Why we started End Close

Imagine there was a process in your business that absorbed about 10% of your team at different times during the day. The work you need to do to complete it is repetitive, low-value, and let’s face it…quite boring. There is software on the market to help with this work but it’s pretty basic and your head of engineering says you’d be better off building an in-house solution. The consensus is this work is so core to your business it shouldn’t be outsourced or handled by third party software.

From edge case to the norm

I lived that scenario for six years at Modern Treasury, where I joined as the sixth employee in 2019. Back then, a company moving that many payments across that many rails was an edge case. Today it’s increasingly the norm, and recon has become a far thornier problem for far more companies.

The world has moved to real time payment rails (RTP, FedNow, etc) which means money moves 24/7, 365. End user expectations of financial services are increasing e.g. buy now, pay later has turned a single checkout into four payments on four dates, each of which can fail, be refunded or be disputed separately. Returns and disputes are ballooning but they trickle in on unpredictable timelines. Data quality varies wildly across banks and processors.

In parallel with those macro changes, more companies are becoming payment companies. There’s a growing cohort of companies where money movement is the product rather than the plumbing behind it: cross-border remittance and global payroll apps getting money to workers in another country, marketplaces settling with thousands of sellers, processors running tens of millions of ACH transactions a year. For these businesses the core promise is that the money arrives, correctly, when it’s meant to, which turns recon into a product problem rather than a back-office one.

Recon is a data problem, not a finance problem

Having built reconciliation for payments at enormous scale, I realised the same approach would be invaluable to any finance team managing high volumes across multiple rails. While there are existing reconciliation solutions on the market they are focused on matching transactions, which is largely a solved problem.

In many organizations, recon is viewed in a very limited way. Engineering or product manage the pipelines and the data is thrown over the fence to the finance or operations team to reconcile everything. That makes it a very hard problem because you’re guessing with limited information.

Finance doesn’t understand how much of this could be automated, so they don’t ask for the data that would make it possible. Engineering doesn’t know where the data lands or what it’s used for, so they don’t know what’s worth passing on. The most useful information gets dropped on the floor and finance is left to match on amounts, dates and whatever the bank might have decided to put in a description field.

It doesn’t have to work that way. Recon works best when you treat it as a single problem that runs end to end, from the moment a payment is created to the moment it’s reflected in your ledger. How you send a payment shapes how easily it can be reconciled, so the data you capture at origination is the most valuable signal you have. Most companies that move money at scale are sitting on that knowledge but not using it to their advantage.

Thinking of it this way, it becomes apparent that recon is a data challenge, which is why I recruited David as my co-founder. A talented engineer I’ve known since college, he was the engineering team lead at PostHog for their error tracking product. While in some ways recon is analytics and error tracking for payments, more importantly he has experience wrangling the kind of high volume data pipelines that underpin End Close.

What we’re building

Our vision for End Close is a reconciliation system that’s flexible, robust and built for scale, and that does the work after the match.

Flexible. End Close connects to the sources you already use (banks, payment processors, data warehouses and your ERP) through APIs and managed connectors. When a bank changes its file format or you add a new rail, it adapts rather than breaks.

Robust. Missing and duplicate payments are flagged the moment they appear, not at month end, and every match and every action is logged in a full audit trail.

Built for scale. End Close reconciles at the transaction level, not the batch level, across millions of transactions a day.

Most importantly, it doesn’t stop at matching. AI agents work the exceptions: straightforward cases are resolved automatically, with edge cases triaged to your team with a suggested resolution.

End Close fundamentally changes the roles of the teams that work on recon. We strongly believe that operations teams should be released from the drudgery of manual reconciliation. It’s repetitive, low paying, high pressure work that leads to high turnover of staff that could be deployed doing higher value work for the business. Even on the best ops teams there’s a real bus factor risk of having all the knowledge around your manual recon residing with a small number of team members.

We also believe engineering and product teams shouldn’t be on the hook for ongoing maintenance and support of recon systems, which should largely be able to set and forget. Most importantly of all, finance teams should have access to all the data they need to understand their position in real time and shouldn’t have to chase after numbers they can trust.

Collectively this all adds up to what we call financial observability. Software engineers would never deploy a large system without deploying an observability solution like Datadog so they can understand exactly what is going on in the system. We’re building the same thing for payment systems.

When every record’s state is current, every discrepancy is flagged the moment it appears, and every resolution is logged with its reasoning, you have something bigger than reconciliation. You can see in real-time exactly where your money is and where it isn’t and closing the books stops being something that happens once a month.

A note on the use of AI

A common reaction when we share End Close with people is “Oh, so you’re just using an LLM to match the transactions. Surely I could vibe code something similar?” End Close would not be possible without AI but our experience with large scale recon systems means we are very clear eyed about where it can and can’t add value.

At its core End Close is built on a deterministic, rules-based matching system. There’s no hallucinations or unexpected matches and you have a full audit trail that you can stand over. AI sits on top of that engine, providing a level of reasoning and judgement, so that many repetitive tasks can be automated. Your playbooks for common scenarios can be baked into agents which can scale up as required – rather than having a finite team of ops personnel working on manual processes, you can now have 1,000 agents working round the clock but with a clear audit trail of what they did.

One example of how we apply AI for judgement rather than direct matching of transactions is rule management. The teams working on recon are naturally cautious about editing or deleting current rules, so the default action is to add additional rules when conditions change. This of course then leads to even more rules which compounds the problem i.e. there’s so many rules, which ones can you edit or even delete? AI on the other hand has the “patience” and context to test the impact of rule changes at scale, a task that humans will avoid for fear of producing incorrect matches or degrading the system.

What’s next?

It’s early days for End Close. Right now we’re focused on building a system that offers the highest matching rate out of the box (99% increasing to 99.999% as the rules are updated for your unique requirements), automates resolution of exceptions, and simplifies the management of recon rules.

We’re building towards a world where every company has real-time trust in its money. Where reconciliation stops being something you chase and becomes the infrastructure you build reporting, controls, and resolution on top of.

Want to learn more?

Don’t just take our word for it. If you’re struggling with the demands of payment reconciliation we’d love to show you how we can help. Get in touch, send us your historical unreconciled data, and we’ll prove to you that we can automate your process end-to-end.