Rulebase Reads the Calls No One Has Time For

Every day, banks and fintechs record thousands of calls, chats, and emails with their customers. A person disputing a charge. Someone locked out of an account. A caller who sounds, to a tired agent at 4 p.m., completely legitimate - and isn't. Somewhere in that pile is the fraud that costs a firm millions and the compliance slip that gets a manager fined. The traditional way to find it is to have a small quality-assurance team listen to a handful of interactions and hope the important ones were in the sample.

Rulebase, a company from Y Combinator's Fall 2024 batch, was built on a blunt observation about that setup: the sample is where the problem hides. Its software reads all of it.

"We review 100% of interactions" is the line the founders lead with, and in a room full of compliance officers it lands harder than any pitch about artificial intelligence. Rulebase's AI - which the company calls a "coworker" - monitors phone, email, and chat in real time, flags compliance breaches and fraud, and then does the paperwork behind the agent: the QA report, the dispute filing, the audit trail. It plugs into the tools support teams already run, like Zendesk, Jira, and Slack, rather than asking anyone to switch.

01Why sampling stopped working

Manual QA was always a compromise. Listening to calls is slow, so teams review a slice - often a low single-digit percentage - and treat it as representative. That worked when the risks were routine. It works less well now.

The reason coverage suddenly matters is that the threat got cheap. Rulebase points to a roughly 700% rise in voice-phishing and deepfake incidents in 2024. By the company's figures, a scammer can rent about 1,000 AI voice-clone agents for under $55. When impersonation costs almost nothing to produce, catching it can't depend on which 2% of calls a human happened to pick.

02From Goldman and Microsoft to a four-person team

Rulebase was founded in 2024 by Gideon Ebose and Chidi Williams, two engineers who met in London and trace their roots to Lagos, Nigeria. Ebose, the CEO, spent years in product - including work at Microsoft on communication platforms for hospitals - and led design at Gatherwell before it was acquired. Williams, the CTO, wrote back-end systems at Goldman Sachs and, on his own time, built Buzz, an open-source speech-to-text tool that has been downloaded more than 300,000 times and holds over 12,000 stars on GitHub.

That last detail is not decorative. The hard part of catching fraud on a call is understanding what was actually said and meant. Williams had already spent years on exactly that problem before it had a business attached to it.

03A coworker, and a copilot

Rulebase ships two closely related things. Coworker is the back-office engine: it monitors interactions for QA and compliance, manages fraud and dispute cases across a team's existing tools, tracks SLAs with internal staff and third-party partners, and produces audit-ready reports. Beam is the front-line version - a real-time copilot that sits with a support agent during a live call, offers recall of prior context, coaches in the moment, and flags regulatory or fraud risk before the conversation ends.

The distinction matters to how the company positions itself. A lot of "AI for support" is sold as a way to remove agents. Rulebase's pitch is the inverse: keep the agent, take away the invisible, unglamorous work behind them.

04Where it sits, and who it's up against

Rulebase competes on two fronts at once. On one side are the manual QA and compliance teams it is trying to automate. On the other are contact-center AI platforms - names like Observe.AI, Level AI, and Cresta - and voice-fraud specialists such as Pindrop. Rulebase's argument is that those tools tend to specialize: one does QA, another does fraud, another does coaching. It bundles QA, compliance, fraud detection, and dispute automation into a single layer built specifically for regulated finance.

Whether the bundle holds up as competitors expand is an open question. What's clear is that the buyer Rulebase is built for - a compliance or fraud lead who can be held personally responsible when something is missed - values total coverage over a single flashy feature. That's a different sale than the one most contact-center vendors are making.

05Customers before scale

In September 2025, Rulebase announced a $2.1 million pre-seed round led by Bowery Capital, with Commerce Ventures, Transpose Platform VC, Y Combinator, and angels joining. The number is ordinary for a pre-seed. The customer list is not: by the time of the raise, Rulebase counted business-banking platform Rho and a Fortune 50 financial institution among its users. In fintech, enterprise logos usually come after the money, not before it - a small signal that the pain here is real.

The business model is straightforward B2B SaaS: subscription software sold to banks and fintechs, sized to the volume of interactions monitored and the workflows automated. Rulebase is a four-person company selling into some of the most heavily regulated buyers in the economy. The team frames that not as a mismatch but as the point - small teams, they argue, can now credibly chase very large problems.

Ten years out, the specific products may look different. The bet underneath them is durable: that the record of what a company said to its customers is the richest, most-ignored dataset it owns - and that reading all of it, not a sample, is worth paying for.

Sources: Y Combinator, rulebase.co, Rulebase blog, Bowery Capital, UrbanGeekz, Fondo, FounderTrace, Dealroom. Figures on cost and escalation reductions are as reported by Rulebase and its customers; treat them as company-provided estimates.