Related service
Product Forge →
Concept to launch-ready, with you at every step.
Work · Marketplace / Two-Sided Platforms
An AI-native two-sided marketplace MVP
A concierge-style MVP for an early-stage two-sided marketplace, built to solve the cold-start liquidity problem before automating matching.
blueprint14 May 2026
8 weeks
Time to validate manual-to-automated matching (concierge MVP pattern)
43%
Startups citing poor product-market fit as the root failure cause
59%
Projected share of e-commerce from marketplaces by 2027 (Edge by Ascential, 2022)
This is a solution blueprint — a reference architecture we can build for your business. The figures above are cited industry benchmarks for this class of system, not results claimed for a named client.
The pipeline we'd build
4 stages
Stage 01 · Concierge
The first transactions are matched by hand inside a lightweight Next.js/Supabase app.
The challenge
An early-stage marketplace founder faced the classic cold-start problem — supply wouldn't join without demand, and demand wouldn't show up without supply — the single most common reason marketplace startups stall before reaching liquidity.
What we did
We build a 'concierge MVP' first: manually curating and matching the first transactions on the harder-to-acquire side (supply) inside a lightweight Next.js/Supabase app, before layering in Stripe Connect payments and AI-assisted matching once repeatable demand is proven.
The outcome
CB Insights' analysis of 431 shut-down VC-backed companies found poor product-market fit is the leading root cause at 43% — the concierge approach exists specifically to prove real demand before automating anything.
Stack
Next.jsTypeScriptSupabaseStripe ConnectVercelOpenAI APITwilio
Every two-sided marketplace faces the same chicken-and-egg problem at launch: buyers won't show up to an empty marketplace, and sellers won't join one with no buyers. CB Insights' post-mortem analysis of 431 shut-down VC-backed companies found poor product-market fit — often a symptom of never actually solving this cold-start problem — is the single largest root cause of startup failure, at 43%.
The concierge MVP: liquidity before automation
The concierge MVP sidesteps the chicken-and-egg problem by brute force: the team manually recruits and matches the first transactions on the harder side of the market (usually supply) inside a lightweight app, proving people actually want the transaction before any matching algorithm exists.
Concierge-to-automated marketplace path
How the platform evolves from manual matching to AI-assisted matching
Client appsCore platformConcierge opsAI matching (phase 2)
Tap any component above for its role and the real tech.
A blueprint of a concierge-MVP path — the AI matching layer is phase 2, added only once manual matching proves the pattern.
- Next.js web app (Client, Buyer + seller portals): The only interface either side sees — the manual matching behind it is invisible to them.
- Supabase (Data, Postgres + Auth + Storage): Backs listings, profiles, and transaction records from day one.
- Stripe Connect (Service, Split payments): Handles marketplace payment splits once transactions are proven repeatable.
- Concierge team (Service, Manual matching console): Humans manually pair the first transactions — the deliberate, temporary bottleneck that proves demand.
- AI matching (phase 2) (Model / AI, OpenAI API, semantic search): Automates the pattern the concierge team proved works, once volume justifies it.
Only once that demand is proven repeatable does AI-assisted matching and Stripe Connect payments layer in at full scale — a similar phased-launch discipline to our vertical healthtech SaaS MVP blueprint, which scopes a narrow workflow first rather than a broad platform.
At a glance
Client
Solution blueprint
Sector
Marketplace / Two-Sided Platforms
Service
Product Forge
Kind
blueprint
Headline result
8 weeks · Time to validate manual-to-automated matching (concierge MVP pattern)
Handover
Documented, tested code in your repository
Questions we were asked
Is 43% product-market-fit failure rate a real statistic?
Yes — it's CB Insights' own analysis of 431 shut-down venture-backed companies' stated failure reasons, a primary-source post-mortem dataset, not an estimate.
Why manually match transactions instead of building the algorithm first?
Because an untested matching algorithm optimizes for a problem you haven't confirmed exists yet. Manually matching the first transactions proves real demand exists before any engineering effort goes into automating it — cheaper to learn you're wrong early.
When do you actually add the AI matching layer?
Once the manual process shows a repeatable pattern — the same few match types working reliably — that pattern is exactly what the AI-assisted matching layer then automates, rather than guessing at rules upfront.
Same problem, different business?
We'll send the architecture and a realistic timeline for your version of this — no obligation.