AI Calls Every Pub in Ireland to Ask for the Price of Guinness—What Small Businesses Should Do in the Era of Market Research Costs Dropping from ¥500,000 to ¥5,000

Humans Take a Year, AI Takes a Few Days—and Only ¥5,000.

Ireland has about 7,000 pubs. Calling each one to ask, “How much is a pint of Guinness?” would take a human about 350 hours if each call takes three minutes. Labor costs alone would easily exceed ¥500,000. Outsourcing to a research company would add even more.

This task was accomplished by an AI voice agent. In just a few days, it called every pub and collected price data. The total cost, including call charges and API usage, was just a few dozen dollars—around ¥5,000. The results were visualized in real-time on a map as the “Guinness Price Index.”

To dismiss this project as merely an “interesting experiment” would be a waste. This demonstrates that the cost structure of market research has decreased to one-hundredth of what it was. This collapse in costs is particularly significant for small businesses rather than large corporations.

What Happened: Real Insights from Six Months of Operation

This project was operational for about six months. The AI agent made calls, analyzed responses using voice recognition, and structured and recorded price data. While it sounds simple, the lessons learned from actual operations were quite gritty.

Lesson 1: Question Design is 90% of the Work
“How much is a pint of Guinness?”—this simple question can lead to disastrous results if poorly designed. If asked, “Is the Guinness draft or canned?” or “Is it a pint or half-pint?” how will the AI respond? Without pre-planned branching scenarios, the conversation can break down, making data collection impossible. Conversely, if you narrow down to one question and design about three branching options, AI can achieve sufficiently practical accuracy.

Lesson 2: The Respondent is Human. Time of Day and Tone Affect Results
The response rate varied significantly depending on the time of day the calls were made. The highest response rates occurred around 11 AM before lunch, while Friday nights were predictably unresponsive. Additionally, the tone of the AI’s speech—if too fast, it would get hung up on, and if too polite, it would raise suspicion. It took several weeks of tuning to find this “just right” balance.

Lesson 3: Don’t Aim for Perfection. 70% Accuracy is Sufficient
There’s no need to get responses from all 7,000 pubs. As long as a statistically significant sample size is achieved, regional price trends can be adequately observed. In fact, the response rate was about 60-70%, but it was still possible to clearly map out price differences by region (e.g., €5.5 to €7 in Dublin city, €4.5 to €5.5 in rural areas).

Why This is a “Game-Changer” for Small Businesses

Now, let’s get to the main point.

Traditionally, market research has been the “privilege of companies with money.” If you asked a research company to investigate competitive pricing, it would cost at least ¥300,000 to ¥1,000,000. Even if done in-house, dedicating staff to phone or web surveys would lead to exorbitant labor costs and time. Consequently, many small businesses have relied on “intuition and experience” to set prices.

Now, this can be done for ¥5,000.

Let’s consider the implications.

Example 1: A Local Construction Company Calls 30 Competitors Within a 50km Radius to Ask for the Price per Square Meter for Exterior Painting
If done by a human, this would take a full day and could be awkward. With an AI agent, it can simultaneously call 30 places asking, “I’d like a reference for a quote; could you tell me the price per square meter for exterior painting?” The cost, including call charges, would be just a few thousand yen. This way, the company can see where its pricing stands relative to the local market.

Example 2: A Restaurant Gathers Lunch Price Ranges from 50 Nearby Establishments
When opening a new location, knowing the surrounding price ranges is crucial. While one could check each listing on food review sites, the actual prices often differ from those listed. If the AI calls and asks, “How much is today’s lunch?” it can obtain real-time pricing.

Example 3: A Manufacturer Regularly Checks Current Prices from 10 Suppliers for Raw Materials
Once a month, the AI automatically calls 10 suppliers to inquire about pricing. By systematizing this, a database of price trends is automatically created, allowing for data-driven decisions on when to negotiate price increases.

The key point is that previously “impossible due to cost” research can now be conducted for a fixed monthly fee of ¥5,000.

The Technical Barriers for “Phone AI” Are Now Low

You might think, “Isn’t that technically difficult?” Two years ago, it certainly was. However, the situation has changed dramatically between 2024 and 2025.

• Text-to-Speech (TTS): Services like ElevenLabs and OpenAI TTS can generate natural-sounding voices for a few thousand yen per month.
• Speech-to-Text (STT): Whisper (OpenAI) can provide real-time transcription at nearly no cost.
• Phone APIs: Services like Twilio and Vonage charge just a few yen to several dozen yen per call.
• Conversation Control: With GPT-4o or Claude 3.5, branching conversations can be designed using just prompts.
• Integrated Platforms: Services like Bland.ai, Vapi, and Retell are emerging that allow for the construction of phone AI agents without coding.

In other words, even without programming skills, there is already an environment where you can build an AI to “make calls and gather information” for under ¥10,000 a month.

This Can Also Be Applied to Agriculture—In Fact, Agriculture Needs It Most

This structure applies directly to agriculture as well.

When a local farmer is trying to determine “how much to sell their vegetables for,” the reliable information is limited. They currently check the JA shipping prices, the prices at roadside stations, and the pricing at nearby direct sales outlets by visiting them in person.

An AI agent could call 20 nearby direct sales outlets every morning asking, “What is today’s price for tomatoes?” This alone would provide real-time access to local market prices. By aggregating weekly, price trends can be observed. Decisions like “Let’s hold back on shipments next week and consolidate for the following week” can be made based on data rather than intuition.

The cost would be just a few thousand yen per month. Farmers could potentially regain the pricing power that has traditionally been left to agricultural cooperatives or markets.

So, What Should Be Done?

For those thinking, “This is an interesting story, but it doesn’t relate to us,” I want to emphasize this:

Step 1: Identify One Instance of “Information Gathering Done Repeatedly by Humans via Phone or Email Each Month”
Checking competitor prices, confirming stock with suppliers, surveying customers—anything is fine. Find one task where “the same questions are asked each time by a human.”

Step 2: Try Calling Just 10 Cases Using Bland.ai or Vapi
There’s no need to start with 100 cases. Call 10 and verify if you can collect data effectively. The initial cost is almost zero; you can test it with just call charges.

Step 3: Create a System to Automatically Record the Collected Data into a Spreadsheet
Using Zapier or Make, set up a system to automatically transfer the data collected by the AI into Google Sheets. This way, “research → recording → analysis” can occur without human intervention.

Up to this point, the cost would be under ¥10,000. Half a day of time would be sufficient.

When Costs Drop to One-Hundredth, the Only Change is Whether to Do It or Not

If market research costs ¥500,000, the decision not to do it was rational. At ¥5,000, there’s no reason not to do it.

The case of the Guinness Price Index illustrates that this is not just about how “amazing AI is.” When the cost of information gathering approaches zero, the gap between those who have information and those who do not becomes decisive.

Large corporations have always had ample budgets for research costs. Small businesses have not been able to do this. But now, that cost difference has vanished. The same information can be obtained with the same accuracy at one-hundredth of the cost.

This is the first opportunity for small businesses to stand on the same playing field as large corporations.

The issue is not the technology. “The only question is whether to let them make 10 calls or not”—that’s all.