Conclusion First. The Era of Receptionists, Front Desks, and Inquiry Responses Operating for Less Than 50,000 Yen a Month Has Arrived

For small businesses, the heaviest fixed cost is labor. In particular, the tasks of “phone answering,” “guest reception,” and “responding to frequently asked questions” require personnel even though they do not directly generate revenue. Hiring one part-time employee costs between 150,000 to 200,000 yen a month, while a full-time employee costs over 250,000 yen.

Now, three technologies have simultaneously crossed the line of being “usable.”

• Voice AI (Low Latency API): Response delay of less than 200ms, handling calls at a pace indistinguishable from humans.
• $399 Open Source Robot: The “Reachy Mini” from Hugging Face, capable of physical reception and guidance actions.
• Local LLM Chatbot: No cloud required, runs on in-house PCs with almost zero monthly costs.

What happens to the “cost of receptionists” in small businesses when these three are combined? Let’s verify with specific numbers.

—

1. Voice AI—Once the 200ms Barrier is Surpassed, “You Can’t Tell if the Person on the Other End is Human or AI”

The dividing line for whether voice AI is usable comes down to just one number: TTFT (Time to First Token)—the time it takes for the AI to respond with the first sound after the user finishes speaking.

In human conversations, the average time for back-and-forth responses is between 200 to 300ms. If it exceeds this, it feels like there is a pause. Conversely, if it is below 200ms, one may not realize they are speaking to an AI.

From late 2024, multiple APIs capable of consistently achieving this sub-200ms response time have emerged. For instance, Groq’s Whisper + LLM pipeline has a TTFT of around 200ms. Cerebras and Fireworks AI are also competing at similar levels. OpenAI’s Real-time API, while expressive, has a cost of $0.06 to $0.24 per minute, which can amount to $60 to $240 a month for small businesses using 1,000 minutes (about 17 hours), translating to approximately 9,000 to 36,000 yen.

On the other hand, Groq and Fireworks’ inference APIs can sometimes keep costs below half of that. Let’s estimate that for 1,000 minutes of phone handling, the cost could be 10,000 to 20,000 yen a month.

What does this mean?

Hiring one person for phone handling costs 150,000 to 200,000 yen a month. With voice AI, it would be 10,000 to 20,000 yen. The cost difference is over tenfold. Moreover, voice AI operates 24/7, never takes sick leave, and won’t miss late-night inquiries.

“But it can’t handle complex inquiries, right?”—That’s true. However, consider this: 70-80% of calls to small businesses are routine inquiries like “What are your hours?” “Do you have this in stock?” or “I’d like to make a reservation.” AI can handle these, leaving only 20% for human staff. This alone could reduce the need for receptionists by 80%.

—

2. The $399 Robot “Reachy Mini”—Covering Situations Where a “Body” is Needed at Reception

The open-source robot announced by Hugging Face in 2025 is officially named Reachy Mini (previously code-named Microduck). The price is $399 (about 60,000 yen).

Let’s summarize the specs:

• Height: Approximately 25cm
• Motors: 15 installed (for upper body arms, neck, and torso)
• Control: Python-based open-source framework
• Learning: Supports imitation learning, allowing it to learn from observing human actions
• Equipped with a camera for object recognition

To be honest, if asked whether a 25cm robot can greet guests at a reception counter right now, I would say “it’s still early.” It’s too small to carry items or hand over documents, and its walking capabilities are still experimental.

However, the focus should be on price and scalability. Traditionally, research humanoids cost several million yen. Now, it’s 60,000 yen. Being open-source means it can be customized in-house. For example—

• Place it at the reception counter to detect guests with its camera → Integrate with voice AI to respond with “Welcome, how can I assist you?”
• Combine it with a tablet to capture guest names → Automatically notify the responsible person via Slack
• Assist with product explanations through simple demo actions

Saying “a robot will handle reception” may sound grand, but what it’s doing is essentially “camera + voice AI + automation of notifications.” The robot’s body is merely an interface to add “physical presence” to that.

With an initial cost of 60,000 yen and monthly running costs including electricity and maintenance of a few thousand yen, let’s assume it to be 5,000 yen a month.

—

3. Local LLM Chatbot—Inquiry Responses with Almost Zero Monthly Costs, No Cloud Involved

When small businesses consider implementing chatbots, the biggest hurdles are cost and risk of information leakage.

Cloud-based chatbot services often cost several tens of thousands of yen per month. Moreover, customer data and internal information must be sent to the cloud. Local small businesses, especially in professions like law and healthcare, have strong concerns about “not wanting to send data externally.”

Here, local LLM becomes an option. As of 2025, open-source models like Llama 3.1, Gemma 2, and Phi-3 can run on a standard desktop PC (ideally with 16GB of RAM, though a GPU is not mandatory). Using tools like Ollama, installation and startup can be completed in 10 minutes.

Here’s a breakdown of costs:

With an initial investment of 50,000 to 80,000 yen, the monthly running cost is approximately 2,000 yen. Compared to cloud services costing 30,000 to 50,000 yen a month, this results in an annual difference of 360,000 to 580,000 yen.

By feeding in-house FAQs and product manuals into RAG (Retrieval-Augmented Generation), the chatbot can provide accurate responses based on company data to inquiries like “What is the delivery time for this part?” or “What is the warranty period?” without sending any data externally.

—

Summing Up Monthly Costs—Will It Really Stay Below 50,000 Yen?

Let’s summarize the monthly running costs when combining these three elements:

The initial investment would be for Reachy Mini (about 60,000 yen) + local PC (about 70,000 yen) + setup labor, totaling around 200,000 yen.

With a monthly cost of 25,000 yen, compared to a part-time employee’s monthly salary of 150,000 to 200,000 yen, this is one-sixth to one-eighth of the cost. This results in an annual saving of 1.5 to 2.1 million yen. The initial investment could be recouped in just two months.

Not only is it “below 50,000 yen,” but it may even fall below 30,000 yen.

—

However, Here Are the Pitfalls

Looking at the numbers alone, it seems like something that should be implemented immediately. However, there are three barriers to consider when introducing it on-site.

1. The Problem of Not Having Someone Who Can Set It Up
API integration for voice AI, building local LLMs, and customizing robots—very few small businesses can handle these in-house. If outsourcing costs are added, the initial costs can skyrocket. This is why providers offering “pre-configured packages” will likely see growth in the future.

2. How to Handle the Remaining 20% That Is Not Automated
Complex inquiries that voice AI cannot handle, complaint responses, and emotional interactions—these require human involvement. Without designing an escalation pathway from AI to human, customer satisfaction may decline. It is essential to establish a flow from “AI handles → if unresolved, transfer to the responsible person’s smartphone” from the outset.

3. The Quality of Japanese Voice AI
While low-latency voice AI below 200ms has entered practical use in English-speaking regions, Japanese technology is still a step behind. Particularly, support for dialects and industry-specific terms is still developing. I expect rapid improvements between late 2025 and 2026, but at this point, it is realistic to focus on “standard Japanese for routine responses.”

—

So, What Should You Do?

There is no need to implement all three at once. If you prioritize, start with—

First, begin with a local LLM chatbot.

The reason is simple. It has the smallest initial investment (you can try it on existing PCs), poses almost zero risk, and its effectiveness is easy to measure. Start by feeding in your company’s FAQs and begin handling internal inquiries. Embed it on your website for customer-facing deployment. This alone will reduce the number of inquiry calls. With fewer calls, deciding to implement voice AI will become easier.

Next, implement voice AI to automate 70-80% of phone handling. This will save you 150,000 to 200,000 yen in labor costs.

The Reachy Mini can be explored as an “interesting experimental tool.” Automating guest reception can be covered 80% by voice AI + tablet. The need for a robot body will arise a bit later.

—

This Is Not About “Eliminating Receptionists”—It’s About Ending the State of “Only Being Able to Be a Receptionist”

Please do not misunderstand. I am not suggesting that AI will replace humans.

What is happening in small businesses is the issue of “people who should be doing sales or planning are being tied up with receptionist duties.” If one person in a five-person company is stuck as a receptionist, 20% of the workforce is lost to fixed costs.

When 80% of phone, reception, and inquiry tasks are automated for 25,000 yen a month, what can that person do instead? They can go out for sales. They can develop new clients. They can spend time on product development.

The essence of reducing costs is that “things that could not be done before can now be done.”

With voice AI, local LLMs, and affordable robots—these three have simultaneously crossed the line of being “cheap, fast, and usable” in 2025, marking a turning point for small businesses in terms of their “fixed cost structure.”

No one will tell you if you just wait. Start by installing Ollama. It will take just 10 minutes.