AI phone agents: which startup is ahead?

Why are AI phone-agent startups suddenly getting so much attention?

AI phone agents are now handling millions of real customer calls every day, and the biggest startups are turning that usage into serious businesses.

In May 2026, TechCrunch reported that Vapi had processed more than one billion calls since launching and was handling between one million and five million daily. Stripe documented Retell AI reaching more than 30 million monthly calls, while PolyAI disclosed more than 2,000 live deployments across over 100 enterprise customers in December 2025.

Those numbers show how quickly the market has changed. A few years ago, voice automation largely meant rigid menus and scripted responses. Today's systems can interrupt naturally, look up customer information, make reservations, update records and transfer complicated requests to humans. Voice that can interrupt, look up records and hand off to a human is the sort of technology we watch gaining traction in our agentic AI report.

The startups are taking different approaches. Vapi and Retell provide infrastructure for businesses to build telephone agents. PolyAI manages deployments for large organizations. Sierra, Decagon and Parloa offer broader customer-service products that work across phone, chat and other channels.

That difference is central to understanding who is ahead. The company processing the most telephone calls may have a smaller business than a competitor selling expensive enterprise contracts.

Which AI phone-agent startup is actually the biggest?

Sierra currently has the strongest commercial position among independent enterprise AI-agent startups, although Retell AI and Vapi are more directly focused on telephone infrastructure.

Sacra estimates that Sierra reached approximately $200 million in annual recurring revenue in May 2026. Retell reported $80 million in annualized revenue by September, while Sacra estimated that Decagon had reached around $100 million in July.

Sierra's lead comes with an important distinction: its revenue includes customer interactions across multiple channels. Retell's business is more closely tied to voice agents, making it easier to judge as a dedicated phone-agent company. A lead that depends on what you count is hard to read, so we sort out who is really ahead in our report on who's ahead in agentic AI.

Vapi has reported enormous call volume but has disclosed less precise financial information. TechCrunch described its annual recurring revenue run rate as being in the eight figures in May 2026, without identifying an exact amount.

PolyAI adds another complication. Its thousands of enterprise deployments establish a substantial operating presence, but we lack a comparable public revenue figure.

Based on the available evidence, we rank Sierra first for enterprise commercial scale, Retell first among dedicated phone-agent platforms, and Vapi as the strongest challenger in developer adoption and reported telephone traffic.

Startup	Strongest available measure	What it tells us
Sierra	~$200M estimated ARR	Broad enterprise commercial leader
Decagon	~$100M estimated annualized revenue	Fast-growing enterprise challenger
Retell AI	$80M reported annualized revenue	Strong dedicated voice-agent business
Vapi	1M–5M calls daily	Major developer and telephone infrastructure platform
PolyAI	2,000+ live deployments	Large managed enterprise footprint
Bland AI	3.5M+ calls weekly	Established large-scale calling operation
Parloa	$50M+ reported ARR	Growing enterprise customer-service platform
ElevenLabs	$500M+ ARR across products	Larger voice technology business, including non-phone services

Is Sierra really ahead of the other AI customer-service startups?

Sierra is currently the clearest revenue leader in enterprise AI customer service, and its growth has been unusually fast even by the standards of the AI industry.

According to Sacra, Sierra grew from approximately $26 million in ARR at the end of 2024 to $130 million at the end of 2025, before reaching an estimated $200 million in May 2026. That represents roughly 7.7 times the original revenue figure in less than two years.

The company also raised $950 million at a $15.8 billion valuation in May 2026, compared with a $10 billion valuation in September 2025. Investors were paying nearly 60% more for Sierra just eight months later.

Its expansion has increasingly involved telephone conversations. Sacra reported that voice overtook text as Sierra's main interaction channel in October 2025. Sierra is now a genuine competitor in phone automation, even though it also serves digital customer-service channels.

Sierra sells more than access to speech technology. Its enterprise contracts cover deployment, integrations and ongoing improvements, with pricing linked to usage or successful resolutions.

We still cannot isolate how much of its revenue comes specifically from phone calls. But for large businesses buying AI agents to handle customer service, Sierra has built the strongest publicly documented commercial position among the independent startups we reviewed.

Is Retell AI the fastest-growing dedicated phone-agent startup?

Retell AI has one of the most impressive revenue growth records in the AI phone-agent market, especially considering how little funding it initially raised.

Retell's official company history says its annualized revenue reached $40 million in November 2025 and doubled to $80 million by September 2026. Sacra estimates an even earlier baseline of around $5.4 million in 2024, suggesting that the business grew approximately fifteenfold over the period.

The company disclosed a $4.6 million seed round in May 2024. Its own history also describes reaching $40 million in annualized revenue with a team of just 25 people and becoming profitable.

That relationship between early funding and revenue is unusual in a sector where competitors have raised hundreds of millions. However, the seed amount should not be treated as a verified total of all capital available to Retell today.

There is evidence beyond the company's own growth announcements. Brex ranked Retell fourth among the fastest-growing software vendors in its 2025 benchmark, based on spending activity from more than 35,000 companies.

Retell's main advantage is that customers can build, test and operate voice agents without developing the underlying calling infrastructure themselves. It serves businesses in healthcare, insurance, financial services, logistics and retail.

For a startup specializing in AI telephone agents, Retell currently has the strongest combination of disclosed revenue, growth and capital efficiency that we found.

Is Vapi processing more AI phone calls than Retell AI?

Vapi appears to be handling at least as much telephone traffic as Retell AI, and possibly considerably more, although their public figures cover different periods.

TechCrunch reported in May 2026 that Vapi was processing one million to five million calls daily. At a constant rate, that would translate into approximately 30 million to 150 million calls over a 30-day period.

Stripe's Retell case study reported more than 30 million monthly calls. Both platforms have therefore reached the same broad order of magnitude, but Vapi's upper daily range suggests substantially greater potential throughput.

Vapi also reported more than one million developers having used its platform. That is an impressive distribution figure, though registrations and experimentation cannot be equated with paying customers.

We should be careful about declaring a definitive traffic winner. The disclosed figures are snapshots from different points in time, and neither company provides a consistently updated, independently audited call-volume series.

Still, the comparison is revealing. Retell has built a large revenue business around voice infrastructure, while Vapi has achieved enormous developer reach and reported calling volume. These are two credible contenders for the lead, with different strengths.

How did Vapi convince Amazon Ring to choose its phone agents?

Vapi's Amazon Ring contract is one of the strongest examples of a dedicated AI phone-agent startup winning a major enterprise customer against serious competition.

According to TechCrunch's May 2026 reporting, Ring evaluated more than 40 voice AI vendors before selecting Vapi. Ring subsequently routed 100% of its inbound calls through the platform. A buyer weighing more than 40 voice vendors needs a short list, and we cover the top companies today in our report covering agentic AI.

That selection process tells us more than an ordinary customer announcement. Vapi had to compete with dozens of alternatives and meet the requirements of a large consumer technology business.

Ring reportedly wanted granular control over how its agents behaved during live conversations. That fits Vapi's approach: developers can configure models, calling workflows, integrations and other components rather than relying entirely on a prebuilt customer-service product.

Vapi has also named New York Life, Intuit, Kavak and Instawork among its enterprise customers. These businesses operate in quite different industries, suggesting that its infrastructure can serve more than one narrow telephone workflow.

The startup raised $50 million in Series B financing at approximately a $500 million valuation shortly after the Ring deployment became public. Total funding reached $72 million.

Vapi has now shown it can win a demanding enterprise procurement while maintaining a broad developer platform. Its next commercial challenge is converting more of those deployments into predictable, high-value revenue.

Is PolyAI further ahead in real enterprise deployments?

PolyAI is one of the strongest AI phone-agent startups for large-scale enterprise deployment, particularly when customers need agents across multiple locations, languages and countries.

In December 2025, PolyAI announced that it served more than 100 enterprises, with over 2,000 live deployments across 45 languages and more than 25 countries.

That is a different kind of scale from Vapi's developer registrations. PolyAI's figures refer to agents already operating in customer environments, often spread across numerous locations.

The company also raised $86 million in Series D funding, taking cumulative financing above $200 million. Its customer portfolio includes Marriott, Caesars Entertainment, PG&E and UniCredit.

The Fogo de Chão restaurant chain provides a particularly useful example. Before adopting PolyAI, the company reportedly missed nearly one-quarter of its millions of annual incoming calls. Its AI receptionist, Selma, now answers U.S. calls, handles reservations through OpenTable and helps customers with loyalty registration.

According to PolyAI's published case study, Fogo achieved 95% guest satisfaction and an 88% booking completion rate. Those are customer-reported deployment results rather than independent industry benchmarks.

PolyAI has already demonstrated that voice agents can operate across large, geographically distributed businesses. We would place it near the top of the market for managed enterprise telephone deployments.

Is Bland AI becoming a serious competitor to Vapi and Retell?

Bland AI is already large enough to challenge established phone-agent platforms, and its focus on difficult conversations gives it a distinct place in the market.

In June 2026, Bland announced that it was handling more than 3.5 million calls weekly. Maintaining that rate would mean more than 182 million calls annually, although that calculation assumes traffic stays constant.

The company also raised $50 million in Series C funding, bringing its total above $100 million. Customers include Samsara, Kin Insurance and CNO Financial Group.

Bland's technical strategy differs from Vapi's. Rather than building primarily around externally supplied foundation models, Bland develops its own models designed specifically for voice conversations.

That approach is particularly relevant to long and unpredictable calls. Bland says some of its conversations last 30 to 45 minutes and involve situations such as insurance assistance or healthcare support, where callers may interrupt, change direction or need escalation.

Its reported volume and industry exposure are substantial. What we cannot yet establish is how Bland compares with Retell on revenue, customer retention and the cost of successfully completing a comparable task.

Bland belongs among the serious competitors today, especially for businesses dealing with complicated telephone conversations.

Can Parloa compete with Sierra for big enterprise customers?

Parloa is one of Sierra's strongest enterprise challengers, particularly for multinational businesses that need voice agents integrated into existing customer-service operations.

In January 2026, Parloa raised $350 million at a $3 billion valuation, tripling its value from the previous funding round eight months earlier.

Its commercial growth is also substantial. In December 2025, Parloa confirmed that it had surpassed $50 million in annual recurring revenue and reported 150% net revenue retention.

That retention figure deserves attention. It means that, on the company's reported measure, customers from an existing cohort were spending 50% more after accounting for expansions, contractions and churn over the measurement period. Companies appear to be adding more usage or services after their initial contracts.

Parloa has named Booking.com, Allianz Partners, IKEA and HealthEquity among its customers. Its platform helps enterprise teams create, test, manage and monitor agents across voice and chat.

The company benefits from selling a complete operating platform rather than only the infrastructure for individual calls. That can make it attractive to businesses with complicated internal systems and strict compliance requirements.

Parloa is still smaller than Sierra on the available revenue figures, but its customer expansion and retention provide stronger evidence of progress than its valuation alone.

Is Decagon catching up with Sierra in AI customer service?

Decagon is growing quickly enough to become a major competitor to Sierra, although Sierra still has roughly twice its publicly estimated annualized revenue.

Sacra estimates that Decagon reached approximately $100 million in annualized revenue in July 2026, up from around $44 million at the end of 2025. That suggests growth of roughly 127% in seven months.

Decagon raised $250 million in Series D financing in January 2026 at a $4.5 billion valuation. Its announced enterprise customers include Avis Budget Group, Block and Deutsche Telekom. We follow who is getting funded in this market, and by whom, well beyond one Series D, in our report on agentic AI.

The competitive overlap with Sierra is clear. Both sell AI agents designed to resolve customer requests through business-system integrations, and both serve organizations that need much more than a conversational interface.

Decagon's growth also illustrates how quickly customer-service agents can expand once large customers begin putting more interactions through them. A contract can start with one workflow and grow as the system handles additional requests.

For now, Sierra has a clear revenue advantage, but Decagon has grown from a much smaller base into a meaningful enterprise rival. The gap is still large enough to give Sierra the lead, while narrowing fast enough to keep the race competitive.

Could ElevenLabs overtake the dedicated AI phone-agent startups?

ElevenLabs has more revenue and financial resources than most phone-agent specialists, but we cannot yet say that its telephone-agent business is larger.

In May 2026, ElevenLabs announced that it had exceeded $500 million in annual recurring revenue, up from $350 million at the end of 2025.

A $300 million employee tender offer announced in September valued the company at $22 billion, double the valuation associated with its February financing.

Those are remarkable figures, but ElevenLabs earns revenue from several products, including speech generation, APIs, conversational agents and other voice technology services.

The company is particularly important because it supplies technology that other developers can use inside their own voice-agent systems. Its commercial opportunity therefore extends beyond selling complete agents directly to businesses.

At the same time, ElevenLabs has built its own conversational-agent platform. That creates direct competition with startups using external speech and language models.

Its growing enterprise relationships include Deutsche Telekom, KPN, Stripe and Klarna. These give ElevenLabs a route into large-scale voice automation without having to build every customer relationship from scratch.

We would currently classify ElevenLabs as the broader voice technology leader rather than the proven leader in dedicated AI telephone agents. Its resources make it one of the most serious long-term threats to the specialists.

Which AI phone-agent startups have raised the most money?

Sierra is the most heavily funded independent enterprise-agent startup in our comparison, while the gap between funding and commercial results is especially striking at Retell.

Sierra has raised approximately $1.58 billion, according to Sacra. Parloa has exceeded $560 million, PolyAI has raised more than $200 million, and Bland has passed $100 million.

Vapi's disclosed total is $72 million, while Retell's company history highlights a $4.6 million seed round. That Retell figure refers to its publicly described early financing and should not be interpreted as a complete, independently audited funding history.

The differences reflect several business strategies. Sierra and Parloa are investing heavily in enterprise sales, implementation and platform development. Vapi and Retell have grown through products that developers can adopt directly.

Funding can help a company expand internationally, improve reliability and support large customers. But the amount invested tells us little about how efficiently the company earns revenue.

The comparison below deliberately separates revenue from financing and valuation. These figures were disclosed or estimated at different points, and annualized revenue is not always the same as contracted ARR.

Startup	Funding disclosed	Latest valuation evidence	Revenue evidence
Sierra	~$1.58B	$15.8B	~$200M ARR estimated
Parloa	$560M+	$3B	$50M+ ARR
PolyAI	$200M+	Not publicly established	Not disclosed
Bland AI	$100M+	Not publicly established	Not disclosed
Vapi	$72M	~$500M	Eight-figure annual run rate reported
Retell AI	$4.6M seed disclosed	Not publicly established	$80M annualized
Decagon	$250M Series D alone	$4.5B	~$100M annualized estimated
ElevenLabs	$500M February Series D alone	$22B tender valuation	$500M+ ARR across products

Are AI phone agents actually saving companies money?

AI phone agents are already producing measurable benefits in selected customer-service operations, particularly where businesses receive more calls than employees can answer.

Fogo de Chão provides one of the clearest cases. PolyAI reported that its voice agent now answers every guest call, with an 88% booking completion rate and 40% loyalty enrollment among guests who discuss the program.

During the initial Mother's Day pilot, the company said restaurants recovered an average of three hours of staff time per location. PolyAI also projected that the system would generate more than 250,000 reservations over twelve months.

A separate case study published by OpenAI described a Parloa implementation in which requests for human assistance declined by 80%. That is a different measure from successful task completion, but it indicates how automation can reduce pressure on contact-center teams.

PolyAI also commissioned a Forrester Total Economic Impact study estimating a 391% return on investment for a modeled customer organization. That result should be understood as a commissioned financial model, rather than an independently observed return across PolyAI's customer base.

What these examples share is a clear operational starting point: missed calls, repetitive work or high human-support demand. Automating those tasks can create measurable savings or additional sales.

The harder question concerns calls involving unusual requests, complaints, identity checks or sensitive decisions. Public case studies still provide limited standardized information on how frequently agents complete those tasks correctly without human intervention.

Are AI phone agents becoming cheap enough for smaller businesses?

AI phone agents are now affordable enough for many smaller businesses to test, but the advertised per-minute price can hide a substantial difference in actual operating costs.

Vapi advertises a $0.05-per-minute platform fee, with speech recognition, language models, voice generation and telephony charged separately. Retell publishes a broader voice-agent price range of approximately $0.07 to $0.31 per minute, depending on the configuration. With fees this small per minute, the real test is which business models win, and we compare them in our agentic AI report.

At 10,000 billable minutes monthly, a business paying $0.10 per minute would spend about $1,000. A more expensive setup at $0.25 would cost $2,500 for the same volume.

At 400,000 minutes monthly, that difference grows to $60,000 every month. So even relatively small changes in cost per minute become important for large call centers.

The cheaper option also needs to resolve calls successfully. A system that repeatedly asks customers to clarify themselves, gives incorrect answers or transfers too many calls to employees can end up costing more despite a lower advertised rate.

These economics create an interesting challenge for Vapi and Retell. As voice models become cheaper, customers may expect lower platform prices too. The companies will have to keep improving their tools, reliability and integrations to justify their margins.

For customers, the useful comparison is ultimately the cost per successfully resolved call, including infrastructure, implementation and human follow-up.

Can bigger tech companies push AI phone-agent startups out of the market?

Established customer-service platforms could take a significant share of the AI phone-agent market because they already sell software to the same companies that startups are trying to win.

NiCE's acquisition of Cognigy is a particularly clear example. Announced in July 2025 and completed in September, the transaction valued Cognigy at approximately $955 million.

NiCE gained conversational AI technology that it could offer through an existing network of more than 25,000 customers. That gives it a distribution advantage over young companies that must win every enterprise account individually.

Salesforce, Genesys and other contact-center software providers also have established customer relationships, data integrations and procurement agreements. They can incorporate AI agents into systems that companies already use.

The specialists still have reasons to exist. Vapi gives developers considerable control over their voice infrastructure. Retell makes it easier to deploy dedicated telephone agents. PolyAI offers direct experience with complicated, multi-location enterprise installations.

Bland's investment in proprietary voice models is another attempt to build something competitors cannot easily reproduce.

We expect the easiest features to become increasingly common across platforms. Answering straightforward calls and generating natural speech are already available from many suppliers. The more difficult work involves complex requests, business integrations, compliance and keeping the system reliable across thousands of conversations.

Those are the areas where specialist startups have the best chance of holding onto customers.

Which AI phone-agent startup is best positioned to stay ahead?

Sierra has the strongest enterprise commercial position today, but Retell and Vapi have particularly attractive positions if businesses increasingly choose to build their own phone agents.

The market could develop in two different directions, and we can already see evidence for both.

Large enterprises often want complete solutions. They need agents that connect to customer records, follow internal policies, complete transactions and work with human employees. Sierra, Decagon, Parloa and PolyAI are well placed to sell that kind of product.

Other businesses prefer to build custom systems around their existing software. Vapi and Retell benefit from that approach because they provide much of the voice infrastructure without requiring customers to adopt an entire customer-service platform.

The largest technology companies introduce another possibility. ElevenLabs can combine speech models with its own agent products, while NiCE and other contact-center incumbents can sell automation to existing customers.

This makes the competitive race harder to predict than the current revenue rankings suggest. The winner in enterprise software may differ from the winner in developer infrastructure.

For Retell, continued revenue expansion and profitability would strengthen its position. For Vapi, converting its developer reach and enterprise traffic into more disclosed recurring revenue would settle a major unanswered question. For Sierra, the challenge is maintaining growth as large customers demand better performance on increasingly complicated interactions.

AI phone agents: which startup is ahead right now?

Sierra leads the broader enterprise AI customer-service market, while Retell AI is our current pick among dedicated AI phone-agent startups. Vapi is its strongest challenger, and PolyAI deserves particular recognition for large enterprise deployments.

Sierra has the clearest evidence of commercial scale, with Sacra estimating roughly $200 million in annual recurring revenue. Its success increasingly includes voice, although the company also operates across digital customer-service channels.

Retell stands out because it has built a substantial phone-focused business with relatively little disclosed early funding. Its reported $80 million annualized revenue gives us a stronger commercial benchmark than traffic or developer registrations alone.

Vapi could overtake it. The company has demonstrated extraordinary telephone usage, attracted a large developer ecosystem and won a major competitive procurement involving Amazon Ring. More precise revenue disclosures would help establish whether its commercial scale matches its technical reach.

PolyAI has taken another route, building a large network of live enterprise deployments, while Bland, Parloa and Decagon remain serious contenders in their respective markets. ElevenLabs has the resources and speech technology to change the competitive picture further.

For now, Sierra is winning the broad enterprise race and Retell leads among dedicated phone-agent platforms. Vapi has the strongest case to challenge Retell, and neither company has locked up the market.