Subscribe to VERTICALS —> 📺 YouTube | 🎧 Spotify | 📖 Here on Substack
AI marketplaces are still human at the edges.
Trust and network effects—key factors of any marketplace—are inextricably human. But AI is collapsing the messy middle: intake, matching, and transaction prep. While it’s delivering on promises of lower friction and faster growth, cost (inference, humans-in-the-loop) and retention questions remain.
This week’s guest, Jack Greco, co-founded ACV Auctions (NYSE:ACVA), one of the most successful vertical marketplaces of the past decade. He walks through how ACV cracked the wholesale-auto market—buyer-first wedge, uniform data, and guarantees—and what founders can learn about building durable trust in a post-LLM world.
I) This Week’s Vertical Market Pulse
AI’s UI Problem (via Derek Xiao @ Menlo Ventures)
LLMs turn enterprise UX from “learn my menus” to “tell me what you want,” reorganizing stacks into systems of record (data stores), systems of engagement (chat/voice front doors), and systems of work (agents that click the buttons for you). For vertical founders, this kills training tax and unlocks net-new users who won’t learn a new UI every quarter.
Implications for operators
Adoption > enablement: budget shifts from onboarding to outcomes; prove “time-to-first-action,” not feature tours.
Agent SLAs, not feature lists: commit to results (“lead entered + tasks scheduled in 30s”), instrument latencies and error rates.
Workflow is the moat: the same chat box everywhere—your edge is domain constraints (forms, compliance, approvals, vendor rails).
Data gravity matters more: the best agent is the one sitting on the cleanest vertical data model.
Pricing flips: charge for automated outcomes or usage bands, not seats; expand via new workflows, not new users.
Watchouts
Trust debt: hallucinations + silent failures erode confidence fast—keep humans at the edges and log every agent action.
Shadow process sprawl: one chat to rule them all can still create ten new back-end automations—govern them like APIs.
Vertical Spend Platforms (Extend Raises $20M via Axios)
Spend management is verticalizing. Brex and Ramp own the horizontal lane, but construction, healthcare, and franchise markets want specialized flows (supplier validation, compliance, and group purchasing power, etc).
Win condition: own the workflow, not just the interchange.
Playbook: embed spend tools inside broader vertical OS platforms where fintech becomes a true value-add and driver of attach.
The AI Gross-Margin Debate (via Janelle Tang @ BVP)
Early AI companies show 25–50 % gross margins because of inference and human-in-the-loop cost. VCs are betting those margins recover, or that market size outweighs it. Hyper-growth masks churn and CS cost, so many investors are underwriting blind.
Bottom line: price discipline still matters. As Luke noted, “Sierra raised at 225× revenue—great if retention’s real.” GRR remains the lifeblood of growth startups: what’s true for SaaS is equally true for “supernova” AI players.
Check out Janelle’s Substack here.
II) Vertical Titan: Jack Greco
(Co-Founder @ ACV Auctions)
The backstory
ACV Auctions—founded in Buffalo in 2014 by Dan Magnuszewski, Joe Neiman, and Jack Greco—built a digital marketplace that replaced traditional wholesale auto auctions. Backed by Bessemer Venture Partners, Tribeca Venture Partners, and Armory Square Ventures, it raised $350M before its successful IPO on NASDAQ at a ~$4B valuation.
Long dominated by players like Manheim and ADESA, auto auctions was never considered a “venture scale market”. ACV proved doubters wrong, connecting franchise dealers (offloading trade-ins) with used-car dealers (buyers) through mobile auctions that ran in minutes, not weeks. Its success came from standardizing vehicle data, guaranteeing inspections, and scaling city-by-city—an archetype of a vertical marketplace with a geographic lane strategy (not dissimilar from Uber).
Jack came to understand marketplaces at an early age through his father’s antique business. These days, he’s a prolific supporter of founders at Greco VC—he is an investor in dozens (if not hundreds) of startups, serves on several growth stage boards, and is an anchor of innovation in upstate NY.
Buy first, then build
ACV went buyer-first: solve demand by removing uncertainty. Inspectors spent ~20 minutes per car, produced standard condition reports, and ACV guaranteed them—if the car wasn’t as described, ACV ate the cost. Listing was free; fees applied only on sales. Trust was the growth engine.
Vertical of choice matters
Used cars are regional “snowflake assets.” Shipping costs (~$1 per mile) force local liquidity. ACV scaled city by city through territory managers closing ~2 dealers per month and becoming the first digital option in each market—those early territories remain ACV’s strongholds.
The hardest part
Recruiting and training inspectors to capture trustworthy data.
Re-educating dealers on pricing and risk.
Running field ops like a service business while maintaining software-grade UX.
Memorable lines from Jack
On scaling: “There’s always a hot wire… the ‘cheap drinks’ for your marketplace.”
On verticals: “Your first love is your true love — be first and you stay first.”
III) Vertical Playbook: AI Marketplaces
Why it works
Trust unblocks demand. Standardized truth + real guarantees shift risk off the buyer and collapse time to yes.
Local density > global theory. When moving/fulfillment is costly or regulated, liquidity is regional. Win watersheds, not the world.
AI compresses the messy middle. Use LLMs/agents for intake, triage, documentation, and CRM hygiene so humans focus on pricing, negotiation, and exception handling.
Short cycles compound. Run, fail fast, re-run in hours. Velocity becomes a moat.
Frictionless money. Simple terms and instant payouts create habit and switching cost.
How to run it (ACV-style blueprint)
Pick “snowflake” inventory. Assets with real variance (condition, compliance, provenance) where standardized truth is scarce.
Define the Canonical Condition Report (CCR). One schema everyone trusts: what’s right + what’s wrong, photos/video/audio, provenance, and known model-specific pitfalls.
Stand behind it—explicitly. Publish a guarantee policy (scope, claim window, remedy tree). Escrow a reserve and instrument claim SLAs.
Go buyer-first. Seed a hungry demand core (the “rabid 100”) before scaling supply. Make listing free; charge only on successful sales.
Enter by watershed. Map regions by real delivery radius. Staff territory managers with a simple cadence: close ~2 new supply partners/month; target ≥X CCRs/day.
Time the incumbents. Launch/visit just before competitor pickup days; re-run unsold inventory same day.
Operationalize pricing. Show sellers views, bids, and comps; negotiate reserves with data, not ego. Coach to market-clearing prices, not vanity.
Place AI where minutes hide.
Auto-draft CCR narratives from structured checks + media.
Flag likely issues by make/model (heuristics + history).
Summarize buyer questions; pre-answer where safe.
Push clean updates into your CRM with zero clicks.
Instrument trust. Track CCR error rate, claim rate, time-to-payout, % repeat buyers, seller NPS. Make “trust debt” visible on the same dashboard as GMV.
Keep humans at the edges. Escalations, price counsel, and deal disputes stay human by design; everything else is automated or batched.
Pay fast, keep contracts light. Speed is your reward loop—money and time close the adoption gap.
Market like a local. Show up physically early; be first and be loud. The first love tends to stick.
Founder litmus tests
Hot-wire: What’s your Day-0 spark (e.g., in-house supply, financed guarantees, priority payouts) that jump-starts one side without breaking unit economics?
Liability: Can you afford to be wrong—and have you modeled the reserve, fraud rate, and recovery mechanics?
Locality: Do shipping/compliance realities force regional networks (good), or does the asset commoditize globally (harder moat)?
AI fit: Exactly which steps become machine-first without eroding trust? Where do humans remain the UI?
Ops DNA: Are you willing to be a service business with tech on the outside until the flywheel spins?
What we debated on-air
UI shift: systems of record → engagement → work (agents that “just do it”).
Fintech wedge: vertical cards win only when they own the workflow, not just the swipe / interchange.
Margins vs. multiples: inference costs today vs. retention-anchored value tomorrow.
Hardware consensus: As a gate to software success when predicated on new device-centric ecosystems like AR or personal devices.
Retention > net new accounts: you have to grow your bucket fast if it’s leaky, and that problem only gets worse over time. Pay attention to GRR early and ensure your product team is on the hook for it.
Next week on VERTICALS
#5: Vertical Community Building with Todd Saunders, Founder / CEO of Broadlume.
Thanks to our partner
VERTICALS is made possible by Parafin: embedded finance & capital for your merchants. Building a vertical platform? Explore how financing can lift retention & revenue and check out Parafin today!
If you’re building in Vertical AI (or considering it) and want a sounding board, the Euclid Ventures team would love to hear from you. DM us here on Substack or ping us on LinkedIn.
