monday.com today began selling Harmony, a voice AI platform that places and answers sales calls for mid-market and enterprise companies, with the Nasdaq-listed software company claiming its agents respond in about 200 milliseconds and reach "a contact rate of up to 70%." The only customer result in the announcement comes from a UK car finance broker, which says it closed 9% more deals in less than a year.
In Short
monday.com, the company known for work management software, has started selling AI phone agents that call sales leads, talk to them like a person would, and pass the promising ones to human staff. That matters to anyone paying for leads through search or social ads, because many of those leads go cold before a salesperson ever calls them back. The speed and conversion figures come from monday.com and one customer, and the announcement does not explain how they were measured.
What monday.com announced
The announcement was issued through Business Wire at 9:00 a.m. Eastern Daylight Time on October 19, 2026, with a dateline of Tel Aviv and New York. According to monday.com, Harmony is "a voice AI agent platform purpose-built for enterprise sales teams." It is sold under its own name, Harmony.ai, rather than inside the monday.com work management product that most marketers know.
The pitch is narrow. Harmony is not positioned as a general customer service bot. According to the company, its agents run "the full sales cycle, from prospecting, discovery, and qualification to objection handling, pitching, and warm transfers to live reps with a full briefing." In other words, the software is meant to phone the people who filled in a form or applied for a product, work out whether they are worth a salesperson's time, and hand the good ones over mid-call.
monday.com frames the commercial case around money already spent. Harmony, the company says, increases revenue "by turning pipelines that businesses have already paid for into closed deals faster and more consistently than any team could on its own." For a performance marketer, that pipeline is the output of paid search, paid social and affiliate spend. The cost of acquiring a lead is sunk by the time the lead arrives; what happens afterwards determines whether the media budget produced revenue.
Vitali Margolin, Head of Harmony, described the target as the harder end of the call market. "We are tackling the more complex, higher-stakes calls where every wrong answer costs money and every missed opportunity risks a potential deal," Margolin said. "Unlike generic solutions that rely on third-party LLMs, our specialized models and structured logic translate into conversations that sound human and a response time that feels human. It's what makes the difference when the goal isn't just completing a call, but growing a business."
How the system is built
The technical description is more specific than most launch releases in this category, though it stops short of benchmarks.
Small models, one job each
According to monday.com, Harmony's agents are "powered by voice technology acquired from OneAI and a suite of proprietary, task-specific LLMs trained on real customer conversations and orchestrated by a real-time predictive engine." Each model handles a single function. The company lists qualifying leads, capturing data and handling objections, followed by "and more," so the full set of models is not disclosed.
The release calls these "specialized, small Large Language Models," a phrase that sits somewhere between a marketing label and an engineering choice. It does not give parameter counts, training data volumes, or how many customer conversations the models were trained on. Nor does it say whose conversations they were, or under what consent terms recordings were used for training. The OneAI acquisition is mentioned without a date or price.
The design follows an argument that has been building in AI research for over a year. In 2025, NVIDIA researchers argued that models under 10 billion parameters could handle an estimated 60-80% of AI agent tasks then assigned to models above 70 billion parameters, with serving costs 10 to 30 times lower. monday.com does not cite that work, but its reasoning runs the same way: narrow models are cheaper to run and easier to constrain than a single general model asked to do everything.
The 200 millisecond figure
Speed is the headline specification. According to the company, "Because they are compact and run largely on Harmony's own infrastructure, Harmony's models deliver approximately 200ms audio-in to audio-out latency, producing speech rhythm that is indistinguishable from humans, with sharper accuracy."
Three qualifiers sit inside that sentence. The figure is "approximately" 200 milliseconds. The models run "largely," not entirely, on Harmony's own infrastructure, which leaves open which parts of a call depend on outside services. And the measurement point, audio in to audio out, is the right one for a phone conversation, because it covers listening, deciding and speaking rather than just the text generation step. What the release does not give is a distribution: whether 200 milliseconds is a median, an average or a best case, and what happens at the slow end when a caller asks something the models were not trained on.
Latency is a familiar constraint in advertising technology, where auctions run against hard deadlines and every added delay costs bids. In voice, the penalty is social rather than financial in the first instance: a pause of a second or more is enough for a caller to notice they are talking to a machine. The claim that Harmony's speech rhythm is "indistinguishable from humans" is monday.com's own characterisation. No listening test, sample size or third-party evaluation is cited.
Twilio underneath, call center on top
The voice models are only part of the product. According to monday.com, Harmony also offers "complete AI call center capabilities including unlimited phone numbers with branded caller ID, predictive and power dialing, inbound routing, built-in consent and compliance controls, and native CRM write-back." These, the company says, are "built on top of Twilio's platform to ensure enterprise grade quality."
That stack matters for how the product would sit inside a marketing operation. Native write-back to the CRM means call outcomes, such as whether a lead qualified or was transferred, land in the same system that many advertisers use to send offline conversions back to Google Ads, Meta and LinkedIn. The release says only that Harmony "integrates with major CRM platforms." It does not name them.
The company also claims Harmony runs customer calls "end-to-end, 24/7, in any market, without limitations, at lower infrastructure cost than general-purpose AI at call-center volume." No price, per-minute rate or cost comparison is published. The phrase "in any market" sits awkwardly next to the availability statement two paragraphs later, which limits the launch to four countries.
The 70% contact rate
The most prominent number in the announcement appears only once, in the subheadline: Harmony's models "produce a contact rate of up to 70%." The body of the release never returns to it.
Contact rate is usually understood as the share of dialled leads who are actually reached and speak to someone. The release does not define it, does not say which customers or call types produced the 70% figure, and does not give the contact rate those customers achieved before Harmony. "Up to" marks it as a ceiling rather than a typical result. Without a baseline, the number does not show whether Harmony reaches more people than a human team calling the same list at the same speed.
For comparison, call analytics vendor Invoca published figures in 2026 showing that 52% of home services calls reach a person, with wide variation by trade. Those were inbound calls to businesses rather than outbound calls to leads, so the two numbers measure different things and are not directly comparable.
One customer, one set of numbers
The only named customer is Bright Car Finance Ltd, which monday.com describes as "an FCA-regulated UK car finance broker." Its director, Richard Stubbs, gave the release its only outcome data.
"We generate around 2,500 approved applications a month but were only converting about 5% of them into finance agreements because leads went cold before a rep could reach them," Stubbs said. "Harmony's AI agent now follows up with every approved applicant right away. It qualifies about half of the conversations it has and hands 80% of those live to our advisers, driving a 9% increase in deals closed in less than a year. It's taken the repetitive work off our team so they can focus on the conversations that really matter."
The figures can be partly reconstructed. On PPC Land's own calculation, 5% of 2,500 approved applications is roughly 125 finance agreements a month before Harmony. If the 9% increase is relative, that would mean roughly 136 a month afterwards, about 11 extra agreements. If it means nine percentage points, conversion would have risen from 5% to 14%, close to tripling output. The quote does not say which, and the two readings differ by an order of magnitude.
The funnel inside the quote also leaves gaps. Half of conversations qualify, and 80% of those are handed to advisers, so roughly 40% of the conversations Harmony holds end in a live transfer, again on PPC Land's arithmetic. But the quote counts conversations, not applicants, and it does not say how many of the 2,500 monthly applicants the agent actually reaches. "Less than a year" also leaves the measurement window and start date open. There is no control group, so the 9% cannot be separated from seasonality, changes in lead volume, interest rates or other changes at the broker over the same period.
None of this makes the result wrong. It means the one outcome figure in the launch is a single customer's description, with no method attached.
Where it is available
According to monday.com, Harmony "is now available for mid-market and enterprise customers in the United States, United Kingdom, Australia, and Singapore, with additional markets and languages to come." The release does not list the languages supported at launch, and it does not publish pricing or a minimum contract size.
The four markets each regulate automated sales calls differently. The company points to "built-in consent and compliance controls" without describing them. In the United States, the Federal Communications Commission ruled in February 2024 that AI-generated voices count as "artificial" under the Telephone Consumer Protection Act, which brings outbound AI calls under the law's consent requirements. Bright Car Finance operates in a sector the UK's Financial Conduct Authority supervises, which is why the release mentions its regulated status. How Harmony handles disclosure that a caller is speaking to software, call recording consent, or opt-outs in each market is not addressed. Recording rules carry real exposure: PPC Land reported in August that a proposed class action against Granola sought statutory damages of $5,000 per violation under California privacy law over an AI notetaker that allegedly recorded participants without telling them.
Why it matters for advertisers
The problem Harmony targets is not new to anyone buying leads. A form fill or a missed call costs the same whether a salesperson follows up in two minutes or two days, and the conversion rate is far better at two minutes. Ad platforms have increasingly priced that gap onto advertisers. Google began charging Local Services Ads advertisers for missed business-hours calls longer than 20 seconds from October 1, 2026, which turned slow phone handling from a sales issue into a media cost.
Vendors have moved quickly into that space. Yelp agreed in January 2026 to buy Hatch for about $270 million in cash to automate lead follow-up for service businesses, and in September it put OpenAI's GPT-Live-1 into its voice agents after more than one million calls. Invoca packaged AI voice and messaging agents for home services on October 6. LSArespond said it passed 2,000 users of its Local Services Ads automation tool on October 11.
Most of those products sit with small local businesses and inbound calls. Harmony is aimed elsewhere: outbound calling to leads at mid-market and enterprise companies, with the dialer and the qualifying logic in one product. Its sharpest point of difference is architectural. Yelp builds on an OpenAI model; Margolin's quote sets Harmony against "generic solutions that rely on third-party LLMs." Whether a stack of small proprietary models performs better on a sales call than a frontier general model is an open question that the release asserts rather than demonstrates. The economics are clearer in principle. Inference cost scales with call volume, and a company running thousands of minutes a day has an incentive to own smaller models rather than pay per token to a general-purpose provider.
There is a measurement consequence too. If AI agents qualify leads before a human touches them, the qualification decision moves from a salesperson's judgment to a model's. Advertisers that feed qualified-lead or lead scoring signals back into bidding algorithms would be training Google's or Meta's systems on a vendor model's verdicts. The release says nothing about how Harmony decides a lead is qualified, or whether those decisions can be audited, which is the part of the system most likely to shape what the ad platforms optimise toward.
The skepticism about autonomous agents that PPC Land has tracked through 2026 applies here as well. A Nylas survey in March found only 4% of teams let AI agents act without any human approval. monday.com's own release acknowledges the risk, warning that generative AI tools "can go off-script, creating legal, reputational, and regulatory damage" and saying enterprises need systems that are "controllable, predictable, debuggable." Harmony is pitched as that system. The announcement offers architecture as evidence; it does not offer error rates, escalation rates or audit results.
The parent company
monday.com describes itself as "the AI workspace where people lead and agents act," and says it is "trusted by nearly 250,000 customers worldwide." Harmony is a departure from its core product, which manages tasks and workflows rather than phone calls. The release does not say whether Harmony will be bundled with or sold into the existing customer base.
The timing places the launch shortly before two scheduled company events listed alongside the release on Business Wire: Elevate '26, monday.com's annual conference, on October 28-29, 2026 at the Javits Center in New York, and third quarter 2026 results on November 9, 2026. The Harmony announcement contains no financial figures, no revenue expectations and no customer count beyond the single named broker.
Timeline
- February 2024 - The FCC rules that AI-generated voices count as "artificial" under the Telephone Consumer Protection Act
- June 2, 2025 - NVIDIA researchers argue small language models can handle 60-80% of agent tasks at 10-30 times lower serving cost
- October 21, 2025 - Yelp releases Yelp Host and Yelp Receptionist AI call answering tools
- January 21, 2026 - Yelp agrees to buy Hatch for about $270 million in cash
- March 10, 2026 - Nylas survey finds only 4% of teams let AI agents act without human approval
- July 30, 2026 - A proposed class action against Granola alleges its AI notetaker recorded people without telling them
- August 25, 2026 - Google tells Local Services Ads advertisers missed calls over 20 seconds become billable
- September 10, 2026 - Yelp puts OpenAI's GPT-Live-1 into its Yelp Host and Hatch voice agents
- October 1, 2026 - Google's missed-call billing rule for Local Services Ads takes effect
- October 6, 2026 - Invoca releases AI voice and messaging agents for home services
- October 11, 2026 - LSArespond says it passed 2,000 users
- October 19, 2026 - monday.com makes Harmony available in the US, UK, Australia and Singapore, claiming about 200 millisecond response times
- October 28-29, 2026 - monday.com's Elevate '26 conference at the Javits Center, New York (scheduled)
- November 9, 2026 - monday.com third quarter 2026 results (scheduled)
Related PPC Land coverage
- Invoca packages AI agents as Google bills missed LSA calls over 20 seconds - Invoca's home services AI agents and its call answer-rate data, released days after Google's missed-call billing began.
- Yelp and Hatch put OpenAI's GPT-Live-1 into voice AI after 1 million calls - Yelp's move to a full-duplex OpenAI voice model in its restaurant and home services call agents.
- Yelp bets $270M on AI with Hatch acquisition to automate services leads - The deal that brought automated lead follow-up into Yelp's services advertising business.
- LSArespond tops 2,000 users as Google moves Local Services Ads into PMax - A smaller vendor automating lead rating, replies and CRM sync for Local Services Ads.
- Google charges Local Services advertisers for missed calls over 20 seconds - The billing change that put a media cost on unanswered phone leads.
- NVIDIA research challenges $57 billion AI infrastructure strategy with small language models - The research case for small, task-specific models in AI agents.
- Agentic AI is already in production - most teams just won't say it - Survey data on how much autonomy companies actually give AI agents.
- Google Ads filters miss fake leads, so advertisers must add 3 form checks - Google's guidance on invalid leads, relevant to what voice agents end up calling.
Summary
Who: monday.com (Nasdaq: MNDY), through its new Harmony unit led by Vitali Margolin, with UK car finance broker Bright Car Finance Ltd as the only named customer.
What: Harmony, a voice AI platform whose agents call and qualify sales leads and transfer them to human staff, built on task-specific proprietary models, voice technology acquired from OneAI, and Twilio's communications platform. monday.com claims about 200 millisecond audio-in to audio-out latency and a contact rate of up to 70%, without a baseline or method; Bright Car Finance says deals closed rose 9% in less than a year, without saying whether that is a relative or percentage-point change.
When: Today, October 19, 2026, at 9:00 a.m. Eastern Daylight Time.
Where: Available to mid-market and enterprise customers in the United States, United Kingdom, Australia and Singapore, with the announcement datelined Tel Aviv and New York.
Why: Leads bought through advertising often go cold before a salesperson calls back, and ad platforms such as Google have started charging for some unanswered calls. monday.com is betting that small, narrow models can handle sales calls faster and more cheaply than general-purpose AI, though the release offers no pricing, error rates or independent testing to support that.
Discussion