Wonderful Hits $5 Billion as Salesforce Joins the Round

Wonderful raised $550 million at a $5 billion valuation on September 2, 2026, and Salesforce was the only new investor in the Series C.

The Amsterdam-based AI company had been priced at $2 billion in March. Insight Partners led again, and the money is earmarked for on-site engineers who sit inside customer firms, not for a bigger chatbot demo.

A Rival Lands on Salesforce’s Cap Table

Wonderful closed a $550 million Series C led by Insight Partners, with Index Ventures, IVP, Vine Ventures, 9Yards, and Bessemer Venture Partners returning. Salesforce wrote the only new check.

That name is doing a lot of work. Salesforce already sells Agentforce, a line it has put above $1.5 billion in annual recurring revenue, and it paid $3.6 billion in June 2026 for Fin, the former Intercom agent platform. It is now on the cap table of a 20-month-old company that tells buyers they can take their agents and leave.

Wonderful’s own run rate is an estimated $70 million, with a target of more than $100 million by the end of 2026. Against a $5 billion price, that is a 50-times multiple on the year-end figure, accepted in cash by the same class of software buyers Salesforce lives off.

Jeff Horing, co-founder and managing director of Insight Partners, framed the round as a bet on a shared layer rather than another departmental tool. Insight lists more than $90 billion in regulatory assets under management as of December 31, 2025, and it has now led Wonderful’s last two primaries.

Many companies are applying AI to individual departments or use cases. Wonderful is building the operating layer that allows enterprises to scale AI across the entire organization. We’ve watched the team execute on that vision, across dozens of markets, with highly successful early deployments to production at some of the world’s largest enterprises.

Jeff Horing, Co-founder and Managing Director, Insight Partners

A Salesforce survey of 2,025 agentic AI leaders found an eight-month average agent ROI for firms already in production, with a 53 percent employee adoption rate and a 29 percent lift in customer satisfaction among that group. Wonderful’s pitch is that the bottleneck is not the model. It is getting a first live workflow in, then spreading it without rebuilding the stack each time.

What the $550 Million Is For

Bar Winkler, Wonderful’s chief executive and co-founder, said the firm went back to investors because demand is high and slowing down would mean missing growth. Roey Lalazar, the chief technology officer and co-founder, said customers can adopt pieces of the platform, pick models per workload, and keep ownership of what they build.

The company put the same brief on X the morning of the round, in the same words it used with the press.

https://x.com/wonderful_ai/status/2095125312929464388

That post is a hiring document wearing a funding headline. The product only works if people sit at the customer site, speak the local language, and wire agents into systems that were never designed for them. Software does not travel that last mile on its own, which is why the raise is being spent on bodies as much as code.

WHERE THE NEW CAPITAL GOES

• Product work: Build out the Wonderful AI OS so agents, workflows, and in-house apps share context, integrations, and controls.
• On-site teams: Grow forward-deployed engineers who ship a first use case, then hand the keys to the customer’s own staff.
• The hiring raid: Bring in hundreds of engineers, on-site specialists, and former founders across more than 35 markets.
• Any-cloud install: Keep the stack able to run in public cloud or on-premise, so security teams are not asked to move the data.

Winkler’s longer version of the pitch is that a shared AI layer stops a new kind of sprawl. If every department buys its own agent, the firm recreates the old software mess with a new label. The operating system, in his telling, is the thing that compounds as more of the company leans on it.

Bank Hapoalim Shipped an Agent in Three Weeks

The cleanest picture of that model is not a slide. It is a savings campaign at Bank Hapoalim, a $35 billion-class retail bank with more than 250 branches and a multilingual customer base.

The bank picked a slice of about 100,000 people with no active savings account, loan, or mortgage, and offered a first deposit of about $150. It assigned a data engineer who had never built an agent. Wonderful’s on-site team had already wired identity, eligibility, and messaging tools for earlier jobs. The bank engineer reused 15 of those tools, added a document store of campaign facts, and had a live agent in three weeks.

THE HAPOALIM BUILD IN FIGURES

• Time to live: A production agent in three weeks, owned by a bank engineer with no prior agent work.
• What it does: Explains the offer, checks ID with a one-time passcode, tests eligibility on live bank systems, and confirms enrollment.
• Load: Aimed at the full 100,000-person slice, with about 40,000 interactions a month expected.
• Voice result: An 88 percent containment rate, across voice and chat, in two languages, with SMS, human handoff, and a survey built in.

The point the case makes is the transfer, not the campaign. The bank did not want a vendor forever in the loop. After the first agents, it had a staffer who could build the next one. Wonderful’s own careers page hires deployment strategists on customer profit-and-loss outcomes, which is the same idea with a job title on it.

In Israel the customer list now also includes Maccabi, Israel Electric Corporation, Bezeq, Bank Leumi, and several card firms, on top of earlier names such as Discount Bank and Menora Mivtachim. The pattern is regulated, non-English, high-volume work, the opposite of a Silicon Valley pilot that never leaves the lab.

Amsterdam on the Letterhead, Israel in the Staff

The press release is datelined Amsterdam. About half of the 650 people sit in Israel, and the Israeli development and implementation centre is slated to triple over the next year. Winkler previously built Approve.com, which Tipalti bought in 2021, and spent time at ironSource. Lalazar previously built Kaps, an AI localisation tool. They founded Wonderful in early 2025 in Tel Aviv, then put the parent letterhead in the Netherlands.

That split is not a trivia line. European buyers, and European law, are allergic to a single American stack owning the operational layer of a bank or a hospital. Lalazar wrote the company’s answer in June, before this round, in an essay that still sits on the blog: export every agent and config, publish a Swagger file with hundreds of endpoints, run headless, and even ship a Python connector that turns Wonderful agents into Azure agents.

Leaving Wonderful should be easy. Every agent, skill, tool, governance config and app can be exported. Individually, in bulk, through our UI, or an API.

Roey Lalazar, CTO and co-founder, Wonderful blog, June 10, 2026

He put the fear in plain words. If a CRM is bad, a firm can migrate it in three years. If an AI platform becomes the way the company runs, getting trapped there is an existential problem. Openness, in that essay, is also a commercial constraint: the firm says it cannot coast on switching costs, because a customer can run a script and leave.

The EU rules on switching providers under the Data Act, Regulation (EU) 2023/2854, move the same idea from a feature list into law for cloud-style data processing. Switching charges are due to fall away entirely from January 12, 2027, after an earlier ban on penalties for changing provider. Wonderful’s Amsterdam base and its easy-exit copy are aligned with that clock, whether or not a given customer ever actually walks.

Lalazar’s Series C quote stays inside that frame. The AI OS is modular, model-agnostic, and meant to sit on systems the customer already owns. Salesforce, on the same cap table, sells a suite whose gravity runs the other way. The article does not have a product-partnership announcement from either side. The check is still a fact.

$170 Million in Early Employee Sales

The $550 million is the primary. Investors also bought $170 million of stock from employees and early angels at the same $5 billion price, a cash-out less than two years after founding. A smaller $15 million secondary sat beside the November 2025 Series A. People inside the company are being allowed to sell on the way up, not after an IPO that may never come on a tidy timetable.

Headcount moved with the price. Wonderful had 350 people at the March round and 650 by September 2, a gain of 300 in six months, with a public target of 1,000 by the end of 2026.

THE DISCLOSED PRIMARY ROUNDS

Those four primaries add up to $834 million. The company says it has raised more than $800 million in total, which matches that path. Annual recurring revenue was $7 million at the end of 2025 and $18 million in the first quarter of 2026, before the estimated $70 million run rate attached to this round. The slope is steep. So is the price.

Winkler told interviewers the firm came back to market because organisations want one system in which to build what they need in AI. That is a land-grab sentence. It also explains why Insight was willing to re-lead so fast, and why a CRM vendor that already sells agents would want a seat rather than watch from outside.

AI Took 60 Percent of Europe’s Checks

Wonderful is incorporated in the Dutch story as a European scale-up and in the Israeli story as a Tel Aviv company. Both can be true at once. The money around it is the European pattern in miniature: fewer companies getting paid, and the ones that do getting paid a lot.

A mid-year tally of European tech deals put funding at €44.1 billion (about $50.6 billion) in the first half of 2026, across just over 1,740 rounds, the lowest half-year deal count since 2020. AI took about 60 percent of that capital. Seed-stage deal counts in the first quarter were down 44 percent year on year, with early-stage volume down 30 percent, while late-stage volume held up.

THE CONCENTRATION IN ONE SCREEN

• H1 2026 total: €44.1 billion into European tech, with deal count at a six-year low.
• AI’s share: About 60 percent of the money, not 60 percent of the companies.
• Early-stage squeeze: Seed deals down 44 percent in the first quarter, Series A volume down 30 percent.
• This round’s place: A $550 million Series C into a 20-month-old firm, six months after a $150 million round at $2 billion.

That is what “fewer deals, bigger bets” looks like when it lands on a single cap table. The losers are not theoretical. They are the seed-stage agent shops in London, Berlin, Paris, and Tel Aviv that now recruit against a firm that can pay, visa, and place engineers in more than 35 markets, and that has already let staff sell stock.

Wonderful’s first commercial edge was non-English customer-service agents, customised market by market, with engineering pods on the ground. The Series C language has moved up a layer, to an operating system that coordinates agents, workflows, and apps. The go-to-market has not moved. It is still people on site, in the customer’s language, on the customer’s stack.

The Hiring Target Is 1,000 People

Six months ago the company told reporters it expected about 900 people by year end. The figure attached to this round is 1,000, with the Israel centre set to triple and deployment teams to grow in every region where it already has a foothold. The official post called the targets “cracked engineers, FDEs, and former founders,” which is a culture signal as much as a headcount plan.

Speed is the metric this market is now using instead of the old unicorn label, and it is also the risk. A $5 billion price on an estimated $70 million run rate only holds if those new hires keep turning first use cases into production systems that spread inside the account. If the on-site model stalls, the multiple has nowhere to hide.

Salesforce has a seat, Insight has led twice in six months, and employees have already taken $170 million off the table. The next test is whether 350 more people, dropped into banks and telcos from Amsterdam and Tel Aviv, can make an AI operating system look as boring and necessary as the cloud layer Winkler keeps comparing it to.