How Much Does Moderation Really Cost: Manual vs. AI (With Numbers)

"How much does moderation actually cost?" is one of the first questions founders ask us. The answers online are either wildly optimistic ("$0.002 per check, AI solves it") or wildly pessimistic ("you need a BPO, budget $2M"). Neither is useful. This post builds a per-item cost model you can apply to your actual traffic, with real numbers we see across our customer base.

The three moderation models and what they cost

In-house human moderators

• Loaded cost per moderator: $45,000 to $120,000 a year depending on geography.
• Productive hours per moderator per year: ~1,700 (accounting for training, breaks, mental-health time).
• Items cleared per hour (text): 150 to 300. Much less for images.
• Effective cost per 1,000 text items: $100 to $470.

Outsourced BPO / vendor (Accenture, TELUS, Majorel, etc.)

• Priced per item, per hour, or per FTE depending on contract.
• Typical rates (low-cost regions): $1.50 to $4.00 per 1,000 items for text at scale.
• Higher for images, video, and 24/7 SLA.
• Minimum contracts usually start at $10k to $30k a month; not suitable for small platforms.

AI API moderation

• Typical rates: $0.10 to $0.50 per 1,000 text items, $1 to $5 per 1,000 images.
• Sub-50ms response, 24/7, no minimums.
• Edge cases still need human review (typically 2% to 8% of flagged content).

A concrete example: 1 million user messages per day

Let us price moderation for a hypothetical platform with 1M messages/day (~30M/month). Assume 3% of messages need classifier review and 0.3% need human review.

Option A: pure human moderation

• 30M messages/month. At 250/hour per moderator, you need ~600 moderator-hours per day, or ~75 full-time moderators.
• At $60k loaded cost each: $4.5M/year, or $12,300/day.
• Per-1000 cost: ~$12.

Option B: BPO handling 100% of messages

• 30M items × $3/1000 = $90,000/month = $1.08M/year.
• Plus contract minimums, often 20% to 30% uplift at this volume for SLA guarantees.
• Realistic total: $1.3M to $1.5M a year.

Option C: AI + targeted human review (the right answer)

• 30M AI calls × $0.30/1000 = $9,000/month in API cost.
• 3% of messages flagged for review = 900K/month.
• Of those, only ~10% actually need human eyes = 90K human reviews/month.
• 90K × $3/1000 (BPO) = $270/month, or one part-time internal reviewer ~$5k/month.
• Total: $14,000/month = $168,000/year.

The AI-plus-humans stack comes in at roughly 1/10th the cost of pure BPO at this volume, while also responding in 50ms instead of hours.

The hidden costs nobody mentions

Policy and training

Writing clear policies, training moderators on them, keeping them up to date. A full-time Trust & Safety ops role at $90k to $150k is realistic from the point you have dedicated moderators.

Tooling

Moderation dashboards, queue management, appeals workflow, transparency reporting. Either you build this (two to three engineer-quarters) or you buy it (roughly $2k to $10k a month for T&S platforms like Checkstep, Unitary, Spectrum Labs, Hive).

Legal and compliance

DSA reporting, GDPR compliance, DPA reviews with vendors. Figure on retainer or in-house counsel time: at least $30k a year for a mid-sized platform.

Mental health support

If you have human reviewers, you have a duty of care. Counselling programmes, rotation policies. Budget 3% to 5% of moderator compensation for ethical operation.

Model failures you did not catch

The cost of a single missed CSAM or terrorism incident (legal, PR, user trust) can dwarf everything else in this model by orders of magnitude. It is hard to price, but it is not zero, and it is not hypothetical.

When each model makes sense

• Pure human moderation: almost never a good choice at scale. Makes sense only at very low volume (<10K items/day) where the overhead of AI integration outweighs the labour.
• Pure BPO: makes sense when you need immediate capacity and do not have the engineering bandwidth to integrate an API. Good bridge, bad destination.
• AI + humans: the right answer for almost every platform above 100K items/day. The economics are not close.

The break-even point

For most B2C platforms, AI-assisted moderation becomes cheaper than pure human moderation somewhere around 50,000 items per month. Below that, the fixed cost of integration and policy design dominates. Above that, the variable cost of humans dominates, and AI wins by a growing margin as volume scales.

The modern question is not "can we afford moderation?". It is "can we afford not to moderate well, when our competitors can do it for pennies per thousand?"