AI Adoption Interviews

Q.What were the challenges before AI?

A.The website was far too old, social media wasn't bringing in customers, and phone handling was a hassle. The menu was handwritten, so every change took effort.

Q.Why PH AI Works?

A.First, I could consult entirely in Japanese. And the office is right near Little Tokyo in Makati — I was amazed to find an AI specialist with this track record in the neighborhood; someone you'd struggle to meet even in Japan. Given the cost of acquiring customers and of labor, AI was actually quite affordable.

Q.What exactly did you put in?

A.First a WordPress rebuild, then SEO/GEO aimed at small keywords — and we started ranking near the top. We use AI for Facebook-centered social marketing too: analytics, posts, and automation. We added an AI chatbot so it handles simple inquiries, and made the menu easy to update from a PC. Down the line I'd like to use AI to analyze rival shops and sharpen our own strategy.

Q.Timeline?

A.About a week of consultation, one to two months to build. We set a training period for staff and went into full operation about two weeks later.

Q.Results?

A.We've held a top 1–5 ranking on small keywords. Staff almost never handle simple inquiries now. SEO/GEO and social marketing worked, customers increased — we hired more (so labor cost rose), but sales rose more, so it's a net win. Honestly I was skeptical at first, but the return on investment has been excellent.

Q.Any difficulties, and a word for others?

A.Almost none — they took my direction and advice as-is. If anything, AI moves fast, so I hesitated over 'which one to adopt,' but deciding that with an expert was a big help. Businesses and shops that don't adopt AI will fall behind.

Q.Opening a salon in the Philippines is a big decision.

A.I spent about 20 years as a stylist in Japan. Seeing the energy of the Philippines on a trip, I decided to open here. I'd like to think this much experience is hard to find even in Japan.

Q.Why PH AI Works?

A.I'd been a customer of the operator before, so I went to them. We could talk in Japanese, and the AI track record was more than enough — no hesitation.

Q.What did you adopt?

A.A WordPress rebuild, with SEO/GEO left to AI. In particular, they automated the Google Analytics and Search Console analysis and improvement. We also added an AI chatbot, AI customer management, and AI marketing management.

Q.Results?

A.The biggest thing is that the labor for inquiries and bookings dropped to almost zero. Before, every time the phone rang mid-cut I'd keep the customer waiting to answer. Now I don't have to. If anything I keep marketing modest — push it and too many people come. The Philippines' population growth helps a lot, too.

Q.A word for those considering AI?

A.It's a world apart from Japan's declining beauty industry. Honestly, I can't imagine running a salon without AI anymore.

Q.What sparked the opening?

A.There was no personal gym in Manila where you could really communicate in Japanese. So we thought, we'll do it ourselves. When we learned the operator is a Nippon Sport Science University graduate, we hit it off completely.

Q.What did you adopt?

A.We built a WordPress site and use AI for SEO/GEO and Google Maps acquisition. Inquiries are handled by an AI chatbot, so the time we spend on them went to zero.

Q.Results?

A.People find us via Google Maps now, and inquiry handling is off our plate — so the trainers can focus on coaching.

Q.What's next?

A.We'd like to use AI to streamline our clients' training and meal plans.

Q.A word?

A.Even a small gym can make acquisition and operations this much easier with AI.

Q.What led to the request?

A.We'd already worked together on another business (video production), and it grew from there.

Q.What kind of system did you build?

A.A fairly large one in Next.js: email send/receive, schedule management, customer management, managing the site admin and multiple instructors, class schedules, evaluation management, instructor evaluations, marketing management, creating and managing teaching materials, linking materials to bookings, even semi-automated blog generation — essentially everything the school's operations need, in one.

Q.Budget and timeline?

A.Over ¥10 million, with development around 3–5 months. Given the scale, we took the time to do it right.

Q.Results?

A.It's a new launch, so there's no 'before vs after,' but customers are coming steadily. Having an efficient operating system from day one is huge.

Q.A word for those considering AI?

A.AI is essential technology for online English going forward. Schools that don't adopt it will, I think, decline.

Q.Tell us about the shop.

A.It's a small phone shop run by a Filipino owner — load (prepaid) sales, repairs, used handsets, accessories; several revenue streams. We're near Little Tokyo in Makati, so we occasionally need to serve Japanese customers who don't speak English.

Q.The challenge?

A.A lot of routine questions come in on Messenger — 'How much is this model?' 'In stock?' — and handling them was quietly draining.

Q.What did you adopt?

A.First, auto/semi-auto replies to routine Messenger inquiries, able to switch among Tagalog, English and Japanese, plus a recommendation feature that asks budget and use to suggest a handset. We also mass-produce Facebook posts — new arrivals, price drops, repair notices — with AI-generated copy and thumbnails.

Q.Results?

A.Inquiry handling got much easier, and we can keep posting on social without it stalling. We can serve Japanese customers without worrying about language.

Q.What's next?

A.Eventually I'd like to use AI smart glasses to smooth in-person service for Japanese customers, too.

Q.What was it like before you used AI?

A.I was running a YouTube channel, but I didn't really understand how to read the analytics numbers or turn them into the next move. Honestly, I was posting videos mostly on instinct.

Q.What exactly did you put in?

A.A setup where, once a week, Claude Code automatically gathers the channel's analytics data and produces an improvement report. From there, we go through it together and fix the content based on what it shows.

Q.Results?

A.In about six months, views roughly doubled and subscribers grew about 1.2×. I was already confident in the content — it was a pretty solid channel to begin with — so the growth rate might look modest. Even so, it's climbing steadily.

Q.What struck you most?

A.Honestly, I wish I'd put this in from the very start of the channel. If I'd been watching the numbers and improving from launch, I could have grown faster and bigger — so I do regret that a little.

Q.What's next?

A.Right now it's just YouTube, but I'd like to analyze the data from TikTok, Instagram and X the same way — auto-reported — and improve the content on each.

Q.A word for those considering it?

A.In the end, this is basically marketing automation. Thanks to it, I no longer need a marketing consultant or a dedicated analytics person. Even a solo creator can keep running data-driven improvements with AI, without hiring specialist staff.

Q.What problem brought you in?

A.I wanted to accept PayPal payments on my site, but I could no longer log into the PayPal account I'd set up earlier, and that's where I was stuck. That account had a Philippine BDO bank account linked to it, but the email domain I'd registered it with had expired. There was nothing I could do on my own — I was completely stuck.

Q.How did you solve it?

A.First, they contacted PayPal support and sorted out the old account that still had the BDO bank account tied to it. Then, using an email on the .com domain I'd already obtained for the site, they created a fresh PayPal account and re-linked the BDO account that had been freed from the old one.

Q.What about the site itself?

A.Once the new PayPal account was usable, they installed a full set of PayPal payment buttons and links on the site. Now customers can pay directly from the site.

Q.What's the AI analytics part?

A.Along with payments, we introduced an AI tool that analyzes the site's sales and marketing. We can look at the numbers together and talk through what to fix and how, which is a big help. That part is still ongoing.

Q.Why PH AI Works?

A.Honestly, the PayPal–BDO linking — with English correspondence involved — was completely beyond me alone. Being able to consult entirely in Japanese, and to hand over everything from the payment recovery to the site setup and the AI analysis beyond that, made a huge difference.

Q.A word for others considering it?

A.When you run into a payment problem abroad, you can really grind to a halt. Having someone you can hand that to makes all the difference — I felt it firsthand.

Q.First, tell us about your business.

A.I run an online shop by myself near Manila. As I started sourcing new products, I ended up dealing with overseas suppliers more and more. Email and chat I can manage with a dictionary, but the real problem was video calls.

Q.Where exactly were you struggling?

A.My English is around B1. If someone speaks slowly, I understand. But in a meeting the other side talks at normal speed, and I have to listen while also thinking about how to reply. My head would fill up, and I'd end up only saying 'Yes' or 'OK' the whole time. During an important price negotiation, I couldn't say even half of what I wanted to — it really hurt, because that conversation directly affects my business.

Q.What solution did you first consider?

A.Honestly, my first idea was to buy smart glasses with a translation feature. But looking into them, they all stop at generic translation - none of them frame a reply around my background or what is at stake in the negotiation. When I talked to PH AI Works, they said, 'With just your PC and Zoom, we can go further than that.' That was a real eye-opener.

Q.How do you use the tool they built?

A.I split my screen in two — the tool on the left, Zoom or Google Meet on the right. Before the meeting I enter my background, my English level, the other person's profile and that day's agenda. When they speak, the tool reads the English captions Zoom shows and translates them into Japanese. On top of that, it automatically suggests three English replies that fit the moment, so I just pick the closest one or tweak it a little into my own words.

Q.Whose AI does it use?

A.I can choose between ChatGPT, Claude and Gemini depending on the situation. For negotiations I use the one that thinks things through carefully, and for chattier meetings the one that responds quickly. I'm not locked into a single option, which is reassuring.

Q.What's the biggest change since you started using it?

A.I'm no longer afraid of meetings in English. Before, just having a meeting on the calendar weighed on me; now I can tell myself, 'It'll suggest replies for me, so I'll be fine.' I stopped missing what the other side says, so I can ask my own questions and negotiate terms too. In the end I did not need dedicated hardware like smart glasses — adding this tool to the PC and Zoom I already had was enough.

Q.We hear there's also a feature for after the meeting.

A.Yes. They added a feature that saves the meeting and lets me organize and analyze it with AI afterward. I can look back at what we discussed with a supplier last time and what I need to prepare for next time, which cuts down on he-said-she-said misunderstandings. When you run a business alone, records like this quietly make a big difference.

Q.What would you like to ask for next?

A.My anxiety about meetings is gone, so now I want to grow the shop itself. I'd love to ask them for SEO and GEO work so my shop gets found through search and AI, plus traffic analysis to see where my customers are coming from.

Q.First, tell us about your business.

A.I run websites out of Manila. They're in English, and I earn affiliate income mainly through ClickBank. I built the sites myself with Next.js and Python. I'm in my 50s, but this is the sort of thing I'd rather do myself than hand off.

Q.What made you want to use AI?

A.Writing blog content. I want to publish more, but I work alone and there's only so much I can do. The trouble was that pasting half-finished drafts and my site structure into a paid AI service felt risky — that's my material going somewhere I don't control. And the more articles I write, the more API usage I pay for. Not knowing what the monthly bill would be bothered me too.

Q.What came out of talking to PH AI Works?

A.They suggested a local LLM. The AI runs entirely on my own machine, so nothing I write goes out, and there's no extra charge no matter how much I generate. Honestly, I didn't even know that was an option.

Q.What setups did you look at?

A.We lined up a few and compared them. A small Raspberry Pi–based box for tens of thousands of yen, a vendor-built external local LLM machine for hundreds of thousands, and LLM services on AWS. After talking it through, I went with the simplest one: installing Ollama straight onto the computer I already use. Nothing special — an ordinary Windows laptop.

Q.How does it actually perform?

A.Compared with the latest paid versions of Claude, ChatGPT or Gemini, the quality is lower. I'll be straight about that. But for writing English blog posts, it's more than good enough. The trick I use is having several models check the same article. One model catches what another missed, and that alone lifted the quality a lot.

Q.How does an article get published?

A.They set it up so an article written in Ollama gets published straight to my Next.js blog. Writing, checking, publishing — the whole loop runs on its own now. In the morning I just look over what came out.

Q.Cost and timeline?

A.About ₱60,000, and roughly two weeks. There's no monthly bill, so the more I write, the more it pays for itself.

Q.A word for anyone considering it?

A.I think local LLMs are going to become the mainstream choice. Machines keep getting faster, storage keeps growing, memory keeps increasing while prices come down. Once that happens, there's less and less reason to hand your drafts to an outside service and pay for the privilege. Nothing leaks, nobody limits what you can write, and the cost is predictable — for someone running sites on their own, those three matter a great deal.

Q.What made you want to start this business in the first place?

A.I read about it in a magazine. The idea was to post AI-generated images of a person on social media and charge for access on a fan site. I'm in my 50s and I know nothing about PCs or IT. The only machine I own is a Windows laptop that cost around ¥100,000. I wanted to find out whether that was even enough to do it.

Q.What came out of the consultation?

A.The first thing they told me was, 'Building the system isn't the hard part. Payments and social media are.' Honestly, I'd assumed the opposite — that writing the program would be the ordeal and taking money would just mean hooking up a bank account. Finding out it was completely the other way around was the biggest surprise.

Q.What kind of site did they build?

A.A membership site in Next.js. Sign-up, age verification, what each plan can see, and the option to buy items one at a time. The images aren't kept anywhere the public can reach — as I understand it, the system checks whether you're a member and then hands out a link that only works for a short window. Non-members only see a lower-resolution, blurred version.

Q.Do you disclose that the person is AI-generated?

A.Yes, from the very beginning. That was something they pushed for strongly. Apparently there's a way of doing this where you hide it and let people believe it's a real person, but I was told, 'That's the most dangerous part of this business, both under the platform rules and legally. Say it's AI and build your fanbase on that — it lasts longer.' So far, being upfront about it hasn't cost me anything.

Q.What was the hardest part?

A.Payments. The images go as far as swimwear or lingerie, so I didn't think of it as adult content. But from the screening side, it gets treated as adult content. The ordinary payment services can't handle it at all under their own terms, so it was a repeated cycle of finding a route that might work and submitting it for review. I'm not disclosing the method or the provider, but that's where the time went.

Q.And social media?

A.Just as hard. All my traffic comes from social, so if an account gets suspended it goes to zero in a day. They taught me that it's less about the images themselves and more about the path that sends people to a paid site — that's what trips the rules — so I'm constantly adjusting how that's presented. And labeling content so it's clear it was made with AI. The platforms tighten this every year, so it's the topic we spend most of our monthly calls on.

Q.Cost and timeline?

A.The build was around ₱800,000 over three months. Now I pay ₱30,000 a month for maintenance and an advisory arrangement. I can't keep up with the rule changes on my own, so those monthly conversations are effectively what keeps the business running.

Q.How do you see it going forward?

A.I do have concerns. Payments and social media are both outside my control, so I don't know whether I can keep this business going indefinitely. Even so, I'm glad I did it. Not many people have taken a business that combines AI with attractive-persona content all the way through — payment screening, platform rules, all of it. Learning firsthand where the line is in a field people view as a bit dubious — what gets through and what gets stopped — was worth a lot. When I launch something similar next time, that experience will count.

Q.A word for anyone considering it?

A.Settle payments and social media before you build anything. I had the order backwards, so being told that at the start saved me. You can have the site finished and still have no business if you can't take the money.

Q.What made you want this tool built?

A.An acquaintance told me prediction markets existed. His point was that during an election, the market where participants stake their own money called it better than the opinion polls did. I'm in my 60s and I know nothing about PCs or IT. The only machine I own is a Windows laptop that cost about ¥100,000. But the idea of using those numbers as an input to my US stock decisions struck me as interesting.

Q.What does the tool actually do?

A.It collects prediction-market information and puts it on screen where I can read it. They built it with Next.js, Python and the APIs from several AI companies. All I do is look at it — I place the actual trades by hand in my broker's screen. We deliberately left out any automated order function.

Q.Why no automated ordering?

A.When we talked it through, they told me, 'A tool that displays and a tool that places orders are an order of magnitude apart in how much has to be built.' Handling dropped connections, reconciling whether an order actually filled — the list of things to worry about grows enormously. My trading runs at maybe once every two or three days to once a week, mostly day and swing trades. I'm no good at scalping by the second, so I don't do it. At that frequency, we concluded placing orders by hand is plenty.

Q.What was the hardest part of building it, as you understand it?

A.Connecting what the prediction markets are asking to actual stock tickers. What's listed on a prediction market is a question in plain language — 'Will there be a rate cut in September?' — with no structural connection to a ticker symbol. As I understand it, they have the AI read that and work out which companies or which sectors the question bears on. But if you let the AI do the arithmetic as well, the same input can give you a different answer, so the probability calculations are pinned down in the program code instead.

Q.Can you take prediction-market prices at face value as probabilities?

A.They warned me about that at the start too. In a market with little trading, the buy and sell prices sit far apart, so the number on display barely means anything as a probability. So they built in weighting based on volume and the size of the spread, and markets that are too thin get filtered out from the beginning. There's quite a lot in there whose whole job is to stop me from taking the on-screen number at face value.

Q.Cost and timeline?

A.The build was around ₱120,000 over a month and a half. Now I pay ₱15,000 a month for maintenance and an advisory arrangement.

Q.How do you use the monthly advisory arrangement?

A.It's proved more useful than I expected. They traded FX for many years, so they know charts and how to draw lines on them very well. Beyond reporting bugs in the tool, being able to talk through 'how would you read this setup?' once a month is valuable to me. I'm glad it wasn't a case of handing over the tool and that being that.

Q.What's changed most since you started using it?

A.I have one more input to work with. Before, it was my instinct and the chart, and nothing else. Now I can set alongside that a number showing how people are reading things when their own money is on the line. It doesn't mean I'm right every time — but I'm certainly thinking with more to go on.

Q.A word for anyone considering it?

A.Don't set out to build automated trading straight away. That's what I pictured at first too, but the advice was to build something you only look at, and see whether it fits the way you actually trade. That order turned out to be exactly right. And with something like this, whether you have someone to consult after it's built makes a real difference.