Magentic raises $18M to expand AI agents for industrial procurement

Why it matters

Magentic is targeting one of enterprise AI's clearest business cases: finding and recovering procurement leakage. The harder task is earning permission for agents to act inside contracts, orders and invoices.

Robin Van Aeken and Odhran O'Donoghue announced an $18M Series A for Magentic on September 17th, giving the Oxford founders fresh capital to put AI agents inside the procurement systems of large manufacturers.

Felicis led the round, with existing investors Sequoia Capital and The Westly Group participating, according to Magentic's funding announcement. The financing follows a $5.5M seed announced in July 2025, bringing Magentic's publicly announced funding to at least $23.5M. Magentic did not attach a valuation to the Series A.

The founders bring an unusually tidy division of experience to the problem. Van Aeken studied economics and management at Oxford before working at McKinsey, where he advised manufacturers and supply-chain leaders. O'Donoghue completed an Oxford DPhil focused on machine-learning methods for electronic health records and later worked on AI research at OpenAI.

They met at an AI and Climate Impact Challenge talent-matching session, according to Ana Bakshi of Oxford. That pairing shaped Magentic's founding thesis: valuable procurement decisions remain scattered across contracts, enterprise resource planning systems, emails, invoices and spreadsheets, leaving conventional workflow software with an incomplete picture of what a manufacturer bought, what it should have paid and what a supplier still owes.

One year from launch to Series A

Magentic launched publicly in July 2025 with its agents, then called "Mages," pitched as digital teammates for procurement and supply-chain departments. Its $5.5M seed came from Sequoia, First Momentum Ventures and The Westly Group.

At that launch, Magentic described a "pay-per-cure" model that charged customers based on the value recovered rather than software seats. That positioning put a measurable financial result at the center of the sale: missed rebates found, overpayments recovered, contract terms enforced or purchases shifted to a less expensive supplier. Magentic's current announcement does not specify whether that pricing structure applies across its expanded product line.

Van Aeken's pitch extends beyond cost-cutting. "The companies that build the best intelligence into every decision they make will be the ones that compound their competitive advantage," he said in the Series A announcement.

The new funding is earmarked for additional procurement and supply-chain workflows and for research into agents that can plan and execute longer jobs across large volumes of mixed enterprise data. Magentic operates from London and New York and names Siemens as a customer reference on its website.

Procurement leakage is the wedge

Magentic's product covers three points in the purchasing cycle. Before a purchase, its agents analyze spend, compare equivalent materials and consolidate demand across sites. During purchasing, they check requisitions against approved suppliers, prices and compliance rules. Afterward, they examine orders, contracts and invoices for overpayments, missed rebates and unfulfilled supplier obligations.

That workflow reaches deeper into company systems than a chatbot answering questions over procurement data. Magentic says its agents can draft supplier messages, route purchases, negotiate contracts, run orders and clear invoices. Its security documentation says humans are brought into key review points and receive evidence trails for important actions.

The distinction matters because procurement errors move money and affect production. An inaccurate summary is inconvenient. An agent that approves the wrong supplier, misreads a contract or delays a critical material can interrupt a factory.

Magentic says one customer processes 1.2M orders through its digital workers each year, while another has identified $4M in savings. Magentic also claims typical savings of 2% to 5% and an average 60% improvement in data quality. Those figures come from Magentic and are not accompanied by customer-level methodologies or independent audits. Revenue, customer count and retention were not included in the announcement.

The claims still show why procurement has become a crowded target for vertical AI companies. Savings can be expressed in dollars, implementations can be attached to existing enterprise systems and the customer already has a budget tied to the work.

A well-funded field of virtual buyers

Magentic is arriving alongside several startups with similar ambitions. Didero announced a $30M Series A on February 12th for agents handling supplier communication, order tracking and exceptions. Lio announced its own $30M Series A on March 5th for a virtual procurement department that processes requests, compares suppliers, negotiates and executes purchases.

Magentic is anchoring its approach in large manufacturers, including direct spending on raw materials as well as indirect categories such as equipment and services. It is also pushing value recovery after a purchase, where fragmented contracts and transaction records can conceal pricing deviations or supplier obligations.

Felicis partner Feyza Haskaraman framed the investment around the difficulty of getting agents to understand and act inside manufacturers' existing systems. That is the technical and commercial hurdle Magentic must clear. Manufacturers have spent decades accumulating ERP customizations, local buying processes and supplier data that rarely line up cleanly across business units.

The Series A gives Van Aeken and O'Donoghue room to expand from individual procurement jobs toward the broader workforce they describe. Their next test is turning company-produced savings figures into repeatable economics while increasing the work their agents can complete without weakening the human controls enterprise customers require.