Supply Chain / Procurement (AI) · Deep dive

Lio

The Munich-based YC S23 company (formerly askLio) selling agentic AI as a virtual procurement workforce — closed a $30M Series A led by a16z on March 5, 2026 with SV Angel, Harry Stebbings and YC participating, bringing total funding to $33M, on the back of Munich Re, Brose and Novozymes as reference customers and a claim of 100+ Fortune 500/Global 2000 enterprises managing billions of dollars in spend through Lio's agents.

emerging

The question that decides it: Does an independent multi-agent orchestration layer sitting on top of ERPs — sold to procurement as a workforce, not a tool — retain enough of the workflow to defend a $30M+ round once SAP Ariba's native agent framework, Coupa's Thoma Bravo–funded AI, and Zip's $371M orchestration platform ship agents that are 'good enough' inside the systems where PO, contract and supplier-master data already live?

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Lio is the three-year-old Munich AI-procurement company that graduated Y Combinator’s S23 batch as askLio, rebranded to Lio, and closed a $30M Series A led by Andreessen Horowitz on March 5, 2026 with participation from SV Angel, Harry Stebbings/20VC and Y Combinator. Total funding through the round sits at $33M per PR Newswire; the a16z announcement claims 100+ enterprise clients including Munich Re, Brose and Novozymes managing billions of dollars in spend through the platform. The product is pitched not as another procurement tool but as a virtual agentic workforce — specialised AI agents that triage requests, source vendors, negotiate, onboard suppliers, cut POs and match invoices across a customer’s existing ERPs, inboxes and contract stores. The Series A funds US expansion and product development. Headcount is roughly 80 in Munich; the round rounds a company that has moved fast on a category that every major procurement incumbent is now attacking simultaneously.

Founding story

Lio was founded in 2023 by Vlad Keil (CEO), Lukas Heinzmann (CTO) and Till Wagner (COO), all based in Munich. Keil is a serial German founder — profitable SaaS business by age 22, then Gabriel AI in 2021, an anti-telephone-fraud startup that ended up working with German federal law enforcement. He spent time consulting enterprises on procurement operations before Lio; the founding observation, cited in a16z’s own March 2026 investment note, is that Fortune 500 procurement teams still transact trillions of dollars of enterprise spend one email and one PDF at a time. Heinzmann is a TUM MSc AI engineer and Keil’s Gabriel co-founder, who cut his teeth at an AI startup Audi acquired and rose to technical lead for the myAudi mobile app before spinning out Lio. Wagner runs commercial and operational strategy. The three joined YC in the Summer 2023 batch under the askLio name, launched initially as an “AI-Copilot for procurement teams”, and pivoted the framing from copilot to autonomous multi-agent workforce as the underlying model capability improved through 2024–2025. The rebrand from askLio to Lio, and the transition to selling a “virtual procurement department” rather than a copilot, is the story of the last 18 months.

How it works

A purchase request lands in Lio — typed into a chat interface, submitted through an intake form, forwarded from an email inbox, or triggered by an ERP event. From that moment a set of specialised agents operate in parallel on the request rather than sequentially. A triage agent classifies the request (indirect, direct, services, IT, marketing), pulls policy and threshold data from the customer’s internal knowledge base, and decides which subsequent agents to activate. A sourcing agent researches vendors across the customer’s preferred-supplier list, the open web and public catalogues; a negotiation agent runs quote comparison, back-and-forth over price, terms and delivery via email; a compliance agent runs risk, credit, ESG and sanctions checks against internal and third-party data; an onboarding agent creates the supplier record in the ERP or MDM; and a purchasing agent commits the PO, tracks fulfilment, and matches the eventual invoice against the PO and receipt. The layer that binds these agents together is what Lio calls Agent Operating Procedures (AOPs) — customer-specific execution templates that encode a given enterprise’s SOP so the agents behave the same way a trained buyer or shared-service-centre analyst would. All of this sits on top of the customer’s ERP (SAP, Oracle, Workday), P2P suite (Coupa, Ariba, Ivalua), inbox, contract repository and web browser rather than replacing any of them.

Product and business overview

Lio ships one product — the agent workforce — with three deployment shapes. First, an agentic intake-to-pay layer that replaces or augments an internal shared-service centre for indirect spend, positioned against Zip and ORO Labs. Second, a direct-procurement companion for manufacturers and materials-heavy businesses (the Brose relationship is the reference), positioned against ERPs that under-serve direct sourcing. Third, an integration-and-onboarding layer sold to enterprises that already own Coupa or Ariba and want to automate the long tail of email-driven supplier interactions that never make it into the P2P system. The a16z blog post frames the offering as “replacing outsourced labour with AI” and cites an 85% reduction in manual buyer work.

Business model and pricing

Lio does not publish pricing. Public disclosures suggest an enterprise annual-subscription contract structured around either seats, agent hours, spend under management or requests processed — the standard enterprise-AI mix. The 85% manual-work-reduction claim in Lio’s press materials and a16z’s investment note is the value story that supports the pricing. There is no public suggestion of a take-rate on the managed spend itself, which would push Lio into the Vendr/Tropic negotiation-as-a-service model; the language (“workforce”, “replaces outsourced labour”) points to a per-request or per-seat book. The BPO-replacement comparison is the pricing anchor Lio wants buyers to reach for — a genuine buyer analyst in Europe costs €70–120K loaded, an outsourced Manila or Bengaluru procurement analyst €25–40K, and Lio’s implicit pitch is that a single agent seat with unlimited-throughput agents undercuts both. No published rate card, no self-serve tier, no marketplace listing beyond an AWS Marketplace stub.

Traction over time

• Jul 2023: askLio launches out of Y Combinator S23 as “AI-Copilot for procurement teams” per YC’s launch post.
• 2023–2024: undisclosed pre-seed/seed extensions from SV Angel, Harry Stebbings and angels; Y Combinator standard deal in seed.
• 2024: initial enterprise pilots close in Germany; Munich Re named as an early reference.
• Late 2024–2025: rebrand from askLio to Lio; positioning shifts from copilot to multi-agent workforce; Brose, Novozymes and a dozen additional Fortune 500/Global 2000 customers named.
• Mar 5, 2026: $30M Series A led by a16z announced (PR Newswire, TechCrunch, a16z blog).
• Mar 2026: US expansion office announced; headcount ~80.

Market analysis

Procure-to-pay software is a $10.3B market in 2026 growing at ~9.3% CAGR to $16B by 2031 per Mordor Intelligence. The broader procurement-software category (including sourcing, contract management, supplier lifecycle) is closer to $23B by 2035 per Precedence. That is the ceiling. The floor is set by three structural forces: (1) most large enterprises already own an ERP and a P2P suite, so agent-native entrants sell into an environment where the incumbent already has the data; (2) generative AI has compressed the cost of building agentic workflows to the point that every P2P suite is now shipping its own; and (3) the CIO buyer is increasingly focused on data sovereignty — 78% of CIOs surveyed by Futurum name data control as the primary barrier to scaling procurement agents, and that pressure favours agent frameworks that run inside the incumbent’s data boundary.

Competitive intel

Zip ($371M raised, $2.2B valuation) is the well-funded orchestration incumbent; wins on breadth of intake-to-pay coverage and Gartner Visionary positioning. ORO Labs ($160M, Goldman/Brighton Park) attacks with a similar agent-first pitch and US enterprise incumbency. Levelpath ($100M, Battery/Redpoint) benefits from the Scout RFP founders’ distribution. SAP Ariba’s next-gen agent layer and Coupa’s Thoma Bravo–funded AI push are the bundled-incumbent risk — they sit on the ERP data that Lio’s agents have to reach into. Vendr, Tropic and Sastrify own the SaaS-spend niche and do not touch direct-material spend. Pactum embeds directly with SAP for autonomous negotiations. Aerchain (Bengaluru) attacks from lower cost. Ramp and Rippling-owned Airbase apply the same bundling threat from the SMB and mid-market end. In this set Lio’s specific edge is a European reference-customer base — Munich Re, Brose, Novozymes are three of the more procurement-mature buyers in Europe — plus a technical claim (AOPs as a codified execution grammar) that has yet to be independently benchmarked.

History and evolution

The company’s history is short and legible. Founded and incorporated 2023 in Munich as askLio. YC S23 launch as an AI copilot for procurement teams; rebrand to Lio during 2025 as the product moved from copilot to autonomous multi-agent workforce. Series A close March 5, 2026. The single most consequential evolution is the positioning shift: an “AI copilot for procurement teams” is a productivity tool sold to procurement seat-owners; an “AI procurement workforce” is a labour replacement sold to the CFO or CPO. The economics — and the price point — of the second are an order of magnitude larger. It is the reframing that made the a16z round possible.

What people say

The case for. a16z’s public investment note (Seema Amble, James da Costa, Eric Zhou, Brian Roberts, March 5, 2026) frames the company as agent-native, sitting on top of the existing procurement stack rather than replacing it, and cites 85% manual-work reduction and 100+ enterprise customers. Procurement Magazine covered Lio’s US Summit workshop in 2026 positively, framing the agents as freeing up procurement teams from tactical work. HELIAD’s investor commentary highlights Munich Re, Brose and Novozymes as blue-chip European references. The YC S23 launch post itself drew 130+ comments on the LinkedIn cross-post — an unusual amount of procurement-community interest for an AI copilot pitch in mid-2023.

The complaints. Independent user reviews on G2, Capterra and Reddit are essentially absent — a red flag not because the product is bad but because it means the customer base is still too small and too enterprise-locked to generate public feedback. The wider procurement-agent category has known complaints that Lio inherits: Futurum’s data show 78% of CIOs cite data sovereignty as a barrier to scaling agents on multi-tenant cloud infrastructure; Evolinq and SpecLens both point out that agent platforms struggle with the long tail of smaller regional suppliers who communicate by email in local languages. Founder-market fit is a fair critique: Keil and Heinzmann came out of consumer AI (Gabriel) and consulting, not procurement operations at scale — an SAP Ariba or Coupa PM has a five-year head start on the domain even if Lio has a three-year head start on agent design. Byrne Hobart’s newsletter has not covered Lio directly but has repeatedly argued that agentic-workflow startups selling to enterprises where the ERP incumbent owns the data face a compressed-margin future.

Outlook: the open question

For Lio to justify the a16z round, three things have to be true. First, the agent-orchestration layer has to matter more than the underlying model or the underlying ERP — Lio’s AOPs have to constitute a real, hard-to-copy execution grammar rather than a wrapper around any capable frontier model. Second, the “workforce, not tool” pricing has to hold: enterprise CFOs must accept a $500K–$5M annual contract for AI agents at a price point that displaces a BPO relationship, not just a seat licence. Third, and hardest, Lio has to reach $10M+ ARR and a US enterprise reference base before SAP Ariba’s native agents and Coupa’s AI push become the default answer inside the systems where the data already lives. If Ariba and Coupa ship “good-enough” agents inside the P2P suite in 2026–2027, the burden of proof on any independent orchestration layer becomes brutal.

How to attack it

An attacker does not go head-on with a broader multi-agent framework. Lio is well-positioned on the agent-orchestration frame; the wedge is a level below or a level above.

One level below: build an agent-development platform for procurement teams that lets an in-house team author its own AOPs on top of any frontier model, sold as infrastructure at a fraction of Lio’s contract price. This is the LangChain-of-procurement play. It exploits Lio’s biggest exposed flank — the AOP framework is proprietary, model-agnostic in claim but Lio-controlled in practice, and enterprise CIOs increasingly want frameworks they can audit and swap. Futurum’s finding that 78% of CIOs cite data sovereignty as the number-one barrier to scaling procurement agents is the source. Lio cannot easily commoditise its own framework without cannibalising the $30M round’s implicit thesis.

One level above: pick a single procurement vertical where Lio’s horizontal “agent workforce” pitch loses. Direct materials for automotive, contract manufacturing for pharma, MRO for oil-and-gas, or public-sector tender compliance in the EU are all categories where the buyer has domain-specific data models, regulator constraints and negotiation grammars that a horizontal agent framework will fail on. A vertical agent business (analogous to Aerchain in India or Pactum in negotiation) can beat Lio on demonstrable ROI in one category by owning the domain data and specialised playbooks, and then expand outward.

Exploitable weaknesses: founder-market fit gap versus a decade-long SAP Ariba PM; European geography limiting US Fortune 100 direct access despite the March 2026 expansion; near-total dependence on the reliability of underlying frontier model providers (OpenAI, Anthropic, Google) whose margin the agent layer must extract; and complete absence of independent G2/Capterra review evidence — a proxy for how enterprise-locked and reference-fragile the customer base still is.

Adjacent-segment play

The same agent-orchestration capability could be repackaged into a sourcing and vendor management layer for mid-market companies buying through Amazon Business, Alibaba, Faire, or SAP Business Network, sold as a lightweight buyer-side agent rather than a full procurement workforce. The customer is a 200–1,000 person company without an ERP or a procurement team, the price point is $30–100K/year, and the value is compressing the buyer’s own time on RFPs and vendor negotiation. Ramp and Rippling would be the ultimate ceiling on that motion — both bundle spend management into a broader finance platform — but the space between YC-style startups running on Ramp and mid-market ERP customers is genuinely unserved.

The other clear adjacent is a compliance and audit-agent business for regulated industries — banking, insurance, pharma — where the same orchestration primitives (parallel specialised agents, AOPs, ERP integration) can be repurposed to run continuous vendor risk, third-party compliance, and Know Your Supplier checks. The buyer is second-line-of-defence risk teams; the price point sits alongside a GRC platform (Archer, ServiceNow IRM), which is $500K–$3M/year at enterprise scale. Munich Re as a reference customer maps naturally into this adjacent — reinsurers already run continuous vendor and cedent monitoring at scale.

The wedge does not generalise into consumer or transaction-payments space; the agent framework is intrinsically B2B and workflow-heavy, and there is no consumer analogue where a multi-agent buyer workforce clears a real user need.

Sources and further reading

• Lio Raises $30M Series A to Bring Agentic AI to Enterprise Procurement — PR Newswire, March 5, 2026
• Investing in Lio — Andreessen Horowitz (Seema Amble, James da Costa, Eric Zhou, Brian Roberts), March 5, 2026
• Lio raises $30M from Andreessen Horowitz and others to automate enterprise procurement — TechCrunch, March 5, 2026
• Lio (formerly askLio): The World’s First Multi-Agent System for Procurement — Y Combinator company page, S23
• Lio raises €25.7 million — Munich Startup, March 2026
• Lio Technologies Raises $30m Series A to Bring Agentic AI to Enterprise Procurement — HELIAD investor highlight, March 2026
• How Lio’s AI Agents are Freeing up Procurement Teams — Procurement Magazine, 2026
• Zip Secures $190 Million Series D at $2.2B valuation — BusinessWire, October 21, 2024
• ORO Labs Raises $100M for Agentic Procurement Orchestration — ORO Labs press, 2024
• Levelpath Raises $55M+ Series B led by Battery Ventures — BusinessWire, June 27, 2025
• Is SAP’s Next-Gen Ariba the Missing Link for Real Agentic Procurement Workflows? — Futurum Group, 2026
• Procure To Pay Software Market Size, Trends — Mordor Intelligence, 2026