An unmoderated marketplace becomes a dumpster in 6 months: fake listings, counterfeit goods, payment scams, banned content. Moderation combines automated tools (Hive AI, Two Hat) and human teams. Budget 3 to 8% of your GMV in trust and safety from year 1.
TL;DR
- Hive AI: automated image/video moderation, USD 0.1–0.5 per 1,000 media
- Two Hat (Microsoft): real-time text/chat moderation
- Scam detection: ML rules + signals (new account, stolen photos, abnormal price)
- Human team: 1 moderator per 100,000 active listings
- Trust and safety budget: 3–8% of GMV at seed, 1–3% at maturity
Why moderation is non-negotiable
A marketplace is a fragile commons. An undetected scam listing costs: buyer refund (direct loss), reputation (buyer never returns), honest seller churn (unfair competitors), brand reputation (Twitter, Google reviews).
The 7 main risks
- Fraudulent listings: non-existent product, ghost seller.
- Counterfeit / fake products: watches, perfumes, Chinese electronics.
- Stolen photos: sellers copying competitor images.
- Payment scam: buyers or sellers diverting from official channel.
- Forbidden content: weapons, drugs, fake medicine.
- Chat harassment: insults, discrimination, spam.
- Fake reviews: auto-generated 5-star reviews.
The standard trust and safety stack
| Tool | Use | Cost |
|---|---|---|
| Hive AI | Image/video moderation | USD 0.1–0.5 / 1,000 media |
| Two Hat (Microsoft) | Text / chat moderation | USD 0.5–2 / 1,000 messages |
| Sift Science | Payment fraud detection | 0.5–3% of monitored transactions |
| Custom rules engine | Behavioral signals | Internal build |
| Human moderation | Complex cases, appeals | XOF 200,000–500,000 / Dakar moderator |
Hive AI in detail
Hive analyzes images and videos to detect: nudity, violence, weapons, drugs, counterfeit, brand logos. Latency ~200ms per image. REST API integration. For a 50,000 listings/month marketplace with 5 photos each = 250,000 images = USD 25–125/month.
Manual vs automated moderation
No 100% auto tool is enough. The empirical rule:
- Auto-reject: 60–70% obvious cases (nudity, weapons, blacklisted keywords).
- Auto-approve: 20–30% safe cases (verified account, clean history).
- Human review: 10–20% ambiguous cases → moderation team.
Team sizing
For 100,000 active listings, plan: 1 senior moderator (XOF 300,000/month) + 2 junior moderators (XOF 180,000/month each) + 1 manager (XOF 500,000/month). About XOF 1.16M/month of pure human trust and safety.
Scam detection: the 8 key signals
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Mature marketplaces build a risk score per seller account, based on:
- Account age: under 7 days = very risky.
- KYC complete or not: NINEA + RCCM + ID = safe score.
- Unique photos vs reverse image search: photos stolen from Google = likely scam.
- Price vs market: iPhone 15 at XOF 80,000 = near-certain scam.
- Geolocation vs declared address: VPN, mismatch = negative signal.
- Listing speed: 30 listings in 10 minutes = bot.
- Chat profile: refusing calls, pressuring to pay off-platform.
- Dispute history: over 5% disputes per seller = automatic ban.
Jumia case: anti-fraud combat
Jumia disclosed in 2024 that 6.8% of its seller accounts were suspended each quarter for fraud. Brutal ratio but necessary to preserve buyer trust. The algorithm combines internal ML + community reports.
FAQ
Q: Should you moderate listings before publication or after?
A: Optimal mix: pre-moderation for new sellers (under 30 days, fewer than 5 sales), post-moderation for established ones. Reduces friction without sacrificing quality.
Q: How to handle buyer-seller disputes?
A: 3-step process: (1) 48h for direct exchange, (2) escalation to marketplace moderation with evidence (photos, messages), (3) binding arbitral decision within 5 days. Document every case in a Dispute table.
Q: Hive AI or Two Hat — how to choose?
A: Hive = image/video. Two Hat = text/chat. They are complementary, not competing. A mature marketplace uses both + an in-house rules engine.
Q: Trust and safety budget for an XOF 1B GMV marketplace?
A: 3–5% of GMV = XOF 30–50M/year. Breakdown: 60% human team, 25% tools (Hive, Two Hat, Sift), 15% legal and compliance.
Conclusion
Trust and safety is not a cost center — it is defensive investment that protects buyer LTV and the seller ecosystem. Kolonell embeds the trust and safety stack into client marketplaces from design. Request a free quote or WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
