Big deals and early results put voice AI on the front line of customer service. The phone line, long the most expensive and least loved channel in customer service, has turned into the place where AI companies are staking their futures, with billions changing hands and the first numbers from real customer calls arriving in the same few weeks. Salesforce perhaps made the loudest move, agreeing to buy voice AI specialist Cresta, and just as early adopters began reporting what AI agents deliver on real customer calls. Companies from hotel chains to an Australian credit union point to gains in bookings and customer retention, which go a long way to explaining why buyers will pay so much for the technology. Those results hold only as long as the agents behave, which is, at the moment, far from guaranteed when new benchmark data shows how often leading models still mishandle a caller. Salesforce picks up the phone. Salesforce signed a definitive agreement this week to acquire Cresta, the Stanford AI Lab spinout whose voice agents resolve complex customer calls end to end and whose software coaches human representatives while a call is still live. Kishan Chetan, EVP and GM of Agentforce and Service Cloud, said voice "is where service is hardest to get right." The appeal for Salesforce lies in a channel it has never fully owned, as Service Cloud Voice has leaned on partner telephony and specialist vendors for live-call intelligence. Customers running Cresta on other contact centre platforms have the most to watch, since Salesforce says Cresta will keep supporting those platforms once the deal closes in early 2027. Money follows the phone line. Salesforce was far from the only buyer in voice AI, with Deepgram agreeing to acquire Daily, the company behind Pipecat, a free framework many developers use to assemble voice agents. Deepgram says Pipecat will stay free and move to an independent foundation, but the deal still hands a speech vendor control over tooling that its competitors' customers depend on. The appetite reached HR software too, as Dayforce agreed to buy Ribbon AI, whose voice agents run first-round job interviews. Ribbon reports more than a million completed interviews, so plenty of candidates already meet an AI interviewer long before they speak to a human recruiter. Investors, meanwhile, kept funding younger voice AI companies, with Broccoli raising $52 million last month for agents that answer phones and book jobs for plumbing, heating and electrical firms. Guava followed this week with a $12.5 million seed round for its voice agent platform, which it sells to enterprises and BPOs. Faster voices, cheaper minutes. All that money is pushing the technology forward, and last month alone brought a run of releases that make AI voices faster to respond and cheaper to run. ElevenLabs and Mistral both launched new models designed for live customer conversations, with Mistral offering to host them in Europe for companies that want call audio kept inside the EU, while Cartesia began offering hosted voice agents at a launch price of three cents a minute. Cheaper and faster does little for a compliance team that cannot see what an agent will say, which is why Rime and Salesforce AI Research both released models that draft each answer in writing before speaking it. Concentrix went another way, adding live translation to its calls so customers and advisors who speak different languages each hear the conversation in their own. Agents on the clock. Cheaper models only strengthen the business case once they hold up on real calls, which is why results from companies already running voice agents at scale speak more than any launch. Wyndham's AI Concierge, for one, now handles calls autonomously at more than 1,700 hotels, without staff stepping in. CEO Geoff Ballotti told analysts last month that the agent has answered over three million calls this year, booking rooms at a higher average daily rate than the hotels' own bookings, while catching calls that used to go unanswered has added direct revenue for franchisees. The results extend to outbound calls, where TELUS Digital reported that an ElevenLabs agent welcoming new TELUS home internet customers cut 30-day cancellations to less than half the rate of other new customers across roughly 2,300 calls. The agent went from vendor selection to first call in about six weeks. How quickly a company moves depends on how much risk it will tolerate. Southern Cross Credit Union in Australia moved fast, launching Nova on Salesforce's Agentforce Voice in six weeks after wait times climbed from under two minutes to almost eight. The agent now takes 7% of the credit union's calls. NatWest took a slower approach, asking 90 staff to record about 3,000 banking questions in Glaswegian, Welsh, Yorkshire and other accents to tune Nvidia's speech recognition. The bank says misheard words fell by a fifth, after testing caught a Scottish speaker's "last quarter" landing as "last water". Reality checks for voice AI. The caution looks sensible next to Sierra's latest benchmark, which tested voice agents in six languages and found Spanish, Hindi and Portuguese trailing English on completing tasks, with Korean and Mandarin further behind. Sierra also found that OpenAI's and Google's live voice models missed a caller's turn, replied slowly or failed to give way in more than 80% of calls. Courts and regulators also weighed in last month, with a federal judge dismissing claims from Home Depot callers who alleged Google's Contact Center AI eavesdropped on their calls, while sending callers to Verizon, Uber and Discover to individual arbitration under those companies' terms. Because the brands' own terms did much of the work, companies deploying voice AI should check their consent language before assuming the ruling protects them. Outbound programmes face a firmer deadline in the FCC's revised rules on revoking consent to robocalls, which take effect later this month. Since the regulator treats AI-generated voices as artificial voices under US telemarketing law, the rules apply to AI agents that phone customers about collections, renewals or onboarding. The recent launches in the space definitely made voice agents cheaper and easier to deploy, yet very little of it made them easier to trust, which leaves continuous testing as the essential step for buyers. CXM use AI tools to support parts of its editorial process, including research, transcription, analysis and content development. All published content is reviewed and approved by its team, reflecting its own editorial judgement, insight and voice, with accountability remaining firmly human.

SoundCloud adds direct publishing for AI music created with ElevenLabs. Time: Nov, 5, 2026 ElevenLabs Brings AI-Generated Music to SoundCloud. Audio AI company ElevenLabs announced Thursday that users will be able to publish AI-generated music created with its ElevenMusic platform directly to SoundCloud. ElevenLabs is also offering SoundCloud Artist Pro users three months of ElevenMusic Pro. The plan allows users to generate up to 1,000 songs, separate songs into stem files up to 500 times, and create five custom styles or vocals per month. ElevenMusic's development and adoption. The company released its first music-generation model in 2025, followed in April 2026 by a consumer app that enables users to generate songs with prompts. One month later, ElevenLabs released a new music model capable of switching genres mid-track and handling more complex vocals. According to the company, users have created more than 25 million tracks with its models to date. SoundCloud's AI Integrations. SoundCloud has also been providing artists with AI tools, although it is taking a different approach. Last month, the company partnered with AI music tooling company Moises, allowing artists to export and import songs to and from SoundCloud. SoundCloud also has integrations with platforms including Google Flow Music and music remix platform Hook. "SoundCloud has always been a place where artists experiment, push boundaries, and turn creative ideas into something bigger. By becoming the first platform to bring ElevenMusic creations directly to fans beyond its own ecosystem, we're opening up new possibilities for artists to experiment, share their work, and build an audience," SoundCloud CEO Eliah Seton said in a statement. ElevenLabs' partnership with Spotify. ElevenLabs also has a partnership with Spotify that allows authors to use its audio models to generate audiobook narrations and publish them on the music streaming service. Although Spotify has partnered with labels on products that allow people to use AI to remix tracks, it has not yet announced any model partners such as ElevenLabs. Streaming platforms' approaches to AI music. Overall, music streaming platforms currently appear more focused on labeling AI-generated music and helping users filter it out, particularly as the technology gradually gains popularity among musicians. Deezer, for example, is introducing a toggle that allows users to exclude AI music entirely, while Bandcamp has imposed a blanket ban on AI-generated tracks. Spotify and YouTube have adopted a more cautious approach to labeling tracks.