Amara Chukwu
“Every tool we ship to help sellers, a fraud ring evaluates for abuse before our own team finishes the retro.”
- Martin Knapic, Product MoatAmara Chukwu works on marketplace trust at Harborlist in Lagos, which in practice means she's in a permanent arms race with the exact same kind of AI tooling her own team uses, just pointed at fraud instead of defense. Seven years in, she was one of the more clear-eyed people in this series about what that arms race actually looks like day to day.
Identity & Background
- Name
- Amara Chukwu
- Current role
- Product Manager, Marketplace Trust
- Current company
- Harborlist
- Location
- Lagos, Nigeria
- Years in product
- 7 years
Your story: how did you end up in product?
I started in customer support at Harborlist, handling disputes between buyers and sellers directly. I got obsessed with the patterns in the disputes that recurred, and asked to move into the team that could actually fix the patterns instead of resolving them one at a time.
Core PM Philosophy
How do you define product success?
Fewer bad interactions per thousand transactions, weighted by severity, and honest users seeing no more friction than last quarter. A safer marketplace that feels slower isn't a win.
What's your decision-making philosophy?
We push verification to the moment it matters - payout, high-value listings - instead of the moment someone signs up. Front-loading friction kills good sellers faster than it stops bad ones.
What's your biggest PM lesson or mistake?
Believing a single clever rule would close a fraud pattern. Fraudsters adapt within days. The lesson was to invest in signals and feedback loops, not one-off rules.
How do you manage stakeholders?
Support, legal, and ops see trust problems before we do. I treat them as sensors, hold a weekly triage with them, and close the loop on every escalation so they keep bringing them.
Product & Decision-Making
What's your favorite product you've shipped?
Response-time badges for sellers. A simple, honest signal outperformed several more elaborate reputation scores we tried before it.
AI & the Future
How is AI changing product management?
Every new fraud pattern our models catch gets fed back into a shared internal playbook within days, not quarters - the model's false positives and false negatives are now a standing weekly review, not an annual audit.
What's one thing about AI everyone gets wrong?
That trust is a detection problem. Most of it is design - how a listing is presented, how messaging works, what a first-time seller is asked to prove - long before any classifier runs.
Beliefs & Opinions
What's your unpopular product opinion?
Appeals infrastructure. Everyone invests heavily in detection and almost nobody invests properly in the appeals path for the person the system got wrong. I think that imbalance becomes a real reputational and regulatory risk for marketplaces as automated enforcement scales.
Company & Product
What does your product solve?
We're a peer-to-peer marketplace for buying and selling used equipment - construction tools, generators, that category - connecting individual sellers with buyers who'd otherwise be stuck with unreliable classifieds listings.
What's the biggest misconception about it?
That trust and safety just slows people down. Done well, it's what makes a stranger willing to buy a used bike from another stranger.
What's the market opportunity?
Our ICP is a small tradesperson or contractor buying or selling equipment they actually intend to use, not a reseller flipping inventory. Against generic classifieds, we win on trust infrastructure specifically - verified sellers and protected payments - which is also exactly what makes us a bigger fraud target.
More from the conversation
What do you do outside work?
I play a lot of chess, mostly online, at a fairly serious amateur level. Trust and safety work is oddly similar - you're constantly trying to think two moves ahead of an opponent who's also adapting to your last move.
Any moment you almost left this line of work?
A stretch that was purely reactive - firefighting fraud spikes with no time for structural fixes - nearly burned me out on trust work specifically. I pushed for a rebalanced mandate that included proactive detection, not just response, and staying became the right call.
How big is the trust team, and the company overall?
Trust and safety is 12 people across product, engineering, and operations, inside a company of about 180. I own the seller-side trust surface specifically - verification, listing quality, and payout risk.
What's the newest threat model in marketplace trust work?
Synthetic listings and reviews generated at a volume and quality that our old heuristics never anticipated. We're now in an arms race where both sides use similar tooling, which is a strange position to build product in.
What do you want the next generation of trust tooling to do?
Explain itself to the person it just penalized. A seller who gets suspended deserves a reason specific enough to act on, not a form-letter policy citation.
Enjoyed this conversation?
A new interview with a product person lands every week. Get it in your inbox.