Leadbay’s AI Yields Tenfold Leads From 0.001% of Available Data

The Blind Spot in Sales Intelligence

Leadbay closed a $4.3 million seed round backed by Y Combinator from its F25 batch and launched a sales intelligence platform built natively on Anthropic's Model Context Protocol — a move that lets enterprise sales teams triple their addressable market and increase qualified leads tenfold by reaching the seven in ten U.S. companies traditional databases miss.

Rebel Ventures, Roosh Ventures, Inovexus Ventures, TS Ventures, Alumni Ventures, Bright Ventures, Transpose Platform, and Deel Ventures joined the round. Operators from Deel, Gusto, and Pennylane wrote angel checks. The investor mix signals more than capital: Bright Ventures, the venture arm of web data infrastructure provider Bright Data, backing a company that extracts intelligence from sparse signals suggests the data infrastructure community sees value in the inference approach. Deel Ventures connects Leadbay to a company that itself sells to SMBs globally. A Sorbonne University research partnership adds an academic dimension to the AI inference model that most sales intelligence startups lack.

Garry Tan, Y Combinator's CEO, personally encouraged the founders to expand into the U.S. after reviewing their growth trajectory in August 2025. "What you developed at Leadbay serves enterprise teams prospecting SMBs," Tan told them. "SMBs represent 40% of U.S. GDP, and nobody can help them. It's a blue ocean and a billion-dollar market."

Ludovic Granger and Milan Stankovic founded the company in 2024 after two years developing technology to qualify businesses that lack the digital signals traditional platforms rely on: plumbers, hairdressers, hotels, restaurants, construction firms, B2B service providers. Stankovic holds a PhD in AI focused on working with limited data. The company operates with eight employees from San Francisco. Their system uses a proprietary multi-modal inference model that analyzes weak and fragmented signals to infer what a business does, who it serves, and how it is growing. Sales teams interact with the platform by engaging with leads, which continuously improves the model's accuracy over time.

Leadbay says platforms like ZoomInfo, Apollo, LinkedIn Sales Navigator, and Clay miss that proportion and cover only one in 100,000 digital signals for these underserved verticals. The Leadbay MCP server is open source and lives at github.com/leadbay/mcp. It exposes lead discovery, qualification, and enrichment as tools that AI assistants can call directly — no manual handoffs between applications. The server ships a full UX/UI component library native to MCP, letting teams build workflows shaped to their context rather than adapting to a vendor's interface.

Early enterprise customers include L'Oréal, Saint-Gobain, Deel, Gerflor USA, Corgi, and Nespresso B2B. The fresh investment will enable the firm to scale its U.S. go-to-market team from San Francisco, advance its AI inference technology through the Sorbonne partnership, and expand engineering capabilities under Stankovic's leadership. Distribution efforts will continue in both France and the United States.

Why MCP Changes the Integration Math

Anthropic introduced the Model Context Protocol in November 2024 as an open standard that lets AI assistants connect securely to external tools and data. The protocol sits between the model and the execution layer: each AI application runs an MCP client, each service runs an MCP server, and the two speak a common language. In a traditional integration model, ten AI applications calling ten services require one hundred bespoke connections. With MCP, the same topology needs roughly twenty pieces — one client per application, one server per service — and a new application simply reuses the existing servers. The standard also wraps local resources (files, terminals, databases) that OpenAPI specifications never cover, and when an underlying API changes, the hosted MCP server absorbs the fix in one place instead of forcing every consumer to update.

Leadbay built its platform on this protocol from the start. The server is also open source and hosted there. It works with any assistant that supports MCP, including Claude Desktop, Claude Cowork, Claude Code, Cursor, and Codex, so a sales rep can pull leads, qualify them, draft outreach, and log activity in plain language without leaving the chat interface. The server automatically uses the user's active lens (the last filter configuration used in Leadbay) and exposes lens management through chat: list lenses, create new ones with sector and size criteria, or adjust an existing audience by name. Sector names resolve against a live taxonomy that the assistant can query.

Conventional lead-scoring tools such as LinkedIn Sales Navigator, ZoomInfo, Clay, and Apollo operate as standalone databases or enrichment layers. They deliver static lists that decay the moment they are exported. Leadbay inverts the model: "Inbox, not a database." Each day the user receives a fresh batch of leads paced by how many leads they have actually acted on recently. Pulling more does not produce more; acting on leads does. The system ships two scoring layers. Every lead carries a basic firmographic score that already correlates well with fit. Roughly the top ten leads in each batch receive an AI-qualified score derived from targeted web research against the user's qualification questions. Leads below the top ten are not worse; the system conserves research budget. The assistant can request deeper qualification or contact enrichment on any lead that looks promising.

This architecture enables a daily rhythm that conventional tools cannot replicate. The assistant works as a daily check-in: pull fresh leads, skim the auto-qualified top, deepen one to three promising candidates, propose outreach, then log what was sent via leadbay_report_outreach. If the host supports scheduling, the run can be automated. Early adopters report that one in five Leadbay customers now use the product exclusively through Claude and consume four to ten times more quota than they previously spent on software licenses. A forward-deployed engineer wires Leadbay to the customer's CRM, ERP, and data lakes, tunes the model on won and lost deals, and proves fivefold prospecting improvement on one team and one market.

The contrast sharpens in data-scarce industries such as construction, hospitality, and B2B services, where traditional databases have thin coverage. Leadbay users report three times more qualified leads than LinkedIn, ZoomInfo, Clay, or Apollo in those segments. The MCP layer makes that advantage portable: the same qualified leads flow into whatever AI assistant the team already uses, without a separate integration project for each destination. Competitors including Amplemarket and Clay have since added MCP servers, but Leadbay's server was designed from inception around the protocol's discovery and execution model, not retrofitted onto an existing API surface.

Enterprise Proof: Triple the Market, Tenfold the Leads

The numbers Leadbay publishes come from named enterprise accounts that had already exhausted conventional prospecting stacks. Teams at those companies have tripled their addressable market and increased qualified leads tenfold versus the tools they used before, the company says. The same customers report that more than half of closed deals now originate from companies that never appeared in their CRM or any third-party database.

"We found in 10 minutes on Leadbay what we couldn't find in 6 months with Clay and a GTM agency," one buyer told the founders. That line appears in multiple public sources — Y Combinator's company page, the Pulse2 funding write-up, and Inovexus's portfolio summary — suggesting it is a recurring refrain, not a cherry-picked testimonial. The repetition across channels signals that the time-to-value gap is the metric buyers care about most.

Those sectors share a structural problem: the businesses they sell to (plumbers, hairdressers, independent hotels, regional restaurants) generate almost no digital exhaust. No LinkedIn pages, no job postings, no funding announcements. Traditional intelligence platforms cover roughly one in 100,000 available digital signals and miss an estimated that proportion, per Leadbay's Y Combinator profile. The inference model the company spent two years building, Lucid-see-V1 (released with the 2.0 launch), analyzes fragmented, weak signals to infer what it does, who it serves, and whether it is growing. Prediction accuracy improved nearly half with that model, Inovexus says.

The forward-deployed engineering motion turns those raw predictions into pipeline fast. On day one, an engineer integrates Leadbay with the customer's systems, tunes the model on historical deals, and demonstrates a fivefold prospecting lift on a single team in a single market. That proof point — 5× prospecting efficiency — is cited on both the YC page and the company's own site. It also explains the 55 percent figure: when the model is trained on a specific sales motion, it surfaces buyers the team would never have thought to target.

Revenue traction backs the usage claims. Leadbay had signed half a million euros in ARR with a team of ten before the seed close, split between France and the United States. The Microsoft Marketplace listing, secured earlier in 2025, opens co-sell routes into enterprise accounts that already transact through Azure Marketplace — a distribution lever most seed-stage AI companies lack. The pattern across those enterprises is consistent: each sells into fragmented SMB markets where conventional data providers have no coverage. Leadbay does not replace LinkedIn or ZoomInfo for high-signal prospects; it unlocks the share those platforms miss. That is the addressable-market expansion. The tenfold qualified-lead increase follows because the scoring model is calibrated on each customer's actual close data, not a generic firmographic heuristic.

No independent third-party audit of these figures exists; they are vendor-reported. But the specificity of the customer list, the recurrence of the "10 minutes vs. 6 months" quote across unrelated publications, and the technical architecture (inference over sparse signals, forward-deployed tuning, MCP-native delivery) make the claims falsifiable in a way generic "AI-powered prospecting" marketing is not. If the 3×/10×/55% numbers were inflated, the next quarter's churn would expose it. The seed investors bet $4.3 million that they won't.

Building the Forward-Deployed Team

The $4.3 million seed round closed in May 2026 did not sit in a bank account. It went straight into headcount, specifically a new class of forward‑deployed engineers who sit inside customer accounts on day one. Leadbay's job postings describe the role bluntly: "On day one, a Forward‑Deployed Engineer integrates Leadbay with your CRM, ERP and data lakes, tunes the model on your won and lost deals, and proves 5× prospecting on a single team in a single market." That is not a support function. It is the delivery mechanism for the gains that those customers have reported.

The company lists three engineering openings on Y Combinator's board. Two are forward‑deployed roles: a U.S. position with a base range of $100K–$150K and a U.S. internship listed at $3K–$5K monthly. The third is a France‑based internship paying €1.5K–€2K monthly. All three carry the same title. The France role is not a token hire; it is the beachhead for the European expansion the founders have signaled since the round announcement. "Objectif : expansion sur le marché français et américain," the French press release stated. Pulse2 and Parsers.vc both confirmed Leadbay will continue expanding distribution efforts there as well.

"Right now that someone is a founder. That is the bottleneck, and it is the entire reason this job exists."

That line appears verbatim in the Y Combinator job description for the France internship. Ludovic Granger and Milan Stankovic, the two founders, have been personally integrating Leadbay into enterprise stacks since the F25 batch. The 3× and 10× numbers are real, the posting adds, "because someone sat with each of those teams and built their specific workflow. Not a template. Theirs." The founders cannot scale that motion themselves. Hence the hiring push.

The forward‑deployed model reflects a deliberate architectural choice. Leadbay's MCP‑native platform includes a complete UX/UI component library that runs inside Claude, ChatGPT, and Copilot. But the platform only produces the 5× prospecting proof point when the model is tuned on a specific company's won and lost deals, data that lives in CRM, ERP, and data lakes behind firewalls. A traditional sales engineer hands off a demo. A forward‑deployed engineer writes the integration, trains the model, and stays until the metric moves. The job postings call for Kotlin, Python, React, PyTorch, LLMs, Postgres, and MongoDB fluency (full‑stack plus ML) because the role spans crawling, enrichment, qualification, and the MCP server surface.

Europe is not an afterthought. The product page lists "GDPR‑compliant. Dedicated DPO and European data compliance" as a core feature, not a compliance checkbox. Sonepar, Lyreco, and Deel, all European‑headquartered, already appear on the customer logo wall alongside the French multinationals. The France internship is the first paid role on the continent; the founding team is French (Granger and Stankovic met at Station F), and the San Francisco office exists to serve U.S. expansion, not the other way around. The company's own announcement reads "Launch Leadbay USA, San Francisco," not "Launch Leadbay, San Francisco."

The hiring plan also reveals the competitive logic. Generic prospecting tools, the major platforms, hand every team the same workflow. "Generic tools hand everyone the same generic workflow, which is exactly why every team using them plateaus in the same place," the U.S. internship posting argues. Leadbay's bet is that the plateau breaks only when the tool adapts to each team's qualification logic. That adaptation requires an engineer in the room. Three roles now. More when the 5× proof point repeats across the next ten accounts.

The MCP Ecosystem Shifts Underfoot

Anthropic launched the protocol in November 2024 to solve what it called the N-times-M connector mess: every AI application needing a custom integration for every data source. By early 2026 the company reported over ten thousand active public MCP servers and nearly 100 million monthly SDK downloads across Python and TypeScript. Salesforce, HubSpot, Notion, GitHub, Slack, and Google Drive have official or community-built servers. Block and Apollo have integrated MCP into their systems. Replit, Codeium, and Sourcegraph are adding support. The protocol has moved from proposal to infrastructure in roughly 15 months.

Sales-focused platforms have reacted at different speeds. Amplemarket and Clay each published MCP implementations during 2025. Amplemarket scored 10 sales MCP servers across five capabilities, including finding prospects, enriching and researching them, building sequences, enrolling them into outreach, and cross-client compatibility with Claude and ChatGPT, on a documented 0-to-3 scale. The volume of third-party evaluations signals that buyers are already comparison-shopping.

Microsoft entered with a first-party play. Its D365 Sales Model Context Protocol lets AI analyze lead data, engagement history, and contextual signals inside Dynamics 365 Sales, surfacing a prioritized list without guesswork or delay. Microsoft also published an MCP lab for lead qualification that enables AI to securely access tools, data, and prompts through the standard protocol. The move matters because Dynamics 365 sits in the CRM layer where lead-scoring decisions actually happen, closer to the revenue event than top-of-funnel prospecting tools.

OpenAI chose a different path. Rather than open-sourcing a protocol, it brought a data-connecting feature called Work with Apps to ChatGPT in late 2024, initially letting the chatbot read code in developer-focused coding apps. The company said it plans to extend the capability to other app categories but will pursue implementations with close partners instead of publishing an open standard. Google followed with the Agent-to-Agent Protocol (A2A), an open standard designed for seamless communication between diverse AI agents. Google explicitly positioned A2A as complementary to MCP: MCP handles the vertical connection from AI agent to business tool, while A2A handles the horizontal connection from AI agent to AI agent. As of Q1 2026, four inter-agent protocols have reached meaningful adoption: MCP, A2A, ACP (Agent Communication Protocol), and UCP (Universal Communication Protocol). Some observers warn that the delineation between MCP and A2A may blur in practice, potentially forcing developers to implement both.

Analyst forecasts underscore the pace. Gartner found that two-thirds of enterprise technology leaders cite integration complexity as the top barrier to deploying agentic AI. The same firm predicts that three-quarters of API gateway vendors and half of iPaaS vendors will have MCP features integrated by the end of 2026. Gartner also forecasts that two-fifths of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5% in 2025. A separate projection puts nine in ten B2B buying as AI agent-intermediated by 2028, driving over $15 trillion of B2B spend through AI agent exchanges. Early adopters report that MCP-based integrations reduce time-to-integration from months to weeks and cut development costs by up to 70%.

For sales AI vendors, the implication is clear: MCP support is shifting from differentiator to baseline expectation. When a company evaluates prospecting or CRM tools, the question is moving from "does it integrate with our CRM?" to "can our AI agents interact with it natively?" Leadbay's MCP-native architecture — built on the protocol from day one rather than retrofitted — positions it ahead of incumbents that must now bolt on compatibility. The founders' next hire will ship from a Paris desk before the window closes.

Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe, and the people building the field.