파이버(FVRR) 2026년 3분기 실적 컨퍼런스 콜: AI의 트래픽 재편 속 가이던스 상향

핵심 요약

• 파이버 인터내셔널(Fiverr International Limited)은 2026년 3분기 매출이 가이던스 범위인 8,000만~8,800만 달러를 상회하는 9,090만 달러를 기록했으며, 조정 EBITDA는 1,450만 달러로 16%의 이익률을 달성했다고 발표했다.
• 총매출은 전년 동기 대비 15.8% 감소했다. 마켓플레이스 매출은 5,540만 달러로 24.7% 감소했으나, 서비스 매출이 3,550만 달러로 3.4% 증가하며 일부 상쇄되었다.
• 지난 12개월(TTM) 기준 활성 구매자 수는 240만 명으로 25.8% 감소했지만, 마켓플레이스가 고부가가치 프로젝트 중심으로 전환됨에 따라 구매자당 지출액은 373달러로 12.9% 증가했다.
• 오토DS(AutoDS) 및 파이버의 광범위한 이커머스 서비스가 분기 실적 호조를 이끌었다. 경영진은 8월 말에 시작된 주요 캠페인이 4분기까지 강세를 유지할 것으로 예상된다고 밝혔다.
• 신규 구매자로부터 창출된 AI 기반 총거래액(GMV)은 전분기 대비 약 64% 증가했다. 파이버는 챗GPT(ChatGPT) 및 클로드(Claude) 내에서 사용할 수 있는 API, CLI 및 MCP 연동 기능을 출시했다.
• 파이버는 2026년 전체 매출 및 조정 EBITDA 가이던스를 상향 조정했지만, 경영진은 2027년까지 마켓플레이스 트래픽 압박이 심화되어 조정 EBITDA 마진에 부담을 줄 것으로 예상하고 있다.

주요 재무 데이터

사업 및 운영 실적

마켓플레이스 실적은 검색 기반 트래픽 감소와 저가형 거래 작업의 지속적인 축소로 인해 계속 압박을 받았다. 경영진은 이것이 인재 공급 문제가 아닌 퍼널(funnel) 문제라며, 트래픽 압박이 2027년까지 이어질 것으로 예상된다고 밝혔다.

지난 12개월 기준 완료된 프로젝트 총주문액 중 1,000달러 이상 프로젝트의 비중은 전분기 15%에서 16%로 증가했다. 1,000달러 이상 완료 프로젝트에서 가장 견조한 카테고리는 프로그래밍·테크와 그래픽·디자인이었으며, 두 분야 모두 전년 동기 대비 10%대 중반의 성장률을 기록했다. 그러나 SEO 관련 상단 퍼널(top-of-funnel) 교란으로 인해 3분기 중 성장이 둔화되었다.

서비스 매출은 오토DS의 수혜를 입었다. 경영진은 이러한 실적이 이커머스 소프트웨어, 스토어 소유주를 위한 AI 툴의 성장과 효율적인 마케팅 이니셔티브 덕분이라고 밝혔다. 대형 이커머스 파트너와 함께 8월 말 시작된 캠페인은 예상보다 좋은 성과를 냈으며 경영진은 4분기까지 이어질 것으로 예상하고 있다.

파이버는 고부가가치 및 고신뢰도 작업과 AI 기반 탐색을 중심으로 마켓플레이스의 위치를 재정립하고 있다. 파이버의 지식 그래프(Knowledge Graph)는 약 7만 개의 하드 스킬에 걸쳐 월 330만 건의 스킬 추출을 처리한다. 회사 측은 매칭 시스템을 통해 더 높은 총주문액, 구매자당 더 많은 주문, 더 적은 취소 건수 및 지원 티켓 감소 효과가 나타나고 있다고 밝혔다.

경영진에 따르면 3분기에 출시된 파이버의 AI 헤드헌터인 미라(Mira)는 일반 파이버 대화와 동일한 비율로 소개를 전환시키고 있다. 미라의 중간 주문액은 5배 더 높으며, 최초 매칭 시 취소율은 약 25% 낮다. 파이버는 2027년에 주문 관리 및 모바일을 포함해 고객 여정 전반으로 미라를 확장하는 한편, 헤드헌터 모델에 적합한 고객을 주요 타겟으로 삼을 계획이다.

경영진 가이던스

연간 조정 EBITDA 가이던스의 중간값은 16.7%의 마진율을 의미한다. 경영진은 재량 지출 통제, 내부 AI 효율성 제고 및 규율 있는 마케팅에 힘입어 2027년 영업비용이 전년 대비 감소할 것으로 예상하고 있다.

하지만 회사는 비용 절감 효과가 지속적인 트래픽 및 수요 역풍으로 상쇄될 것으로 예상한다. 따라서 경영진은 2027년 조정 EBITDA 마진에 하방 압력이 있을 것으로 예상하며 4분기 실적 발표와 함께 2027년 전체 가이던스를 제공할 계획이다.

리스크 및 주목할 점

• 경영진은 기존 검색에서 AI 기반 탐색으로의 전환으로 인해 2027년까지 마켓플레이스 트래픽 감소폭이 더욱 가파라질 것으로 예상하고 있다.
• 광고 밀도 상승과 고객 획득 비용 증가로 인해 그동안 검색 중심이었던 파이버의 마케팅 모델 효과가 감소하고 있다.
• 경영진은 확립된 운영 모델이나 업계 표준이 없기 때문에 AI 에이전트를 통한 고객 도입의 시기와 규모를 예측하기 어렵다고 밝혔다.
• AI 에이전트는 플랫폼 충성도보다 결과를 우선시하기 때문에 직접적인 브랜드 친화력을 약화시킬 수 있다. 파이버는 에이전트 기반 채널 내에서 자사의 서비스를 차별화하기 위해 거래 데이터, 매칭 능력, 신뢰도 및 이행 실적에 의존하고 있다.
• 대형 프로젝트 중심으로 비중이 이동함에 따라 구매자당 지출액은 계속 증가할 것으로 예상되지만, 경영진은 마켓플레이스 전반의 트래픽 압박으로 인해 성장 속도가 둔화될 수 있음을 내비쳤다.

애널리스트 Q&A 하이라이트

AI 기반 고객 획득: 경영진은 대형언어모델(LLM)을 통해 유입되는 트래픽의 양은 여전히 적지만, 이용자들이 대체로 보다 명확한 의도를 가지고 유입되기 때문에 품질은 더 높다고 밝혔다. 파이버는 궁극적으로 이용자가 AI 어시스턴트를 벗어나지 않고도 에이전트가 프로젝트 범위를 설정하고, 인재를 선정하며, 주문 및 결제까지 완료할 것으로 기대하고 있다. 이러한 연동의 수익 구조는 파이버의 기존 모델과 동일하게 유지된다.

오토DS의 지속 가능성: 경영진은 오토DS의 성장이 단일 캠페인이 아닌 이커머스 스토어용 소프트웨어, AI 콘텐츠 툴, 마케팅 파트너십 등 다수의 동인에 의해 뒷받침되고 있다고 설명했다. 3분기 실적에 기여한 캠페인은 4분기에도 지속될 것으로 예상된다.

미라 확장: 파이버는 2027년까지 미라를 단계적으로 확장할 계획이다. 경영진은 헤드헌터 방식의 경험이 특정 고객 및 프로젝트에 더 적합하기 때문에 이 제품이 반드시 모든 사용자에게 노출되는 것은 아니라는 점을 강조했다.

자본 배분: 파이버의 최우선 과제는 마켓플레이스 전환 자금 조달과 전략적·재무적 유연성 유지다. 에스티 다돈(Esti Dadon) CFO는 현재 M&A는 고려하고 있지 않다고 밝혔다.

실적 발표 전화회의 전문

---

전체 실적 발표 컨퍼런스 콜 녹취록

경영진 발표

Operator

Good day, and welcome to the Fiverr International Limited Q3 2026 Earnings Conference Call.

[Operator Instructions]

Please note, this event is being recorded. I would now like to turn the conference over to Emily Greenstein, Senior IR Manager. Please go ahead.

Emily Greenstein

Thank you, operator, and good morning, everyone. Thank you for joining us on Fiverr's earnings conference call for the third quarter that ended September 30, 2026.

Joining me on the call today are Micha Kaufman, Founder and CEO, and Esti Dadon, CFO.

Before we start, I would like to remind you that during this call, we may make forward-looking statements and that these statements are based on our current expectations and assumptions as of today, and Fiverr assumes no obligation to update or revise them.

A discussion of some of the important risk factors that could cause actual results to differ materially from any forward-looking statements can be found under the Risk Factors section in Fiverr's most recent Form 20-F and other filings with the SEC.

During this call, we'll be referring to some key performance metrics and non-GAAP financial measures, including adjusted EBITDA, adjusted EBITDA margin, and free cash flow.

Further explanation and a reconciliation of each of the non-GAAP financial measures to the most directly comparable GAAP measures is provided in the earnings release we issued today and our shareholder letter, each of which is available on our website at investors.fiverr.com.

And now I will turn the call over to Micha.

Micha Kaufman

Thank you, Emily. Good morning, everyone, and thank you for joining us.

Let me start with the headline. Q3 came in ahead of our guidance. We delivered revenue of $90.9 million, above the top end of the $80 million to $88 million range we gave you in July.

Adjusted EBITDA was $14.5 million, a 16% margin, also well above our range. The beat was led by our services business, where AutoDS and our e-commerce offering performed better than expected, and the momentum is carrying into Q4, along with tight cost management across the company.

Three months ago, we told you we would run this business with discipline through a hard transition. This quarter, we did exactly that. This quarter is also a good reminder of how we run Fiverr.

We are one business with diversified lines of revenue and profit, and we manage it as one to make the top line as large and fast-growing as possible and as profitable as possible.

Within that, we are working to turn around our marketplace, and that is exactly what the transformation I'm about to walk you through is for.

Esti will walk through the details, but I want to use my time on 2 things: what is changing in how work gets found and how we plan to win in the new world.

What is changing? For 17 years, Fiverr built one of the best performance marketing engines on the Internet on top of search. That world is changing fast.

More and more journeys now start and end inside an AI assistant. Organic clicks are falling across the web and acquisition costs are rising for everyone who depends on them.

We expect that pressure on marketplace traffic to continue into 2027. But here is what has not changed. The work still needs people. AI is automating simple tasks, and the work that remains is more complex and more valuable.

That work needs a human who owns the outcome. Our top talent community remains highly engaged. This year alone, we hosted 62 events with more than 3,500 freelancers.

Most of them are top talent. Put simply, this is not a talent problem. It is a funnel problem, and funnels can be rebuilt. So how are we going to do it? Our North Star is unchanged: to become the most trusted platform for high-value, high-trust work.

Let me show you the progress across our 4 pillars. First, our push upmarket. High-value work now represents 16% of completed projects' gross order amount on a trailing 12-month basis, up from 15% last quarter.

And we continue to see businesses bring us multiphase mission-critical work. For example, a London-based private investment firm that manages institutional real estate assets came to us to build an end-to-end digital operations system.

They assembled a team of senior engineers through Fiverr and worked alongside them directly from start to finish. An outdoor recreation equipment manufacturer is building a real-time 3D design engine, so both consumers and commercial buyers can configure large play installations before they order them.

It has to carry real-world compliance logic, safety zones, material libraries, and parametric constraints. And our freelance team is optimizing close to 300 3D assets to render fast in a browser.

A European packaging and industrial manufacturer is replacing its legacy systems with a single enterprise e-commerce platform that ties together sales, real-time inventory, and fulfillment.

A Fiverr Pro team is working through continuous cycles with the client's own leadership. These are not one-off gigs. They are exactly the engagements we are building Fiverr for.

Second, matching. In an agentic world, an AI agent does not want 20 options. It wants the right one. Our knowledge graph ranks talent on what they actually delivered, not on who got clicked.

It performs 3.3 million skill extractions a month across roughly 70,000 distinct hard skills, and every order makes it sharper. The results are already showing higher gross order value, more orders per buyer, fewer cancellations, and fewer support tickets.

Third, the work platform. In Q3, we launched Mira, our AI headhunter. Mira turns a vague brief into a clear scope and a precise match. Early results are encouraging.

A Mira introduction converts at the same rate as a normal conversation on Fiverr, and the median order value for the headhunter experience is 5x larger.

Among first-time matches, cancellations on Mira work run about 1/4 lower. Every brief makes Mira and the Knowledge Graph smarter.

In 2027, Mira is expected to run across the full journey from the first conversation to order management, including mobile.

Fourth, our growth engines. This is where the new funnel is being built. We launched our first public APIs, CLIs, and MCP integrations with Fiverr, now live inside Claud and ChatGPT.

AI-sourced GMV for first-time buyers grew approximately 64% from last quarter. Where this goes is a full agentic experience. An agent finds the talent, scopes the work, places the order, and pays for it without ever leaving the assistant.

Whether a business arrives through a browser or an AI agent, it should find one place to get serious work done. One leadership update. As you know, Jinjin was planning to relocate to Israel, where our Chief Business Officer role is based.

She decided moving her family isn't the right step for them right now, so she stepped down. We respect that decision and thank her for her contribution.

The search for her successor is well advanced, and we expect to announce our new Chief Business Officer soon. In the meantime, Esti and I are running the business directly.

To take a step back, the search era built Fiverr; the AI era is where we intend to lead. We have the talent, the data, the balance sheet, and the focus to get there.

With that, I'll turn it over to Esti for the financial details.

Esti Dadon

Thank you, Micha. Q3 was a stronger-than-expected quarter where our operational discipline and strategic execution shone.

Both revenue and adjusted EBITDA came in above the guidance range we gave in July, and we are raising our full-year guidance for both revenue and adjusted EBITDA.

Today, I will review our third quarter results, update you on our cost structure and balance sheet, and share our fourth quarter and full-year 2026 guidance, along with some early color on 2027.

Third quarter revenue was $90.9 million, down 15.8% year-over-year, reflecting the continued shift away from low-value transactional work and the search-driven traffic pressure Micha described.

This was partly offset by the increase in services revenue. Adjusted EBITDA was $14.5 million, a 16% margin.

Two things drove that outperformance: a strong quarter in our e-commerce services and the cost discipline we have built into how we run the company, including tight management of discretionary spend, internal AI efficiencies, and a balanced approach to marketing.

Q3 marketplace revenue was $55.4 million, down 24.7% year-over-year, driven by 2.4 million active buyers, $373 in average spend per buyer, and a 28.2% marketplace take rate.

Spend per buyer rose 12.9% year-over-year, driven by an ongoing mix shift across the marketplace from low-value transactional projects to $1,000-plus projects.

The 25.8% decline in active buyers on a TTM basis was mainly due to the ongoing compression in low-value transactional work as well as the continued traffic and demand pressure from shifts in search that continued through Q3.

At the end of Q3, projects at or above $1,000 represented 16% of completed projects' gross order amount on a TTM basis. Programming and tech and graphics and design were the strongest verticals for completed $1,000-plus projects, both up mid-teens year-over-year.

Growth rates in these verticals slowed in Q3 due to top-of-funnel issues related to SEO disruption. Services revenue was $35.5 million, up 3.4% year-over-year and accounted for 39% of total revenue.

The upside was led by AutoDS, where a major e-commerce campaign launched in late August. The campaign is performing extremely well and is expected to continue through the rest of the year.

This is a good example of why we manage Fiverr as a whole, not as separate segments of revenue and profit.

Non-GAAP operating expenses were $62.2 million, down from $66.7 million a year ago and from $65.3 million last quarter, with the largest savings in R&D and sales and marketing.

We will continue to run a lean organization through Q4 and into 2027, and we expect operating expenses to be lower year-over-year in 2027. We ended the quarter with $316.2 million in cash and investments and generated $9.4 million in free cash flow.

Our capital allocation priorities are straightforward. First, funding the transformation that positions our marketplace for how work is found in the AI era.

Second, keeping strategic and financial flexibility in a fast-moving environment. A strong balance sheet is a real advantage right now, and we intend to keep it that way.

Now for guidance. For the full year 2026, we are raising our revenue guidance to $374 million to $380 million, up from $356 million to $372 million last quarter.

We are also raising our adjusted EBITDA guidance to $61 million to $65 million, a 16.7% margin at the midpoint and up from $52 million to $62 million last quarter.

For the fourth quarter, we expect revenue of $80 million to $86 million and adjusted EBITDA of $6 million to $10 million. That outlook includes continued strength from our e-commerce business, offset by the continued decline in marketplace traffic we have discussed.

We are 2 quarters into the transformation. We expect the shift from search to AI-driven discovery to drive a steeper decline in marketplace traffic through 2027.

We will meet that with lower operating expenses and continued discipline. That said, while we anticipate lower year-over-year OpEx in 2027, the reduction is expected to be offset by the ongoing traffic and demand headwinds, resulting in downward pressure on 2027 adjusted EBITDA margin.

We will provide full 2027 guidance on our fourth quarter call. To wrap up, we beat on revenue and profit. We raised the year on both, and we did it while lowering our cost base.

We are running a lean, disciplined company, and we are moving through this transformation with a strong balance sheet and a clear plan.

With that, we will turn the call over to the operator for questions.

Operator

[Operator Instructions]

The first question is from the line of Ron Josey with Citi.

질의응답

Unknown Analyst

This is Jake on for Ron. Micha, I just wanted to double-check on the fourth pillar, the growth engine. You mentioned AI-sourced GMV for first-time buyers grew 64% quarter-over-quarter.

Just talk to us about what you're seeing from buyers who arrived through those channels. And how do you think about the path for AI-sourced work to become a more meaningful share of new buyer acquisition over the next few quarters?

And then I had a follow-up.

Micha Kaufman

Jake, thanks for the question. I will start maybe by saying that we're witnessing something that is no less than an earthquake on the Internet or the web as we know it.

And I've been talking about this for over a year. And I think that we're starting to see this. The Internet, as we've known it for the past 20-plus years, is going away. And it's going to completely change. It's dying.

It used to be that customers searched for things, then clicked on links. Then came to destination platforms or websites, then made a decision or an exploration, and finally placed an order or an option.

All of the layers that I've described right now are going to disappear. And there is a changing in the interest of customers between the discovery layer and the focus on the actual output.

This is changing the complete economics of search. And I think that we've seen that with the latest earnings from Google. And I believe that this is going to have a domino effect on every possible business we know now, including fiber.

No one is immune. There have been businesses that have been built over the past 20 years, and this includes all of the Magnificent 7 as well, but definitely all of the other companies, ourselves included, where right now, what we should be focusing on is that we don't have 20 years to get adjusted to this new reality.

I don't know if we have a year, but all of us need to accelerate and understand that the way customers are going to interact is getting appended. And as a result, I've called out the focus that Fiverr is putting on these new experiences, these new channels.

I've called out the fact that we're live on ChatGPT and Claude and hopefully soon on Gemini as well. And the idea is to provide customers, without leaving the convenience of their agents, the full experience.

We don't expect them to continue using the Internet as they used to. And it's very important that we emphasize the first-principles moats that we've accumulated over the years to make sure that their experience is optimized for the best output.

Now where I call this an earthquake is because unlike before, agents don't have brand loyalty. If you, as a customer, send your agent to purchase something, that agent is not loyal to Amazon or Walmart or your local business. They're loyal to the business that is going to maximize the output.

And we can talk about what it means. But the reality is that everything is changing, and the entire discovery, search, link, ads- all of this is completely changing.

As I've said, I've called it out more than a year ago. And I'm not sure I'm such a prophet, but it's more about the fact that as a marketplace, we're super sensitive to changes.

And this is why we called out that the way we do business, like everybody else, like the booking business or the e-commerce business or the SaaS business, has treated Google Search as the Internet. And this is what we optimized for, which means SEO and SEM.

SEO is going away. Because there's less traffic coming to traditional search and more coming to LLMs, which increases the ad density, which means that ads are now more expensive, which means that their efficiency is going down.

Now, if you think about what's going on in LLMs, and I'm coming back to your question about AI customers that are generating GMV through LLMs, what we're seeing is, yes, at the amount level, and this is prior to us releasing the MCP and API, which just happened.

But prior to that, what we're seeing is we're seeing less traffic coming from LLMs, but we're seeing a higher quality traffic because these are customers that have had experiences with their agents, so they come with a better, more formed intent to us.

Now we don't even expect them to come to us. We expect to bring fiber to them inside their agents and ultimately allow their agents to be autonomous in working with fiber. This is why we think that this is so important.

Obviously, I'm happy to talk about this further, but I'll stop here.

Unknown Analyst

Just pivoting a little bit on services. AutoDS is driving momentum here. Just talk to us about what's driving that e-commerce demand?

How durable do you think it is? And then maybe how you think about services as a growth engine going forward?

Micha Kaufman

Sure. I mentioned in my opening comments; I've sent a clear message. Fiverr thinks about its business as a whole.

Now, some vectors in our business are growing. Make no mistake, AutoDS and the team that joined Fiverr a couple of years ago are killing it. They're doing extremely well.

Yes, we're seeing some weaknesses, and we explain why on our marketplace and what we do to mitigate that. But we think about the business as a whole. And therefore, the opportunities that we have to increase businesses and grow businesses faster and contribute the top line into the bottom line- this is where we double click.

And I recommend our shareholders look at it the same way. It's a whole business with different avenues of opportunities. And those who have an opportunity to grow, we grow as fast as possible.

Operator

The next question is from the line of Bernie McTernan with Needham.

Stefanos Crist

This is Stefanos Crist on for Bernie. I wanted to follow up on AutoDS. I believe there was something similar in the first quarter this year where there was a marketing campaign that drove the beat and that pulled forward revenue.

Is that similar to what's happening this quarter? And if not, can you just walk us through the dynamic there? Like, are revenues supposed to be lumpy? Are they just driven by these campaigns?

Any more color there would be great.

Micha Kaufman

Absolutely. So as I said earlier, AutoDS is growing extremely well. Within our e-commerce business that AutoDS and its team are leading, there are a number of initiatives that are working really well.

So first, their software solution for e-commerce stores is doing extremely well. They've created AI tools that help those store owners generate content that is growing extremely well.

And on top of that, we are very successful, fortunately, at running very efficient marketing and capturing marketing opportunities very efficiently that are also contributing to growth.

So again, it's a mix of different things. It's not relying on just one leg, but actually a multi-vector growth opportunity that the team is capturing very well.

Esti Dadon

Yes. And I'll add that the campaign that we referred to started in late August, and it was very successful and better than expected. And it's a partnership with a large e-commerce player, and we expect that to continue into Q4.

So that's what led us to be able to update our guidance on the top line.

Stefanos Crist

And then on the LLM partnerships, I think you mentioned ChatGPT and Claude.

Can you just talk about the economic model there? What's the take rate? Who owns the customer? And I know it's early, but how do you expect to handle repeat buyers?

Micha Kaufman

Yes. So the economics are unchanged. We are thinking about those integrations as the new top of the funnel, which is, as I've said before at the beginning, very focused on the outcome and very, very friendly to the agent of the customer.

It is not changing the model. It is true that we've been one of the first 100 brands to also get exposure to advertising on ChatGPT, and I've called out some initial information there, but I think it's very early on that model, and it is probably going to be embraced by other LLM providers.

But the main thought around this is we want to meet the customers where they are. And they're shifting from their usual browsing experience and search and clicking links into an outcome-based experience where they anticipate receiving the entire experience without leaving their agent.

And by the way, rightfully so. And that is why we're doing these integrations, these deep integrations with all major foundational models and agents.

Operator

The next question is from the line of Marvin Fong with BTIG.

Marvin Fong

Maybe to start with on Mira. You talked about expanding that in 2027. Should we expect that to apply to all your categories and traffic by then?

Or how is that exposure going to kind of phase in in 2027?

Micha Kaufman

Thanks, Marvin. So the idea behind Mira is the understanding that there are fundamentally 2 ways to experience working with the talent.

One is a discovery layer. And that allows customers to do more on their own when they think about picking the right talent or the right service and the right price and everything.

The other is an approach that you should think of as a headhunter. A headhunter really does a lot of the work of qualifying the 2 sides of a transaction.

For clients, it's qualifying talent. But for talent, it is qualifying that the client that stands across from them is the right client for the services that they offer. This saves both sides a tremendous amount of hassle and time.

Now, headhunting is not necessarily the right experience for everyone. And for us, it's really important to surface Mira for the right customers at the right time.

Now, because we've developed the knowledge graph and we collect a tremendous amount of information about both the clients and, obviously, about the talent, we can do very bespoke types of matching. And this is what Mira really focuses on.

So as we introduce this, we've begun slowly to ensure that the quality is actually getting delivered. And the way we measure it is we measure it by success and success is getting measured by the happiness of the fulfillment that we can deliver.

Once we saw that this was actually outperforming everything else we've created in the space of headhunting or sophisticated matching, we started expanding it across entry points.

Again, we want to be very cautious because if you introduce Mira to the wrong person, it's like introducing a headhunter for someone who doesn't want headhunting; you don't get the right results.

So our expectation over 2027 is continue expanding this offering as we learn and we qualify which clients Mira is great for. And that's exactly what we're doing.

We're seeing some expansion right now. It's very successful, and we'll continue scaling this over time.

Marvin Fong

And a follow-up question on the quarter: the third quarter, the spend per buyer, it remains very strong here. But on a dollar basis, I guess, sequentially, it rose $5.

That's maybe a bit less than in recent quarters, and I know you called out the growth in the higher-value projects, but could you just square those 2 metrics?

Are we seeing some weakness somewhere else in the spend per buyer metrics?

Esti Dadon

Yes. Thanks, Marvin. So spend per buyer is growing because we're seeing higher 100-plus projects. We said now it's 16% of our GOA.

But that being said, the traffic headwinds that Micha described are affecting the entire marketplace. So what we're seeing is, and what we're expecting is, the spend per buyer to continue to grow.

However, there would be some effect on the entire transactions, so it would be on a slower base.

Operator

The next question is from the line of Jason Holstein with Oppenheimer.

Unknown Analyst

This is Julian on for Jason. You've outlined a number of initiatives to adapt to how AI is changing the top of the funnel.

How should we think about those efforts translating into improved acquisition trends in 2027? And then I have a follow-up.

Micha Kaufman

Julian, thanks for the question. The answer is it's very hard to tell. When you think about AI integrations, there is no blueprint for how this is done. There is no best practice on how this is done.

As I've said at the beginning, agents do not have brand loyalty. They don't care. So you need to optimize for things that they do care about. I'm not going to go into the details for obvious reasons, but it's uncharted territory, much like doing SEO 20-something years ago was uncharted.

So it's very hard to extrapolate what the implications of this on the top funnel is going to be. Now, as I've also said, the traffic that we're seeing coming from agents is increasing.

The business is increasing, and the quality is increasing. And building on that, we're making it much simpler for those agents to actually perform actions on behalf of the clients, their owners on fiber.

We are full steam ahead on this. So whenever I have something specific, or at least solid enough to make some projections, I promise to share those with the market.

Unknown Analyst

And then I know you guys have already discussed AI agents a lot during this call, but could you try to like size what the opportunity here is for the top of the funnel?

And then over what time frame do you think this could really grow? Is this just like a '28, 2030 type of thing? Or is this more imminent?

Micha Kaufman

Well, the necessary speed is yesterday. The horizon is harder to anticipate. Now look, I've called out this domino effect. And it's actually much more complex than we can get into in such a short call.

But essentially, it is changing pretty much everything when you think about the movement from top of funnel to the discovery or the destination experiences.

Even if you think about things like booking a hotel or shopping for something. Everything is changing to the outcome, not the experience that gets you there.

The same applies to software. You don't care as much about the software that is doing the job. You are only focused on the output regardless of what happens behind the scenes.

So this is going to change the software, including the enterprise software space as well. Now, the rates of change, unlike 20 years ago, are extremely fast, which means that the ability to offset these changes is very complex.

I'll give you an example. Think about Microsoft and their dominance in enterprise software. Now, if they put a lot of emphasis on Copilot as a channel for customers, and Copilot can actually deliver the output, then it's going to be harder to sell enterprise software because you care less about the software; you care more about what the software can output.

Can you offset by crazy growth on Copilot on the decrease on enterprise software decline or potential decline? I don't know. Each brand would have to figure out how to do this.

So the urgency of doing this is immediate. The horizon of how fast customers are going to embrace these new types of experiences and funnels is unknown, but the assumption is it's going to happen very fast.

And again, you see Meta with Muse, and you see Instinct. And then you see the announcement of Anthropic with Duo, and you see ChatGPT with Dots. And there's a long line of agents that should do things on behalf of their users.

This is going to happen extremely, extremely fast. That is the assumption, which makes it very urgent to act on and realize that the entire ecosystem is changing.

And definitely, by the way, the ad space. Please again, agents don't care about ads. They don't click on ads. The click to destination has 0 meaning; they want to get to the output, to the result.

So it sounds bombastic, but again, I've been talking about this for over a year, and it is happening super, super fast as we've just seen in the past 2, 3 days.

Operator

The next question is from the line of Matt Condon with Citizens.

Matthew Condon

My first one: can you talk about what you're currently seeing as far as the SEO disruption?

My point is, have we seen any sort of stabilization after an impact? Or are you continuing to see trends worsen here for SEO? And then, Micha, my second question: I know we've put on AI agents a bunch, but just another follow-up here.

Just as they continue to handle the discovery and purchase process for buyers, how are you thinking about the potential implications for Fiverr's direct customer relationships, as buyers are increasingly transacting with third-party agents rather than going to Fiverr directly?

Does that create a risk of losing brand affinity and customer loyalty over time and change the marketplace into an even more transactional type of marketplace versus something that people come back to time and time again?

Micha Kaufman

Thanks, Mat. So on SEO, traditional SEO, or the impact of SEO, is decreasing to the point where it's going to be dead very, very soon.

And the reason is that when you have less traffic coming to traditional search, or the traffic that comes to traditional search wants the Gemini component, wants the AI component rather than the links, there's a need to increase ad density. I think we heard that coming from the largest share search engine.

When there is very high ad density, there's no space for organic results. And so the impact of SEO is likely to decrease to the point where it's no longer important.

Now, we used to think I say used to think this was just 6 or 12 months ago that GEO was going to replace SEO, which means generative engine optimization.

But I don't think that this is the case at all because, again, what agents want to do is to provide outcomes for their users. They don't want to replace search by having search done within their LLMs because customers are not interested in getting the search inside LLMs and then going to destinations.

They want everything in the same place. And therefore, the integrations need to be much tighter and the understanding that the agent is actually doing the work on behalf of the customer.

You asked a question about brand. And I already think that I talked about this. I think that brand loyalty is changing. And yes, it is a very high risk because agents don't care and they don't have brand loyalty.

Therefore, you should surface the differentiating layer into the agent layer. Luckily, over the past 17 years, we've built a long list of these moats that I kept talking about in that context. And sure, it's going to be harder because it's not going to be dependent on brand loyalty.

But if you use your asset correctly, then you are likely to get the output chosen by the agent with your platform. I think all brands have to play that game, understanding that the powers are changing in this ecosystem.

But we started, and we're doing this, and I'm confident that the assets, the data assets, the fact that we have so much nuanced transactional understanding, the fact that we know how to guide how a project should be managed and orchestrated, the fact that we know where the importance of human judgment is within that ecosystem, the trust, the quality, the reliable fulfillment- all of these are increasingly important sources of differentiation.

And so we believe that this transition could happen significantly faster than previous transformations on the Internet. And this is why it's critical to play in that game as aggressively as possible, as fast as possible.

Operator

We'll take our last question from Brianna Kanju with UBS.

Unknown Analyst

This is Brianna Kanju on for Josh Chan. A couple of questions on my end. First, can you talk through how Marketplace trended in Q3 versus expectations?

And a follow-up to that, when you say steeper decline in marketplace, does it mean steeper than expected even 3 months ago? And can you go into more detail on why that may be?

Esti Dadon

Yes. So in the last quarter, we called out the traffic headwinds, and we saw those coming into effect in Q3.

So it was within our expectation, our expectation first-quarter range, and it was there. What we are seeing is continuing in Q4, and we expect that traffic deterioration to continue and to increase by the end of the year and also towards 2027.

And that's the reason we're investing in all the things that Micha just talked about: opening these new traffic channels, the transformation that we are undertaking, and upgrading our marketplace in order to support high-value transactions. So all of these are actually being done in order to solve those trends that we are seeing.

Unknown Analyst

And then I guess a follow-up. On capital allocation, I know a lot of investments being made on the platform, but curious if more M&A is appealing at all right now, for example, like with AutoDS?

Esti Dadon

M&A is not on the table now. Our focus is investing in the transformation.

Operator

This concludes our question-and-answer session. I would like to turn the conference back over to management for any closing remarks.

Micha Kaufman

Thank you, Maria, for moderating today's call, and thank you to everyone who dialed in. Wishing you all a great day, and look forward to chatting with you later.

Operator

The conference has now concluded. Thank you for attending today's presentation. You may now disconnect.