Monk co-founders George Kurdin and Joe Zhou have added outbound phone calls and inbound callbacks to Julia, their AI accounts receivable agent, extending a product built around email into a channel where mistakes can carry a higher financial and reputational cost.
Monk described the feature in its Voice Collections product post and issued a formal launch release on July 23. Each customer organization receives a dedicated number that Julia can use to call accounts selected by a collections playbook. Customers can call the same number back with questions about invoices, payments and bank details.
The New York software maker is betting that the phone can become an automated extension of the collection record instead of a separate manual queue. Calls appear alongside email history, invoice information, payment updates and earlier collection activity. That gives a finance worker reviewing an escalation the conversation that preceded it, rather than a note reconstructed after the call.
The rollout also pushes Monk into a harder version of the problem Kurdin and Zhou chose when they started automating accounts receivable. A poorly worded email can irritate a customer. A voice agent can disclose financial information, misunderstand a dispute or make an unauthorized commitment in real time. Monk's approach is to give Julia a narrower job than the product's conversational interface might suggest.
A phone agent designed to say no
Julia is read-only during calls. Monk says the agent can confirm an invoice amount or due date, check whether a payment has arrived, summarize why the collections number called and route a question to a person. Julia cannot rewrite an invoice, change its payment status or accept sensitive payment changes by phone.
The lookup process also requires references supplied by the caller. Someone asking about an invoice must provide both the company name and invoice number. For payment or ACH questions, Monk asks for details such as the relevant company and exact amount. Julia will not use a single piece of information to search broadly across customer records.
Zhou's fingerprints are clearest in those restrictions. Before Monk, he worked on financial data at Intuit, Android infrastructure at Google and growth systems at Snap. Monk says he became a top-30 contributor to Snap's codebase and built infrastructure used by dozens of internal teams. He and Kurdin began exploring accounts receivable through roughly 300 conversations with CFOs, controllers and finance leaders, according to a July profile published by Monk.
Zhou has described a 1% error rate in financial work as unacceptable. Voice Collections applies that view by limiting what the model can retrieve and reserving actions that alter money or records for a person. Escalation rules can assign calls needing judgment to a designated employee, while the transcript and call outcome remain attached to the account.
Customers can configure the agent's displayed name, voice, caller ID, speaking pace, voicemail behavior and escalation rules through Monk's voice settings. Monk enables the feature on an opt-in basis and works with each organization to configure the number and calling behavior before activation.
Kurdin's bet on controlled automation
Kurdin arrived at finance software by an unusual route. He worked at D.E. Shaw, Minecraft and Streamlabs, and Monk says he spent six years playing professional poker, logging more than 1.5 million hands. He has used that experience to draw a distinction between accepting variance at a poker table and building software that touches customer money.
His product thesis is that models need a surrounding "agent harness": deterministic workflows, logging and escalation rules that constrain where an AI system can act. Monk runs those checks before Julia drafts collection communications, according to Kurdin's explanation of the architecture. The same principle now governs phone calls.
That matters because Monk's evidence for Julia's performance comes from the existing collections product, rather than Voice Collections itself. Across its first 100 customers, Monk says Julia produced a 24% higher response rate than standard dunning and resolved 88.2% of collections without human intervention. Monk has not disclosed how many customers are using the voice feature or published a voice-specific payment recovery rate.
Monk also says more than $1.5 billion in receivables is managed on its platform. At Pump, which handles more than 1,500 customers, Monk claims it helped collect over $10 million in recent months. Those figures are company-reported, and Monk has not disclosed revenue, pricing or retention.
Phone calls become an AR software feature
Voice Collections enters a market where phone automation is moving into broader finance products. Billtrust introduced Agentic VoIP in January, focusing on assisting human collectors and automatically documenting calls. Monk is taking the automation further by letting Julia conduct outbound conversations and answer callbacks.
The wider accounts receivable category has also attracted substantial venture funding. Tabs raised a $55 million Series B in September 2025 to expand agents across billing and collections. Fazeshift announced a Series A in May that brought its total funding to $22 million.
Monk joined that funding cycle in April with a $25 million Series A co-led by Footwork and Acrew Capital, with Better Tomorrow Ventures returning. Monk's original round announcement put total funding at $29 million, including a $4 million seed, although the July 23 Voice Collections release says Monk has raised $25 million. No valuation was disclosed.
The central competitive question is how much authority AR platforms give voice agents as phone automation becomes a standard feature. Generic voice infrastructure can place a call. Monk is trying to own the financial context around that call: the invoice, payment history, collection playbook, prior emails and audit trail.
Kurdin and Zhou now have to show that the restrictions are tight enough for finance leaders while leaving Julia useful enough to collect money. The early metrics Monk cites measure email-based collections across its customer base. Voice Collections will be judged on a more demanding set of outcomes: whether customers answer, whether calls produce payment commitments, how often Julia escalates and whether the agent can protect a commercial relationship while asking for overdue cash.