The New York start-up is pushing autonomous agents into the long-horizon onboarding and reactivation work that regulated fintechs usually leave half-finished – though its headline performance figures are unaudited and self-reported.
Rulebase, a New York start-up building AI agents for financial services, has launched a product it calls Revenue Agents: autonomous software that takes ownership of a customer onboarding or reactivation case and works it over the days or weeks a conversion can take.
Rulebase puts the impact at a 120% lift in recovered accounts and a halving of time-to-activation across fintechs in the US, EU and Africa.
The pitch targets a familiar gap in regulated onboarding. A merchant clears most of a verification check, stalls on one rejected document and is never chased to the finish. An approved account never funds. A customer transacts once and goes quiet.
The argument is that the tools most fintechs run are built to close a single support interaction and move on — not to follow a customer across the weeks it takes to turn an approval into a live, paying account.
One agent per case
Revenue Agents sit on a new layer Rulebase calls the Customer Agent Runtime. Each agent is trained on a customer’s own policies, queues and data, and treats every objective – document collection, exception triage, reactivation outreach or service recovery – as a single long-lived task attached to the underlying business object: an application, a ticket, a transaction, a document. The runtime holds the state, the evidence, the next step, the governing policy and the condition that counts as done.
The design is event-driven rather than a continuously running model. An agent wakes on a signal, takes or recommends the next action, records the result and waits, an approach Rulebase says keeps compute cost roughly in line with the value of each case rather than burning idle model loops. An action policy sets what each agent may draft, execute or escalate, and consequential customer-facing actions require a human to sign off.
None of it has been independently tested. Rulebase has not said how many fintechs run the runtime in production, at what volume, or shown outside verification – so how the architecture behaves in a live regulated queue is not yet something a buyer can check.
“Every fintech is leaking revenue it has already won,” said Gideon Ebose, chief executive and co-founder of Rulebase. “Customers who clear verification but never go live. Accounts approved but never funded. Customers who transacted once and quietly went dormant. That revenue isn’t lost to bad decisions, but to follow-up work that nobody owns long enough to finish.”
Agents move into regulated work
The launch lands amid a wider shift. Through 2026, agentic AI, systems that plan and act across a workflow rather than answer a single prompt, has moved from experimentation towards operational reality in banking back offices, running parts of KYC, AML screening and customer operations that were until recently manual, according to industry analysis of agentic AI in financial services.
That migration into regulated territory is where the caution sits. Model risk, auditability and the point at which a human signs off remain unsettled, and agents acting on onboarding and reactivation touch KYC, KYB and consumer-protection rules directly.
“If you don’t solve the guard function, I don’t see AI at scale in banks at all,” said Jouk Pleiter, founder and chief executive of banking software firm Backbase, on agentic compliance in banking.
Rulebase’s answer is the built-in approval step and the action policy — human oversight where it counts, in Ebose’s phrasing. Whether that holds up under a regulator’s scrutiny, rather than a vendor’s, is not something a product launch can settle.
Built on an existing product
Revenue Agents extend a narrower product Rulebase already sells: AI that reviews 100% of customer interactions across voice, email and chat, against the 3% to 5% a manual quality-assurance team typically samples.
Its one named reference is US business banking platform Rho, which Rulebase says used that coverage to cut manual QA by roughly 90% while flagging four times more issues.
The wider company is young and lightly capitalised — founded in 2024, through Y Combinator’s Fall 2024 batch, on a $2.1 million pre-seed round led by Bowery Capital.
Set against that stage, it is a sizeable ask: hand an agent a live, regulated onboarding queue on the strength of the company’s own unaudited figures.
