cultural alignment with Western content policies (no translation needed for nuance), and lower cost. A Kenya-based moderator costs £7,500–£9,500/year. Same accuracy metrics. Better wellbeing outcomes because we implement structured breaks, mental health support, and rotation through different moderation types (combat fatigue).
Types of Moderation and Accuracy Requirements
Not all moderation is equal. Different content types require different skill levels and accuracy standards.
Type 1: UGC Moderation (Text + Image)
Review product listings, posts, comments against policy compliance (harassment, hate speech, misinformation, spam). Accuracy requirement: 92–95% (false positives are acceptable; false negatives are costly).
Moderator skill: Medium. Requires cultural understanding of context and intent, not deep technical knowledge.
Volume: 200–500 items per moderator per 8-hour shift.
Type 2: Fraud Detection
Identify suspicious seller accounts, payment mismatches, identity fraud. Accuracy requirement: 98%+ (false positives disrupt legitimate users; false negatives enable fraud).
Moderator skill: High. Requires financial literacy, pattern recognition, and judgment.
Volume: 50–100 items per moderator per shift (much lower throughput due to complexity).
Type 3: Abuse & Harm Assessment
- Determine if content violates abuse/CSAM policies. Accuracy requirement: 99%+ (false negatives enable real harm).
- Moderator skill: High. Requires sensitivity training, knowledge of harm indicators, escalation judgment.
- Volume: 50–150 items per moderator per shift.
Type 4: Appeal Reviews
- Re-review content that users dispute. Required: independent eye, consistency check. Accuracy: 90–95%.
- Moderator skill: Medium-high (requires nuanced judgment).
- Volume: 100–250 appeals per moderator per shift.
Assembling Your Kenya Moderation Team
Comparison
| Role | Responsibility | Typical Cost (Kenya) |
|---|---|---|
| UGC Moderator (Tier 1) | Reviews product listings, comments, posts. Applies content policies consistently. Escalates borderline cases to Tier 2. | £7,500–£9,000/year |
| Fraud Analyst (Tier 2) | Investigates suspicious accounts, payment patterns, identity fraud. Works with payment partners on chargebacks. High accuracy, low volume. | £9,500–£12,000/year |
| Harm Reviewer (Tier 3) | CSAM, abuse, violence assessment. Trained on harm indicators. Mental health support and 3-day rotation max per week. | £10,000–£13,000/year |
| Appeals Reviewer (Tier 2) | Re-reviews user appeals against content removal. Independent eye. Ensures consistency and fairness. | £9,000–£11,000/year |
| Moderation Manager (Kenya, full-time) | Quality oversight (random audits of 10–15% of decisions), training, escalation judgment, team morale. Liaison with UK legal/policy team. | £12,000–£15,000/year |
| Policy & Training Lead (UK, part-time) | Updates moderation guidelines, conducts monthly calibration sessions, reviews edge cases for policy consistency. | £5,000–£8,000/year (10 hrs/week) |
- Headcount calculation: For a marketplace with 5,000 items/day needing moderation:
- 3 UGC moderators (900 items/day @ 300/moderator)
- 1 fraud analyst (300 items/day)
- 1 harm reviewer (200 items/day)
- 1 appeals reviewer (500 appeals/week)
- 1 moderation manager
- Total: 7 FTEs + 1 part-time UK policy lead. Annual cost: ~£75K–£90K (Kenya) vs. £210K–£280K (UK). Saving: 65–70%.
Accuracy Metrics and SLA Frameworks
Moderation quality is measured by three metrics: accuracy, consistency, and appeal rate.
Accuracy (Primary Metric)
- Percentage of decisions that align with policy. Measured via random audit and user appeals. Target by moderation type:
- UGC moderation: 92–95% accuracy
- Fraud detection: 97–99%
- Abuse/harm: 98–99%
- Appeals: 90–95%
Consistency (Secondary Metric)
Two moderators should make the same decision on the same content. Measured by comparing decisions on duplicates or policy edge cases. Target: 90%+. Inconsistency indicates unclear policy or training gaps.
Appeal Rate (Outcome Metric)
Percentage of decisions users dispute. If users appeal 5–8% of removals, it suggests decisions are borderline or unclear. If appeal rate climbs above 10%, investigate training issues or policy drift.
SLA Framework
- UGC: 99% of items reviewed within 4 hours of upload.
- Fraud: 95% of flagged accounts reviewed within 2 hours (time-sensitive to prevent fraud spreading).
- Appeals: 90% resolved within 24 hours.
- Escalations: 100% of harm/CSAM cases escalated to UK legal team within 30 minutes.
- Quality audits: 10% of decisions random-sampled weekly. 5% re-reviewed after appeal.
Moderator Wellbeing and Rotation
Content moderation burnout is real. Research shows moderators viewing disturbing content have PTSD rates of 10–15%. We combat this via structured rotation and support.
Rotation Strategy
Moderators rotate between content types weekly. Example week:
- Mon–Tue: UGC moderation (lower emotional load)
- Wed: Fraud detection (analytical, less emotional)
- Thu: Appeals (intellectual challenge, lower stress)
- Fri: Training + calibration (no live moderation)
- This prevents 'content fatigue' where a moderator sees too much abuse in a single week. Different content types engage different cognitive skills.
Support Structures
Peer debrief: 30-minute team sync daily to discuss difficult cases
- Mental health access: quarterly counseling support (budget £100–150 per moderator per quarter)
- Time off: unlimited PTO (Kenya cultural norm); 10 days mandatory annual leave with pay
- Training: 4 hours/month professional development (tone diversity, trauma-informed communication)
- Reporting: moderators flag burnout risk to manager. Early intervention prevents turnover.
- Result: Turnover for offshore moderation is 30–35% annually (vs. 70%+ for in-house teams). Better retention = better accuracy and lower training cost.
Moderation Tooling and Integration
Moderation works only with the right tools. You need a platform that surfaces content for review, logs decisions, and tracks metrics.
Essential tooling:
Review queue system (Crisp Thinking, Two Hat Security, or custom dashboard) that prioritises content by risk (flagged keywords, new sellers, high-volume accounts)
Decision logging (every decision is timestamped, moderator-attributed, reason-tagged for audit trail)
Appeal workflow (users can contest removal; appeals route to senior reviewer, then policy lead)
Integration with CMS/marketplace (moderation decision auto-applies: remove content, suspend account, flag for manual review)
Bulk action (when policy changes, retroactively review past decisions; e.g. 'find all listings with these keywords')
Analytics dashboard (accuracy, throughput, appeal rate tracked in real-time)
Most UK brands use a combination: public moderation platforms (Crisp, Two Hat) + in-house CMS integration. Treba provides the Kenya team + manages tooling setup.
Deployment Timeline and Training
Rolling out a moderation team takes 6–10 weeks depending on complexity.
Week 1–2: Hire and baseline training
Recruit UGC moderators, fraud analysts, and harm reviewers from Kenya. All candidates interview on English proficiency and cultural fit. You meet final candidates. They complete baseline training (platform orientation, company values, basic moderation rules).
Week 3–4: Policy documentation and deep training
Your policy lead (UK team) documents ALL moderation policies in writing: What content is allowed? What's harassment? What's spam? What's misinformation? Treba Kenya team learns policies via live training sessions. They review 100–200 historical cases and discuss edge cases.
Week 5–6: Calibration and shadow run
Kenya team and your policy lead review the same 50 items and discuss decisions. Goal: achieve 90%+ agreement on borderline cases before going live. Kenya team watches your current team (if in-house) or Treba's UK moderators handle live queue. They flag questions real-time.
Week 7–8: Parallel run
Kenya team reviews 50% of incoming queue independently. Every decision is checked by UK policy lead before applying. Accuracy tracked. If accuracy drops below 90%, you pause and retrain.
Week 9–10: Go live and stabilise
Kenya team handles 100% of queue. Accuracy audits run weekly (10% sample-check). If accuracy stays above 92%, continue. If it dips, pause and investigate (training gap? policy misunderstanding?). By week 10, team should be stable.
Key takeaways
• Content moderation at scale (5K+ items/day) requires 3–7 moderators.
In-house UK teams cost £90K–£280K/year; Kenya-based teams cost £75K–£90K (65% saving). • Moderation isn't one skill: UGC is 92–95% accuracy, fraud is 98%+, harm/abuse is 99%.
Your team needs specialists (Tier 1 UGC, Tier 2 fraud, Tier 3 harm reviewers). • Kenya has 85%+ English proficiency and cultural alignment with Western content policies.
No translation needed.
Accuracy matches UK standards when training is rigorous. • Moderator burnout is avoidable: rotate between content types, limit exposure to harm content (3 days/week max), provide mental health support, ensure flexibility.
This reduces turnover from 70% (in-house) to 30% (offshore). • Deployment is 6–10 weeks.
Weeks 1–4 hire + train, Weeks 5–6 calibrate, Weeks 7–8 parallel run, Weeks 9–10 go live.
Weekly accuracy audits prevent surprises.
Written by
Treba Research
Treba editorial team — expert analysis on outsourcing, compliance, and building distributed UK–Kenya teams.
