Fragment Raises $56m for Procurement AI Agents

Fragment, a San Francisco start-up building AI agents for corporate purchasing, has launched with $56m in total funding, including a Series A led by 8VC and a seed round co-led by General Catalyst and A*. Its software finds and fixes the mismatched orders, partial deliveries and invoice errors that procurement staff at large manufacturers now sort out by hand.

The company announced the round in a release, on its blog and in a post by co-founder and chief executive Pranav Mulgund. It names Graybar, an electrical and communications distributor, and CNX as customers.

An 8VC-led Series A and a list of angels

Fragment did not say how the $56m splits between the seed and Series A rounds, or give a valuation. Its angel investors include Roelof Botha, Nikesh Arora, Jeff Wilke, Eric Wu and Dennis Cinelli.

“Most software companies earn their way into the enterprise over time. Fragment started there,” said Jack Moshkovich, a partner at 8VC. He said building from the start for companies with “thousands of dependencies and immense complexity” sets a higher bar from the outset.

The company says the money will pay for further development of its core product, expansion “across source-to-pay”, meaning the full cycle from finding suppliers to paying them, and deployments at more large companies.

An engine that acts rather than flags

Fragment’s product is what it calls the Autonomous Context Engine, or ACE. It links data from a company’s existing systems and records the policies, supplier relationships and unwritten rules that experienced staff rely on. Fragment’s agents then use that picture to investigate an exception, decide what to do and “carry out authorized resolutions end-to-end, rather than flagging issues or recommending next steps”.

It works with data where it already sits, so customers can skip moving records into a new system. Mulgund’s blog names SAP, Ariba and Coupa as the kind of software companies have spent billions on, while people still step in when something goes wrong.

The scale of that work is the pitch. Graybar processes roughly 3m invoices a year, and 80% of them need some form of resolution, according to Fragment. “That collaborative approach helped us tackle some deeply entrenched challenges,” said Paul Ferguson, Graybar’s vice-president of global finance shared services. Neither company said how many of those invoices Fragment now handles.

“We deliberately started with direct procurement at large enterprises with the most complex supply chains because solving that challenge could fundamentally change what enterprise software can do,” Mulgund said. Direct procurement is the buying of materials and parts that go into a manufacturer’s products.

Fragment also prices differently from most business software. Rather than charging per user seat, its model “is built around work completed by its agents”.

Paying for work done, not seats

Charging for completed work is a bet that AI agents will cut the number of people using procurement software, which would erode seat-based revenue. It also ties Fragment’s income to how much its agents actually resolve, which gives buyers a clear test. The same shift is visible elsewhere: Salesforce has added spending caps to its agent platform, a sign that buyers want limits on what agents cost.

The harder question is trust. Fixing an invoice mismatch can mean changing an order or approving a payment, and “authorized” carries a lot of weight in Fragment’s description. Payment networks are already testing how far to let agents go, as in the Visa, BILL and Billtrust pilot where agents pay only invoices that people have approved.

Investors are also putting money into US manufacturing, from Re:Build’s $2.1bn round to robotics. Fragment targets the paperwork around that industry rather than the factory floor. The figure to watch is the share of exceptions its agents close without a person stepping in, which Fragment has not yet published.