Monday.com Launches Harmony Voice Agents for Sales Calls and Human Handoffs

Harmony combines automated sales conversations with transfers to human representatives. Its launch puts the focus on what happens between reaching a lead and giving a salesperson a useful conversation to continue.

monday.com announced Harmony on October 19, 2026, targeting mid-market and enterprise customers in the United States, United Kingdom, Australia and Singapore. India is absent from the initial availability list.

The company says the platform handles qualification, follow-ups, objection handling and live transfers with a briefing for the receiving representative. It also includes dialing, inbound routing and CRM updates, with telephony capabilities built on Twilio.

For sales teams, the important question is whether those steps stay connected. Reaching someone quickly has limited value if the agent records the wrong requirement, promises something the business cannot deliver or transfers the caller without context.

The technology behind Harmony

According to monday.com, Harmony uses proprietary language models assigned to individual functions, coordinated by a predictive engine. The models run largely on its own infrastructure. The company reports approximately 200 milliseconds of audio-in to audio-out latency.

That is a company performance claim. It does not establish how consistently the system responds during interruptions, background noise or a slow external lookup. A short response time also says little about whether the response is correct.

The launch follows monday.com’s May announcement of an agreement to acquire OneAI. October’s release identifies acquired OneAI voice technology as part of Harmony. The acquisition announcement and the product launch are separate events.

A human handoff is part of the product

Harmony’s product FAQ describes an agent that handles outreach and transfers interested prospects to a human who closes the sale. It says agents identify themselves as AI by default, with configurable disclosure. The site also describes consent capture, recording controls, transcripts and audit trails.

Those controls need to be evaluated in the actual calling workflow. A caller who asks for a person should reach the intended team, and the receiving representative should see the caller’s needs, unresolved questions and any commitments already made.

Consider a prospective customer who first asks about a product, then corrects their budget and requests a callback tomorrow. A useful handoff preserves the corrected budget and callback preference. A summary that keeps the original answer could cause the human team to continue the conversation on the wrong basis.

The same issue applies to CRM updates. Buyers should check which fields an agent can change, how it handles conflicting information and what happens when a write fails. A fluent call should not conceal an incomplete business record.

The launch metrics need separate denominators

The Next Web’s launch report notes monday.com’s claim of contact rates up to 70%, while pointing out that the measurement method was not disclosed. It also reports that pricing was not published.

A contact rate measures a different stage from qualification, a completed transfer or a sale. Buyers need the denominator: unique people, individual attempts or another population. Repeated calls to the same lead can make those measures difficult to compare.

The named customer example is Bright Car Finance. In a statement also carried by Business Chief, director Richard Stubbs says the agent qualifies roughly half of its conversations and transfers 80% of those qualified conversations live to advisers. He reports a 9% increase in deals closed in less than a year.

This is customer testimony presented with the launch, rather than a controlled performance study. The 80% figure applies to qualified conversations, not all leads. The reported 9% increase in deals also should not be rewritten as a nine-percentage-point increase in conversion rate.

For a prospective customer, the useful comparison is the number of additional suitable opportunities that reach a representative and become business, after accounting for the cost of calling and human follow-up.

Disclosure and permissions belong in the evaluation

Harmony’s FAQ says compliance depends substantially on configuration. Its description of controls does not establish that every deployment will meet every applicable requirement.

For product evaluation, teams should make the agent’s authority concrete. Can it only explain approved terms, or can it offer a discount? Can it schedule a callback, or also commit a representative to a particular time? What should happen when the customer declines further contact?

These are related to the permission boundaries explored in AgentsCalling’s analysis of what persistent approval allows an AI agent to do. A sales agent needs limits that remain clear as a conversation changes direction.

What a useful pilot should measure

A controlled pilot could use consenting internal testers and fictional customer records before reaching real prospects. Include a caller who interrupts, corrects a detail, declines the offer, requests a person or asks for a promise outside the approved script.

Track whether each conversation ends with the intended outcome. Check the CRM record against the call, confirm that the representative receives the right briefing and inspect what happens when no representative is available.

Measure successful handoffs and qualified opportunities alongside response speed. Include platform charges, telephony, setup and review time when calculating the cost per useful outcome. Compare equivalent lead groups and follow-up windows so that a better lead list is not mistaken for better agent performance.

Harmony’s opportunity is to shorten the distance between customer interest and a productive human conversation. Its value for an individual sales team will depend on how accurately it carries information, preferences and responsibility across that distance.