Content Moderation Cost at Scale: Stream AI vs. Manual Review
TL;DR
- Reviewers can't read everything, so a typical team checks only the 10% that users report or a keyword list flags. At 10M items a month, that costs about $155,700 manually. Meanwhile, Stream AI Moderation checks all 10M for about $21,900 at list pay-as-you-go rates, which is 86% less.
- Gumtree saw about 80% fewer fraudulent messages after turning on Stream AI Moderation. At its volume, reviewing 10% by hand would cost about $137,000 a month.
- Manual review costs more with every reviewer you hire. AI moderation costs less per item as volume grows.
- AI handles the volume, and human reviewers handle the edge cases.
A platform handling 10 million pieces of user content a month (chat messages, comments, listings, reviews, profile photos, uploads) needs about 33 full-time reviewers just to check 10% of it by hand. That team costs about $155,700 a month based on our estimates in the sections below.
Stream AI Moderation checks all of that content for about $21,900 a month at list rates. Enterprise plans are up to 75% off list, which would bring that to about $5,500.
This post shows the full cost comparison, what Stream customers saved after switching, and how Stream moderation pricing works.
What Manual Review Costs
The cost of manual review comes down to four numbers: how much content you get, how much of it a person has to look at, how fast a reviewer works, and what an hour of their time costs.
Monthly cost = items reviewed ÷ items per hour × loaded hourly cost
Hourly cost. The average US content moderator earns $20.44 an hour in base pay (Payscale). Wages make up 70% of what US private employers spend on compensation. The other 30% is benefits, payroll taxes and insurance (BLS). That puts the loaded cost of a reviewer at about $29.20 an hour.
Speed. Researchers found moderators are given about 1,500 items per 8-hour shift, which leaves 15 to 20 seconds per decision (University of Queensland). That works out to about 187 items an hour, with very little time to read context.
Coverage. At that speed, most teams can't look at everything. In our model, the team reviews 10% of content: items users report and items that hit a keyword list. The other 90% goes live without anyone checking it.
Then there are the costs that aren't in reviewer salaries:
- Response time. Harmful content stays up until someone gets to it. A report filed at 2 a.m. waits for the morning shift unless you staff overnight.
- Coverage. 24/7 shifts, and reviewers who can read every language your users write in.
- Burnout and turnover. Reviewers exposed to graphic content need wellness support, and every reviewer who leaves means recruiting and training a new one.
- Consistency. Two reviewers looking at the same message often make different calls, especially at 15 seconds per decision.
Stream vs. Manual Review by Content Volume
| Monthly content items | What that looks like | Items reviewed by hand (10%) | Reviewers needed | Manual cost / month | Stream cost / month | Savings |
|---|---|---|---|---|---|---|
| 100K | Startup or early launch app | 10,000 | 1 | $4,700 | $120 (after the $100 free credit) | 97% |
| 1M | About a 50K MAU social app, or a marketplace with 200K users | 100,000 | 3.3 | $15,600 | $2,100 | 87% |
| 10M | About a 500K MAU social app, or a marketplace with 2M users | 1,000,000 | 33 | $155,700 | $21,900 | 86% |
| 100M | About a 5M MAU social app, or a marketplace with 20M users | 10,000,000 | 333 | $1,557,300 | $219,900 | 86% |
How we calculated this: the manual team reviews 10% of content at 187.5 items per hour, 160 working hours per reviewer per month and $29.20 an hour. Stream cost uses list pay-as-you-go rates ($2.00 per 1,000 messages, $4.00 per 1,000 images) on a mix of 90% text and 10% images, minus the $100 monthly credit.
Platform examples assume 20 items per user per month for a social app and five for a marketplace. The manual cost leaves out management, QA, tooling and overnight shifts, so the real number is higher.
We measure by content volume instead of users because content per user varies widely. A live chat community can generate hundreds of messages per user each month, while a marketplace might see a few listings and a handful of buyer messages. Both manual review and Stream are priced per item, so content volume is what sets the cost.
The savings percentage holds steady across the table because both columns use list rates. In practice the gap widens as you grow: platforms at 100M items a month move to enterprise rates, which lower Stream's per-item price by up to 75%, while a 333-person review team adds managers, QA leads and recruiting on top of the hourly cost.
What Stream Customers Saw
At a Glance
| Customer | Monthly volume | Manual review cost at same volume | Stream list price at same volume | Savings with Stream | Result with Stream |
|---|---|---|---|---|---|
| Gumtree | 8.8M+ text moderations | About $137,000 a month (29 reviewers) | About $17,500 a month | 87% | About 80% fewer fraudulent messages |
| Peerspace | 1M+ text moderations | About $15,600 a month (3 reviewers) | About $1,900 a month | 88% | Phishing and off-app bookings routed to trust and safety before a booking leaves the platform |
| Tradeblock | Hundreds of scam messages a day | About $21,300 a month (5 reviewers to keep one on shift 24/7) | Pending volume | Pending volume | Daily scam messages down to near zero |
| Collectibles marketplace | About 1.27M items (1.2M text, 68K images) | About $19,700 a month (4 reviewers) | About $2,600 a month | 87% | 21,600 automatic actions in 30 days; 0.56% reversed by moderators |
Gumtree
Gumtree, a classifieds marketplace in Australia and the UK, dealt with fraud that kept changing: impersonation, phishing and QR code scams that came back in a new form every time one method was shut down. Buyers and sellers also shared contact details in chat to move deals off the platform.
Gumtree uses Stream's AI text moderation, OCR to read scams hidden inside images, and PII masking. Fraudulent messages dropped by about 80%, and Stream ran more than 8.8 million text moderations in a single month, catching scammers who sent payment and contact details past text filters.
At that volume, reviewing even 10% by hand would take about 29 full-time reviewers and $137,000 a month. At Stream's list rates, the same volume costs about $17,500 a month, 87% less.
Peerspace
Peerspace, a marketplace for booking venues by the hour, saw hosts targeted by phishing and by guests proposing off-app bookings to avoid fees. Peerspace set up a custom "Platform Bypass" AI label with an automated ticketing workflow. More than 1 million text moderations a month are now flagged and routed to the trust and safety team before a booking leaves the platform.
Reviewing 10% of that volume by hand would take about 3 reviewers and $15,600 a month. Stream's list cost at the same volume is about $1,900.
Tradeblock
Tradeblock, a sneaker trading marketplace, was getting hundreds of scam messages a day and had no moderation team to fight them. With LLM detection tuned to marketplace language, daily scam messages went from hundreds to near zero. Tradeblock never had to hire a review team. Phishing waves arrived at all hours, so catching them by hand would have meant keeping at least one reviewer on shift around the clock: about 5 full-time reviewers and $21,300 a month.
A Collectibles Marketplace
In 30 days, Stream took 21,600 automatic actions on a collectibles marketplace's chat. Moderators reversed 0.56% of them. That's about 115 hours of reviewer time, or $3,400 a month, the team didn't spend on clear-cut cases. Reviewing 10% of its roughly 1.27 million monthly items by hand would take about 4 reviewers and $19,700 a month, compared with about $2,600 at Stream's list rates.
How Stream Moderation Pricing Works
Stream AI Moderation is priced per item, by content type. These are the published pay-as-you-go rates:
| Content type | List price |
|---|---|
| Text | $2.00 per 1,000 |
| Images | $4.00 per 1,000 |
| Images + CSAM detection | $6.00 per 1,000 |
| Images + OCR (text inside images) | $7.60 per 1,000 |
| Full detection (images + OCR + CSAM) | $10.00 per 1,000 |
| Uploaded video | $0.80 per minute |
| Live video | $4.00 per 1,000 frames ($0.24 per minute at 1 frame per second; you choose the sampling rate) |
| Audio | $0.06 per minute |
Every account gets $100 in free credits each month, enough to moderate approximately 50,000 text messages. Credits can be used across text, images, video, and audio. The free tier includes the moderator dashboard, 40 AI harm engines and LLM review, so you can test on your own traffic before you pay anything.
Here is the 10M example worked through: 9 million text items at $2.00 per 1,000 is $18,000, and 1 million images at $4.00 per 1,000 is $4,000. Subtract the $100 credit and the total is $21,900 a month.
Enterprise plans commit to a yearly volume and get one flat rate per content type, with discounts up to 75% off list. The more you moderate, the lower your rate.
See current rates at getstream.io/moderation/pricing.
Where Human Reviewers Still Fit
For most platforms, AI moderation changes what people spend their time on. Stream removes clear violations automatically, lets clean content through, and sends uncertain cases to the review queue with the label and context attached. Reviewers handle edge cases, appeals, and policy decisions.
The result is a smaller team, with each reviewer spending time on the hardest decisions instead of clicking through obvious spam. In the collectibles marketplace example, moderators reversed fewer than 1 in 150 automatic actions.
Some platforms run with no moderators at all. Tradeblock had no moderation team when it started with Stream, and LLM detection took daily scam messages from hundreds to near zero on its own. Platforms with a narrow set of harms, like scams on a marketplace, can often let AI take every action.
FAQ
How much does content moderation cost per message, image or video?
Manual review costs about $0.16 per item reviewed ($29.20 an hour ÷ 187.5 items an hour). Stream's list price is $0.002 per message, $0.004 per image, $0.80 per minute of uploaded video and $0.06 per minute of audio. Enterprise plans are up to 75% off those rates.
Is AI moderation cheaper than outsourcing to a BPO?
BPO rates aren't published, so the math depends on your quote: multiply their per-item price by the number of items you'd send them. Even at half the in-house rate, reviewing 10% of 10M items a month would cost about $78,000, compared with about $21,900 for Stream to check all of it.
How accurate is AI moderation compared with human reviewers?
Human reviewers working at 15 to 20 seconds per decision often disagree with each other on the same content. Stream detects 35+ harms across text, image, video and audio in 50+ languages, with 98% accuracy in under 50ms, and applies the same policy to every item.
At what volume does switching from manual review pay off?
Almost immediately. At 1M items a month, a 10% review rate already needs about 3 reviewers and $15,600 a month, compared with about $2,100 for Stream. Under about 45,000 items a month, the $100 free credit covers it.
The Bottom Line
Manual moderation costs go up with every reviewer you hire. AI moderation costs less per item as you grow, checks everything instead of a sample, and leaves your team the decisions that need a person.