Japan recorded 10,505 corporate bankruptcies in fiscal 2025, a 3.6% rise according to Tokyo Shoko Research data reported by Kyodo News, and about 90% of them were firms with fewer than 10 employees. The service sector had more failures than any other industry, with 3,585 cases. Behind that number is a scene I know well: a small business owner answering customer messages at 11pm, three apps open, nobody else on staff to help. If you’re comparing a marketing automation agency with an AI agency in Tokyo, that’s the real question you’re answering. It isn’t about which software to buy. It’s about who, or what, picks up the work your missing staff used to do.

An AI agency in Tokyo builds autonomous AI agents that read unstructured messages, reason over context and complete multi-step tasks such as answering, qualifying and routing inquiries. A marketing automation agency configures rule-based workflows, like scheduled emails and lead scoring, that fire identical actions from fixed triggers. Automation runs predictable campaigns. AI agents handle judgment calls at scale.

Why do so many Tokyo SMEs abandon their first AI tool?

They abandon it because the tool sat on top of weak infrastructure. A Research Institute of Economy, Trade and Industry (RIETI) study analyzed actual payment records for generative AI services from 2022 to 2025 across roughly 87,000 Japanese SMEs. It found that first adoption happened almost simultaneously across industries, with a sharp jump in early 2023 after major model releases. What separated the sectors afterwards was industry structure and the digital tools companies already used, and sectors with stronger digital infrastructure kept using AI at higher rates.

That one finding explains a pattern I see constantly in Tokyo. A business signs up for a chatbot or an AI pilot, uses it for three months, then quietly goes back to spreadsheets and static email blasts. The tools were never the problem. The architecture underneath them was. RIETI’s follow-up column notes that the sector gaps appear whether or not a business depends on face-to-face service, which points to differences in core business tasks and digital infrastructure. In other words, a ramen shop and an accounting firm can fail at AI for the same reason.

What is the technical difference between a marketing automation agency and an AI agency?

The difference is architectural, not cosmetic. A deterministic workflow is a linear sequence where a given trigger always produces the same pre-programmed action, with no interpretation of context. That is how tools like HubSpot workflows, Marketo and basic Zapier automations operate. AWS’s executive guide to AI agents versus automation explains that agents act autonomously, reasoning and adapting to changing inputs, while non-agent systems stick to predefined paths even when advanced AI powers them.

In practice, a marketing automation agency configures trigger-action sequences inside a SaaS platform. An AI agency designs the reasoning layer itself, then connects it to the systems a business already runs, usually through custom AI agents built around one specific business process.

What is changing in 2026, and why does it matter now?

Three shifts are happening at the same time, and together they make your choice of partner more important than your choice of platform.

Is “agent washing” a real problem?

Yes, and it is widespread. Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027 because of rising costs, unclear business value or inadequate risk controls. Gartner also calls out “agent washing,” where vendors rebrand AI assistants, RPA tools and chatbots as agentic AI without real autonomy, and it estimates that only about 130 of the thousands of vendors claiming agentic capabilities actually deliver them. Gartner analyst Anushree Verma described most current agentic projects as “mostly driven by hype and are often misapplied.”

For a Tokyo SME, the translation is blunt: a lot of what is being sold to you as an “AI agent” is a decision tree wearing a nicer jacket.

Are most companies actually using AI agents yet?

Not at scale. McKinsey’s State of AI 2025 survey found that 88% of respondents use AI regularly in at least one business function, up from 78% in 2024. But only 23% say their organizations are scaling an agentic AI system somewhere in the business, another 39% are experimenting, and in any single business function no more than 10% report scaling agents. The window to get ahead of competitors is still open, but it won’t stay open long. Gartner expects at least 15% of day-to-day work decisions to be made autonomously by agentic AI by 2028, compared with 0% in 2024, and expects 33% of enterprise software applications to include agentic AI by then.

Are marketing automation platforms turning into AI agent platforms?

They are, and this is the shift most “automation vs. AI” articles miss. According to CX Today’s coverage of HubSpot’s Q1 2026 earnings, HubSpot’s AI Customer Agent now resolves 70% of support conversations autonomously, up from roughly one in five a year earlier. HubSpot also moved the Customer Agent to outcome-based pricing in April 2026, at roughly $0.50 per resolved conversation, as Macha’s Breeze guide details.

So the line between “automation vendor” and “AI vendor” is blurring fast. That makes my point stronger, not weaker. When every platform claims to have agents, the deciding factor becomes the people who scope the problem, connect the systems, write the guardrails and test the agent against your real customers. Software is becoming a commodity. Judgment is not.

Which global AI agencies are setting the pace?

The world’s largest agency groups have spent the last two years rebuilding themselves around AI agents. Watching what they do tells you where the whole market is heading.

WPP launched Agent Hub inside its WPP Open platform in January 2026. AdNews reported that the hub gives clients and WPP’s 100,000-person workforce a set of “Super Agents,” including brand analytics, behavioural science and creative ideation agents. Fortune reported that more than 85,000 of WPP’s 108,000 employees use WPP Open monthly, and that its staff had created over 75,000 AI agents by the end of 2025.

Publicis Groupe put AI at the center of its organization with CoreAI. MarTech reported a €300 million investment to turn the holding company into what it calls an “intelligent system,” unifying data from 2.3 billion consumer profiles with its existing Marcel AI platform. 

Accenture built its AI Refinery platform for enterprise agents and, more recently, went after smaller companies. On July 7, 2026, Accenture announced that its new Accenture Edge business and Google Cloud would offer a suite of agentic solutions for mid-market companies. 

Dentsu Japan, closer to home, is the clearest signal for Tokyo businesses. In May 2026 it announced “AI for Growth 3.0,” moving into full implementation with a platform of 10 specialized marketing AI agents, backed by a dentsu Japan AI Center of roughly 1,000 people launched in July 2025. In February 2026, dentsu Japan also began a strategic partnership with OpenAI to push AI further into corporate marketing in Japan.

What do the giants’ moves mean for a 10-person business in Tokyo?

They confirm the direction and expose a gap. Accenture’s own description says Accenture Edge serves mid-market companies with annual revenues between $300 million and $3 billion. Read that again. In Accenture’s world, “mid-market” starts at roughly ¥45 billion a year in revenue. Meanwhile, nine in ten of the companies going bankrupt in Japan have fewer than 10 employees. 

Every major agency now agrees AI agents are the future of how businesses operate. Almost none of them is structured to build one for a kimono rental shop in Asakusa, a dental clinic in Setagaya or a kominka inn in Nagano. That’s not a criticism. Their economics simply don’t work at that size. It does mean small Japanese businesses need a different kind of partner: one that works at their scale and speaks their customers’ languages.

Can a marketing automation agency solve Tokyo’s labor shortage, or only run campaigns?

A marketing automation agency cannot resolve a labor shortage, because its tools run pre-written campaign logic, not the judgment-based tasks that are actually understaffed. Jiji Press reported that the government’s 2026 White Paper on Small and Medium Enterprises highlights the challenge of securing staff as labor shortages worsen, urges SMEs to use AI, and warns that “maintaining the status quo is the greatest risk” for management today. In fiscal 2025 alone, 442 Japanese bankruptcies were attributed to labor shortages, with rising labor costs behind 195 of them and a lack of job applicants behind 139. EXCLUSIVE: Japan Govt Wants Small Firms to Enhance Biz Literacy

This is the gap autonomous AI agents are built to close. Not another campaign scheduler, but a system that absorbs the inquiry handling, qualification and routing that used to need a staffed front desk or a dedicated sales inbox.

What happened when I built an agent for my own business?

My friend’s photography business gets inquiries from tourists in the US, Latin America and Europe, mostly through WhatsApp, often while he’s on a shoot in Asakusa or asleep in Meguro. His average reply time was over 15 hours. In tourism that’s fatal, because the client has already booked someone else by breakfast.

The WhatsApp agent Ongito built cut response time from over 15 hours to under 30 seconds. No rule-based sequence could have done that, because no fixed trigger can read “hi, we’re a family of four arriving Thursday, is the kimono included and can you shoot near Senso-ji at sunrise?” and answer correctly. That message contains a date, a group size, a package question and a location preference, all in one sentence. A workflow sees one unknown string. An agent sees four things to do.

When does an SME outgrow rule-based tools like HubSpot?

An SME outgrows HubSpot workflows or Marketo the moment its processes have to interpret unstructured input instead of reacting to a fixed field or click. A deterministic system can email a lead after a form submission. It cannot read a free-text LINE message in Japanese, an English email and a Spanish WhatsApp voice note, figure out what each customer actually wants and route it correctly. That needs probabilistic reasoning, not static conditions.

Asking a classic automation workflow to handle a LINE message written in casual Kansai-ben is like asking a vending machine to recommend a good izakaya. It will give you something, with total confidence, and it will be a can of coffee.

Businesses that juggle walk-in inquiries across LINE and WhatsApp at the same time usually hit this wall first, since no single automation platform was built to unify both channels with contextual judgment.

How do costs compare between SaaS retainers and an AI agency in Tokyo?

SaaS marketing automation often comes with mandatory onboarding fees before the monthly software cost even starts. HubSpot’s own Marketing Hub pricing guide lists a $3,000 one-time onboarding fee for Professional and $7,000 for Enterprise, with Professional at $800 per month on annual billing and Enterprise at $3,600 per month. Contact tiers add another layer: a Professional plan that goes from 2,000 to 2,001 contacts jumps to the next tier and adds roughly $850 a month.

Custom AI infrastructure is scoped per function instead of per contact tier, which avoids the pricing cliffs that come with a growing contact list. Here’s Ongito’s published pricing for comparison:

HubSpot’s $7,000 Enterprise onboarding fee alone is higher than the setup cost of any single Ongito service. The structural difference is that an AI agency’s pricing attaches to the task performed, not to how many contacts sit in your database. To be fair to HubSpot, its move to outcome-based agent pricing shows the whole industry is heading the same way. The question for a small business is whether it needs an enterprise platform to get there, or just one well-built agent.

How do AI agencies handle Japanese-language data that automation tools miss?

AI agencies build integration layers that read and act on unstructured Japanese input, something standard connectors in Western marketing automation suites were never designed to parse. Think of keigo-heavy business emails, handwritten-style LINE messages, katakana product names typed three different ways, or a customer who switches between Japanese and English mid-sentence.

This connects directly to the RIETI finding: sustained AI adoption depends on existing digital infrastructure. An SME running legacy accounting software, a paper reservation book or a disconnected point-of-sale system needs an agency that scopes the integration work itself, rather than assuming a plugin will just work. A review and reputation system is a concrete example: it has to read multilingual Google reviews, understand whether a three-star review in Japanese is polite disappointment or a real complaint, and respond with judgment rather than a fixed template.

Why does working with people who understand your business matter more than the tool?

Because the hard part of AI agents was never the model. It’s the workflow around it. Gartner’s own guidance is to pursue agentic AI only where it delivers clear value or ROI, and it notes that rethinking workflows from the ground up is often the ideal route to a successful implementation. McKinsey’s survey points the same way: the organizations getting real value redesign how work gets done instead of bolting AI onto old processes.

Here’s my point of view after almost two decades in marketing: the person who builds your agent needs to understand your customers before they understand your tech stack. They need to know that a tourist asking “is it raining in Kyoto next week?” is really asking whether to reschedule, and that a Japanese client who writes “検討します” probably isn’t coming back. No platform ships with that knowledge. People bring it.

How to choose an AI agency in Tokyo: 6 questions to ask before you sign

1. Can you show me a live agent you built, not a demo? Ask to message it yourself. Real agents handle messy questions. Demos handle rehearsed ones.
   
   
2. What happens when the agent doesn’t know the answer? A good agency designs a clear handoff to a human. A weak one lets the agent guess.
   
   
3. Which of my existing systems will you connect to? If they can’t name your booking tool, POS or CRM in the first meeting, they’re selling a template.
   
   
4. Does it work in Japanese and in my customers’ other languages? Test it with real past messages from your inbox.
   
   
5. Are you pricing per contact, per seat, per function or per outcome? Know what makes the bill go up before it does.
   
   
6. What will we measure in the first 30 days? Response time, resolved inquiries and booked leads are good answers. “Engagement” alone is not.

What does a working AI agent deployment actually look like?

A working deployment follows a short, repeatable sequence instead of a sprawling campaign build:

1. Audit the SME’s existing inquiry channels, whether WhatsApp, LINE or web chat, along with its current software.
   
   
2. Pick the single highest-friction task, such as inquiry response time or lead qualification.
   
   
3. Build and deploy the autonomous agent against that one task first, rather than automating an entire funnel at once.
   
   
4. Connect the agent to a custom CRM dashboard so results are visible and measurable.
   
   
5. Expand scope once the first deployment proves itself, following the same pattern as Ongito’s chatbot deployment timeline.

Ongito’s real estate client, saw a 300% increase in leads and a 400% increase in time on page after this kind of staged deployment. That came not from a broader marketing campaign, but from a system that engaged visitors with contextual responses instead of static forms. A Japanese kominka inquiry agent in Nagano follows the same logic: one specific, high-friction inquiry process, handled by reasoning instead of a scripted sequence.

Common mistakes Tokyo SMEs make when choosing between automation and AI

Treating “marketing automation” and “AI” as the same thing. Most agency content uses the words as synonyms instead of naming the architectural split between rule-based triggers and probabilistic reasoning. If a vendor can’t explain that split clearly, they probably can’t build across it.

Buying an “agent” that is really a chatbot. Given Gartner’s estimate that only around 130 vendors offer genuine agentic capabilities, ask what the system can actually decide and do on its own, beyond answering FAQs.

Importing Western case studies wholesale. Most automation success stories are built around US lead-nurturing funnels and ignore Japanese realities like LINE-first communication, keigo and an acute labor shortage.

Signing a SaaS contract without modeling the full cost. Onboarding fees, seat costs and contact-tier jumps often only show up months later.

Automating everything at once. The projects that get canceled usually start too broad. The ones that survive start with one task and prove it.

Assuming any AI vendor can read Japanese business data out of the box. In practice it usually takes scoped integration work tied to your existing systems, which is exactly what an infrastructure audit is designed to catch early.

Why does this distinction matter for Japanese SMEs specifically?

It matters because SMEs are the Japanese economy. According to the World Economic Forum, Japan’s more than 3.36 million SMEs make up 99.7% of all businesses and roughly 70% of total employment.

The government is also putting real money behind their modernization. The OECD reports that Japan’s SME Productivity Revolution Programme had an annual budget of about JPY 340 billion in fiscal 2024, with typical grants of JPY 500,000 to JPY 2,500,000 per project, and a further JPY 340 billion supplementary budget was announced in December 2025.

Yet adoption is still early. A Rakuten survey of Japanese SMEs found that only 16% currently use AI, with the service industry ahead of other sectors at 21%. Among non-users, 41% couldn’t identify specific ways to use AI and 40% were unaware of its potential benefits. That’s the real opportunity. The barrier for most small businesses isn’t budget or willingness. It’s knowing where AI fits, and having someone who can build it properly.

Frequently Asked Questions

Question: What does an AI agency in Tokyo actually do?
Answer: An AI agency in Tokyo designs, builds and maintains autonomous AI agents that handle judgment-based work such as answering inquiries, qualifying leads and routing requests across channels like LINE, WhatsApp and web chat. Unlike a marketing automation agency, it builds the reasoning layer itself and connects it to the systems the business already uses.

Question: Is marketing automation still worth it for small businesses in Japan?
Answer: Yes, for predictable work. Scheduled newsletters, follow-up sequences and lead scoring based on fixed fields are still handled well and affordably by rule-based automation. The limits appear when the work requires reading free-text messages or making judgment calls, which is where AI agents take over.

Question: How much does it cost to hire an AI agency for an SME?
Answer: Costs depend on scope. Ongito’s published pricing starts at $1,400 setup and $420 per month for a review and reputation system, while an AI chatbot agent costs $2,100 setup and $560 per month. For comparison, HubSpot’s Marketing Hub Enterprise requires a one-time $7,000 onboarding fee before any monthly software cost.

Question: Why do so many agentic AI projects fail?
Answer: Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value or weak risk controls. Most failures come from starting too broad, choosing vendors that rebrand old chatbots as agents, and skipping integration work with existing systems.

Question: Can AI agents handle Japanese, English and Spanish customer messages?
Answer: Yes. AI agents built on modern large language models can read and reply to customer messages in Japanese, English, Spanish and other languages within the same channel. Quality depends on grounding the agent in the business’s own information and testing it on real customer messages before launch.

Conclusion

The choice between a marketing automation agency and an AI agency comes down to one question: does your business need predictable campaigns executed on schedule, or judgment-based work handled in real time across the channels your customers actually use? Japan’s own data, from RIETI’s adoption study to record labor-shortage bankruptcies, points in the same direction. Infrastructure and reasoning capability matter more than software subscriptions, and the people who build your system matter more than the platform it runs on.

The global giants have already made their bet on AI agents. The businesses that will thrive in Japan over the next five years are the small ones that make the same bet, at their own scale, with a partner who understands them. If you’re looking for an AI agency in Tokyo that builds for SMEs rather than for Fortune 500 budgets, let’s talk about your first agent.