Fractional AI departments, shaped for two-sided platforms.

A marketplace at twenty to fifty employees does not look like a normal company at the same size. A normal company has one sales motion, one support queue, one ops backend. A marketplace has two of each. Supply acquisition is a four-person team chasing sellers, vendors, drivers, hosts, or whatever the supply side is called in your category. Demand acquisition is another four-person team chasing buyers, riders, guests, or whatever the demand side is called. Both motions ship outbound. Both motions write content. Both motions own retention. The org chart looks like two startups stapled together at the take rate, because that is structurally what it is.

Support is where the duplication gets brutal. Two reps cover the supply side, answering questions about payouts, listing rejections, KYC docs, dispute outcomes, Stripe Connect onboarding errors, vendor finance reconciliation. Two more reps cover the demand side, answering refund requests, order status, account recovery, payment failures, escrow release timing. Four reps for a twenty-person company, and they cannot cover nights, weekends, or the second language a real two-sided platform always needs. So you outsource part of it, the outsourced shop does not understand the platform mechanics, sellers churn quietly because their Tuesday payout question sat for thirty hours, and you find out at the next QBR when GMV is flat and supply attrition is up.

Ops is the third doubling and it is the one that breaks the founder. A normal company reconciles invoices. A marketplace reconciles platform fees, take rate cuts, refunded transactions, chargebacks, payout splits between vendor and platform, escrow holds, and dispute outcomes that change the books retroactively. One ops manager is doing the work of three because every transaction touches both sides. The first month of every quarter is firefighting. The marketplace fix nobody admits to is hiring six more people, which the unit economics do not survive past Series B. We covered the structural shape of the same problem in What is a Fractional AI Department. The short version: marketplaces are the cleanest case for fractional AI because the labor problem is not a function problem, it is a duplication problem, and duplication is exactly where AI departments compound.

The standard marketplace mistake is treating supply and demand acquisition as one motion. They are not. Supply acquisition is high-touch, low-volume, long sales cycle. A vendor signing onto your platform is making a multi-month decision about their distribution channel, their take rate exposure, their KYC compliance, their payout cadence. The conversation is closer to enterprise sales than to consumer growth. You can do twenty real supply conversations a week and that is a strong week. The metric that matters is supply conversion rate, supply onboarding time, and first-month vendor GMV. The labor problem on the supply side is research depth, not volume. Every outbound message has to know what the seller currently does, where they sell now, why your take rate is competitive against the alternatives, and what your category density is in their region.

Demand acquisition is the opposite shape. High-volume, low-touch, short cycle. Buyers, riders, guests, customers. The labor problem on the demand side is volume and language coverage. Programmatic SEO around the categories your supply side already covers. Performance creative on the channels your demand lives on. Email and push retention for the cohort that bought once and never came back. The metric that matters is cost per first transaction, time to second transaction, and ninety-day repeat rate. Treating these two motions with one growth team is why most marketplaces under fifty headcount are bad at both. The supply team writes generic outreach because they are also running demand. The demand team underinvests in performance because they are also writing supply emails.

Fractional AI separates the cadence cleanly. The supply side runs through the AI Sales Department configured for high-research, low-volume, multi-touch enterprise-style outreach against your vendor ICP. The demand side runs through content and performance loops configured for programmatic SEO, paid creative cadence, and retention flows in every market language. Same monthly invoice. Two clocks running in parallel. The supply team stops being the bottleneck on the demand team, the demand team stops cannibalizing the supply team, and the org chart finally stops looking like two startups stapled together.

Marketplaces are trust businesses. Network effects compound when both sides believe the platform protects them. Network effects unravel when one side starts telling the other that the platform does not have their back. The mechanism for that unraveling is the support queue, and the marketplaces that lose this most often are the ones that staffed one side and outsourced the other. Sellers find a Slack community within a week of joining, and the first thing they share is which platforms answer payout questions in under an hour and which ones make them wait three days. Buyers find a subreddit, and the first thing they share is which platforms process refund disputes in twenty-four hours and which ones bury the request. The trust signal travels faster than the marketing.

The fractional AI model treats both queues as one trust surface. Every ticket on either side is read with the policy logic and the transaction context that ticket actually requires. The seller asking about a delayed payout sees an answer that already knows their last three payout dates, their current Stripe Connect status, and whether there is a hold flag on the account. The buyer asking about a refund sees an answer that already knows the seller policy, the escrow state, and the regional consumer protection rule that applies. Both sides get the same response speed, in their own language, at three in the morning. The trust signal travels in your direction for once.

The side-aware routing is the part that makes this real. A normal support stack has one queue and one set of policy logic. A marketplace stack has two queues with completely different policy stacks and completely different escalation paths. A seller dispute about a chargeback escalates to your trust and safety team. A buyer dispute about a damaged item escalates to your seller team. A KYC question on the supply side escalates to compliance. A refund question on the demand side escalates to your finance ops. The routing is the moat. The fractional model encodes the routing at kickoff and applies it per ticket per side automatically, so the human escalations that hit your team are already at the right person, with the right context, already triaged. The AI Support Department page covers the mechanism in more depth for single-sided businesses. The marketplace version doubles the surface area and the routing complexity, and the engine scales to it without doubling the cost.

The reason marketplace ops eats a senior person two days a week is that every transaction touches four different ledgers. The order ledger from the demand side. The payout ledger to the supply side. The platform fee ledger that captures your take rate. The dispute ledger that holds the escrow until the outcome is decided. A clean order moves through all four cleanly. A messy order, which is most of them at scale, generates a refund here, a chargeback there, a partial payout, a take rate adjustment, a Stripe Connect transfer reversal, and a manual journal entry to make the books match. Multiply by ten thousand orders a month and the senior ops person is doing nothing except matching transactions all week.

The fractional Ops engine reads all four ledgers continuously and closes them on a live pipeline. Stripe Connect transfers reconcile against the order ledger automatically. Platform fees split between revenue and pass-through automatically. Chargebacks update the take rate exposure automatically. Disputes that change the books retroactively trigger the right journal entries automatically. The vendor finance side gets a clean payout statement every cycle in their language. The platform finance side gets a clean take rate report every cycle. The board reporting that used to wait until the Sunday before the meeting refreshes every minute, the same way the AI Ops Department closes the books for Printdeal across four production hubs. Marketplaces are the heavier version of the same pattern. Same engine.

The piece that matters most is dispute handling, because disputes are the place where reconciliation, support, and trust all converge. A dispute opens with one side claiming the other side breached the policy. Both sides have to be heard. Evidence has to be collected. Escrow has to be held. A decision has to be made within a deadline the platform set in the terms of service. The fractional Ops engine handles the workflow, the Support engine handles the communication with both sides in their own language, and the human escalation is only the final decision call. Marketplaces that previously sat on disputes for two weeks close them inside seventy-two hours on the engine, and the trust signal on both sides moves in the right direction for the first time.