End Close's $3.5m seed round backs rules-first payment reconciliation
Irish-founded End Close raised $3.5m in seed funding to grow its payment reconciliation platform, keeping rules-based matching at the core and AI agents on top.

Reconciliation is one of those back-office jobs that rarely gets attention until it breaks. For high-volume payments companies, it means matching every transaction against what the banks, processors, card networks and ledgers say happened, and then explaining the gaps. End Close, an Irish-founded start-up, has just closed a $3.5m seed round to build out its answer to that problem, and its pitch is a notable one: the matching itself should be deterministic and rules-based, with AI handling the messy exceptions around it.


From left: Sean Bolton and David Newell. Image: End Close

Who is backing it
The round was led in terms of size by Foundation Capital, which is End Close's largest institutional investor. Y Combinator, Liquid 2 Ventures, Sunflower Capital and Angel Invest also took part, alongside a group of experienced angel investors. That's a fairly typical syndicate for a seed-stage B2B infrastructure company, and the Y Combinator connection stands out because End Close was part of its winter 2026 batch.

The company was founded in early 2026 and is headquartered in San Francisco, US. It describes itself as a payment reconciliation platform aimed at high-volume payments companies, which narrows the target market considerably. Smaller merchants usually don't generate the transaction volumes or exception queues that make a dedicated reconciliation tool worth the switch. Companies processing large numbers of payments across several banks and processors do.

The founders' backgrounds
CEO Sean Bolton spent six years at Modern Treasury, where he was tech lead of payments engineering before building and running its reconciliation division. That's direct operating experience with the exact problem End Close is tackling, and it's the sort of background that tends to matter when a buyer's finance team asks how a tool handles edge cases at scale.

CTO David Newell has a different profile. He was a team lead at PostHog for its error tracking product and founded Unflow. He has also worked as a product engineer at Intercom and Basecamp. Taken together, the pairing is payments domain depth on one side and product engineering on the other, which is a reasonable mix for a company trying to sell infrastructure software to finance and operations teams.

What the money is for
End Close plans to put the funding into growing its engineering, go-to-market and implementation teams. The implementation side deserves more attention than it usually gets. Reconciliation tools only work once they've been connected to a customer's specific stack of banks, processors, card networks, enterprise resource planning systems and data warehouses. Hiring for implementation means the company is betting that onboarding capacity, not just the software, will limit how quickly it can grow.

The company is currently recruiting for engineering and enterprise sales roles. For a seed-stage firm, that combination suggests it's already seeing demand from larger accounts and wants sales coverage in place before it's overwhelmed.
The product split: rules for matching, agents for exceptions
The most interesting part of End Close's positioning is how it divides the work. In its own words, the platform is built around the small share of transactions that generate most of the effort. It reconciles continuously, transaction by transaction, across the systems a customer already uses. The matching engine is deterministic and rules-based, so every match can be explained and audited.
AI agents sit on top of that engine. They investigate exceptions, resolve the routine ones automatically and pass the remainder to the finance team with a suggested fix. The company's argument is blunt: matching millions of transactions is a job for rules, not model reasoning, and auditors need a trail they can follow.
My read is that this is the right instinct for this category. A language model that occasionally produces a plausible but wrong match is a poor fit for ledger work, where the answer has to be reproducible on demand. Letting the deterministic layer handle the bulk and the agent handle the leftovers keeps the audit trail intact while still cutting the manual queue. Whether that holds up in practice will depend on how well the exception handling performs on the ugly cases, such as partial payments, currency differences and timing mismatches between systems.
What's coming next
Within the coming months, End Close intends to release two new elements. The first is RecBench, a benchmark for how well AI models handle reconciliation tasks. A public benchmark in this niche is useful well beyond End Close's own sales pitch, because there's little standard measurement of how models perform on matching and exception work. The second is a standalone version of the company's agent, which teams could use to automate simpler reconciliation work without adopting the full platform.
That second release is a telling move. It suggests End Close sees a route where finance teams try the agent first, which lowers the switching cost, and then extend into the deterministic engine as volumes grow. Teams evaluating reconciliation tools should watch whether the standalone agent works with the matching engine's audit trail or sits outside it, since that determines how much of the audit story carries over.
For IT and finance operations leads, the practical takeaway is to treat End Close as an early-stage vendor with a clear architectural stance. The rules-first design answers a real audit concern, but the sales and implementation build-out is still underway. If you're assessing tools in this space, ask how exceptions are logged, how a match can be traced back to its rule, and what the integration timeline looks like for your specific banks and processors.