POJ Studio and Discover Japan: 16 Growth Ideas
This report lists ideas for each company, in priority order. The ideas use technology and AI in the back office and keep customer contact human.
Bottom line
Both companies should keep their customer experience as it is. The best opportunities are in back-office work.
- POJ has more demand than trained hands. Repairs are taking 6 to 8 months and two 2027 KAP terms are full. More marketing would mainly make the wait longer. The priority order is P1, P2, P10, P9, P3 and P4.
- Discover Japan does not show its past work to public clients. It claims 1,000+ cases and shows none (screenshot). Start with D1, D2 and D4. All three are internal, so they do not change the company culture.
Two points to check.
POJ already uses technology. Wallpaper reports that founder Tina Koyama spent six years at Twitter (screenshot), and POJ's About page cites a decade in digital product design. Its storefront runs Shopify, Klaviyo, Appointo and more. Ask her which tasks she would automate first.
Public records do not confirm Discover Japan's digital work. Digital media for local governments may be its largest activity. The records I found mostly show print, tie-ups and guidebooks. Digital work may be less visible in public records. The revenue split is not public.
Where to use technology
Customers should deal with people, not automated systems. POJ says it believes in wabi-sabi. Takahashi-san compares editors to chefs (screenshot). Both companies depend on human work.
Do not automate
- Artisan and restorer judgment
- Stories, editing, bylines
- Product and editorial photos
- Shop staff and teachers
- Curing, firing, any craft time
- In-person relationships with regions
Can automate
- Intake and quotes for review
- Stock, counts, transfers
- Duties, labels, paperwork
- Scheduling and waitlists
- Tender search and proposal admin
- Reports and rights tracking
Test for each idea: if a customer learned how it worked, would they feel misled? A list of customers to thank is acceptable. A machine that imitates handwriting is not.
POJ Studio: 6 ideas
POJ's site already shows the offers below. Each idea extends one of them, so POJ does not need to start from zero.
Kyoto classes and apprenticeships as listed on November 1, 2026. Classes, KAP, Advanced, woodworking.
| Idea | Effort (my estimate) | |
|---|---|---|
| P1 | Repair desk ★ Prepare a draft quote for each repair request. The restorer reviews and approves it. | Small |
| P2 | Personal notes list Software lists which customers should receive a note. A staff member writes each note by hand. | Small |
| P10 | Creator visit Invite one Japan-focused video creator to take a real class and share the experience. | Small |
| P9 | Remote kintsugi course Add a remote course priced between the $884 class and the $9,000 programmes. | Medium |
| P3 | Kintsugi for leaders ★ A two-day management course in the US about leading after a project failure. | Large |
| P4 | Group event desk Reply to each group inquiry within one day with a quote. | Small |
Management training and education budgets
Short answer: it can work as a designed management programme bought by a company. It will not work as a kintsugi class on an expense form. I would treat it as a small additional business until a pilot shows demand.
Will not qualify (my reading)
- The kit, or a weekend class
- Kyoto hotels, meals and flights under Section 127
- An apprenticeship paid as job training, because it points to a new trade
What a programme needs
- Outcomes tied to work, such as leading after a failure
- Syllabus with contact hours
- Attendance and a certificate
- Separate invoice lines for tuition and materials
The description must always be true, and POJ should never coach anyone on getting approval. Have a US tax adviser review this before any sales copy mentions tax.
Discover Japan: 10 ideas
The site's own page for companies and government lists 13 services. Each idea below extends one of these services or supports it. Public records show real contracts, from ¥999,000 in Okinawa (screenshot) to ¥20,000,000 at TCVB (screenshot).
| Idea | Effort (my estimate) | |
|---|---|---|
| D1 | Private case library ★ Create an internal, searchable record of past projects. | Small |
| D2 | Weekly tender list ★ Send the team a one-page weekly list of relevant public tenders. | Small |
| D3 | Proposal desk and win-loss book Automate the administrative parts of a proposal so editors have more time for the concept. | Medium |
| D4 | KPI report pack ★ Give each public client a short report with the numbers it must report each year. | Medium |
| D5 | AI-use log Check each contract for AI rules, and keep a simple record of any AI use. | Small |
| D6 | Translation desk and /worldwide Improve the translation process, and restart the English section one region at a time. | Medium |
| D7 | Photo and rights index Record the usage rights for each photo, and make photos searchable for staff. | Medium |
| D8 | Regional back-issue books Create booklets from past articles about one region for use by that region. | Medium |
| D9 | Lab and shop back office Record which customers bought from each maker, and tell them about future exhibitions. | Small |
| D10 | Creator matching for regions Match regions with video creators, and report the results. | Medium |
D1 Private case library Start here
Create an internal, searchable record of past projects.
- Builds on
/advertising claims over 1,000 cases, but WORKS shows one entry: a contact prompt with a broken image. (screenshot)
- Back end
Gather old proposals, reports and PR articles. Tag them with the company's own 13 service names, region, client type, budget and result. Editors search it in Japanese. Later, an editor may hand-pick 12 real cases for the public page.
- Stays human
Everything public stays editor-chosen. This tool is for internal search only. It does not create content.
- Evidence
- The sample Toyama tender (screenshot) asks for 2 to 3 past examples.
- My crawl found 652 PR-labelled articles out of 5,499, so there is a large amount of material.
- First test
Index 50 past projects. Use it on one live proposal.
- Measure
Minutes to find three relevant cases. Whether editors keep using it.
- Risk
Old client files may have confidentiality terms. Keep the data on company systems.
D2 Weekly tender list Start here
Send the team a one-page weekly list of relevant public tenders.
- Builds on
The 企業・行政の方へ (for companies and government) page shows the company already serves government. It already won a ¥20,000,000 guidebook (screenshot) through competition.
- Back end
Use the free national tender API, filter by region, theme and service, and add DMO reports. A producer reads the list each Monday and chooses what fits. Check local-office requirements first.
- Stays human
The choice of what to pursue, and all contact with regions, which still happens in person.
- Evidence
- My count from the free 官公需 (government procurement) API: 166 unique tourism titles from 92 organisations in 12 months, 40 naming a proposal process. This is a minimum count, not the total market.
- There are 350 DMOs (screenshot) (regional tourism organisations), and they publish reports.
- NJSS and 入札王 are paid tender search services. Their prices are by quote only, and they may cover more. Start free.
- First test
Run for 8 weeks alongside how the team finds work today.
- Measure
Relevant tenders the team would have missed.
- Risk
The Toyama example requires an office in Toyama, Ishikawa or Fukui. A 20-person Tokyo firm may need local partners.
D3 Proposal desk and win-loss book
Automate the administrative parts of a proposal so editors have more time for the concept.
- Builds on
コンセプトワーク (concept work) and アドバイザー (advisory) are listed services. Takahashi-san has said 目指すのは、量より質です (screenshot): quality over volume.
- Back end
Read the tender and turn it into a checklist mapped to the scorecard. Pull relevant cases. Fill in company details, schedule and team sections. Record each win or loss by scorecard line.
- Stays human
Editors keep full control of concept, design and writing. These also receive high scores in tenders.
- Evidence
- Toyama's scorecard (screenshot): design 25 of 100, past work only 5.
- Price was 5% to 10% of the score in the small sample, for example Ishinomaki (screenshot). Proposal quality matters more than price in these examples.
- A Japan Tourism Agency pilot (screenshot) cut the time to draft a hotel marketing proposal from 2 hours to 15 minutes. That is not a public tender proposal, so it is only a rough guide. Users also reported problems with accuracy and stability (screenshot).
- First test
Redo one past tender and compare hours and gaps.
- Measure
Hours per proposal, win rate over time.
- Risk
Check each contract's AI rules first (see D5).
D4 KPI report pack Start here
Give each public client a short report with the numbers it must report each year.
- Builds on
Grant-funded work such as Gifu's plan naming Discover Japan (screenshot). The company does not publish results data.
- Back end
After each job, produce a short report: reach, then signs of interest, then the client's own KPI. Written in language an official can reuse. An editor adds a page on what the region should do next.
- Stays human
The editor's analysis and the follow-up visit. Takahashi-san has said the job carries a responsibility afterwards (screenshot).
- Evidence
- The 地域未来交付金 (regional grant) rules (screenshot) require indicators to be verified 毎年度, every year.
- Under the revised rules, registered DMOs renewing must report (screenshot) tourism spending, overnight guests and visitor satisfaction (trade press column).
- Kumano Travel grew bookings from ¥40M in 2011 to ¥521M in 2019 (screenshot) using local storytelling and a simple booking system.
- First test
Build one report for a finished job and ask that client if it would have helped.
- Measure
Repeat commissions. Officials reusing the text.
- Risk
Page views do not equal overnight stays. I found no proof that magazine reader data drives visits, so reports should not claim this.
D5 AI-use log
Check each contract for AI rules, and keep a simple record of any AI use.
- Builds on
The company's trust with public clients, and its clear PR labels.
- Back end
For each job, read the special terms, then log the tool used, what for, and who reviewed it. Keep AI to research, admin and checks. Final text and images stay human-made.
- Stays human
Authors stay responsible for all writing. Paid and editorial work stay separate.
- Evidence
- Osaka Waterworks Bureau's contractor guideline (screenshot) (January 2026) requires prior confirmation by form, limits AI to work support, bans consumer tools such as ChatGPT, and asks contractors to keep usage logs (screenshot). Other Osaka departments have their own versions. I found no national standard.
- Local government adoption (screenshot) reached 100% for prefectures and designated cities, 46% for others, as of October 2025.
- Agency for Cultural Affairs (文化庁) guidance points to human creative input for copyright. Contracts that transfer rights need this.
- First test
Check the last 10 contracts for AI clauses.
- Measure
Contracts checked before use: 100%.
- Risk
Low. This should come before D3 and D6.
D6 Translation desk and /worldwide
Improve the translation process, and restart the English section one region at a time.
- Builds on
多言語 (multilingual) is a listed service. The English section promises an insider's guide, but its footer says 2018 (screenshot) and its numbered issues stop at Vol.23 (screenshot).
- Back end
Build a shared glossary for craft, food and regional terms. Machine translation creates a first draft. A human translator and an editor then review it. Sell faster language versions on municipal work, then restart /worldwide one region at a time, with an editor choosing each story.
- Stays human
Editors choose and sign off every piece. No bulk translation of the archive.
- Evidence
- Competitor: MATCHA Locally (screenshot) now delivers municipal posts in six languages automatically. Basic translation may become cheaper, so editorial quality must be the main difference.
- Inbound visitors reached 42.68M in 2025 (screenshot), while September 2026 fell 5.0% (screenshot). The 2030 target (screenshot) for foreign nights in the regions is 130M, against 58.73M in 2025.
- DeepL Business and glossaries are cheap. No credible public test covers Japanese culture writing.
- First test
Blind test on 20 of its own articles, graded by its editors.
- Measure
Editor minutes per 1,000 characters. Editor preference.
- Risk
Some contracts or clients may ban machine translation. See D5.
D7 Photo and rights index
Record the usage rights for each photo, and make photos searchable for staff.
- Builds on
180 issues since 2008, a 5,499-article site, and named photographers on each piece.
- Back end
Audit rights by shoot: photographer, contract, permitted use, expiry. Then build a private image search for staff. Show expiry warnings before reuse.
- Stays human
Photography itself. No generated images in editorial or client work.
- Evidence
- Image search can run locally. Immich shows multilingual search works, but it is a family photo tool, so only a proof of concept.
- Impress had to exclude articles (screenshot) without permission from its own digital back catalogue (2005). Checking rights is the main task.
- First test
Audit rights on one region: all Ishikawa shoots.
- Measure
Share of images with clear reuse rights. Search time.
- Risk
Older photo contracts may be unclear, or may date from the company's 枻出版社 era.
D8 Regional back-issue books
Create booklets from past articles about one region for use by that region.
- Builds on
抜き刷り冊子制作 (reprint booklets) and カスタム冊子制作 (custom booklets) are listed services. Nagato (screenshot) planned to use its special issue in schools, libraries and trade meetings.
- Back end
Using D1 and D7, find everything published on one town. An editor assembles a booklet with a new introduction. Software only finds the articles and checks rights.
- Stays human
Article selection, new text and the printed booklet.
- Evidence
- The print market is shrinking: magazines fell 10.0% to ¥370.8bn in 2025 (screenshot). Reusing content could create income without a new photo shoot.
- Takahashi-san has spoken about re-editing what a place already has (screenshot). This idea follows that approach.
- First test
One region with deep coverage. One booklet dummy for one existing client.
- Measure
Booklet orders from past clients.
- Risk
Consumer sales of old issues are unlikely to matter. Sell to towns first.
D9 Lab and shop back office
Record which customers bought from each maker, and tell them about future exhibitions.
- Builds on
Discover Japan Lab. at Shibuya PARCO runs maker shows in sequence, and stock then moves to a shop with 4,029 products. The subscription is ¥13,310 a year.
- Back end
Link in-store and online buyers by maker, with consent. Before a show, list who to tell. Before a subscription ends, list who to thank. Staff write those messages. Remove outdated items, such as the public sample collection (screenshot).
- Stays human
The exhibitions, the shop staff and the messages. The exhibitions stay focused on the maker.
- Evidence
- Hobonichi is an example of a personal, content-led business at scale: sales of ¥8.677bn in FY8/2025 (screenshot), with no ads on its site.
- No loyalty or CRM tools were visible. The newsletter size is unknown.
- First test
Two exhibitions, with a manual invitation list built from purchase data.
- Measure
Repeat buyers per show. Annual renewal rate.
- Risk
Personal data needs clear consent, especially in store.
D10 Creator matching for regions
Match regions with video creators, and report the results.
- Builds on
動画制作 (video production) and 有識者コーディネート (expert coordination) are listed services. Its own YouTube has 379 subscribers (screenshot), so it should connect regions with creators, not compete with them.
- Back end
Use the company's editors and regional trust to brief creators on regional stories. Back-end tools cover contracts, disclosure, and the same KPI pack as D4.
- Stays human
Takahashi-san's relationships and an editor's story brief. The creator decides what to say.
- Evidence
- Hyogo Prefecture sponsored Sharla. The Kanto District Transport Bureau sponsored Paolo, with 612,922 views.
- Tokamachi's 2025 spec (screenshot) shows the buyer format: creators, a heritage story, reports with views and screenshots.
- Winkreative, the agency linked to Monocle, lists the Government of Thailand under Place. Discover Japan should keep paid and editorial work clearly separate.
- First test
One existing client, one creator, one disclosed video.
- Measure
Reach in the target market, and client renewal.
- Risk
Start with a partner the editors already trust.
Similar businesses
The sweets reviewer is not confirmed. Most likely Emmy Cho (emmymade), 3.16M subscribers, who started with a Japanese candy kit and has done wagashi taste tests. Either way, Sharla Hinskens is the better business model. The closest format is 正直レビュー at 444K, but its avatar is a cartoon man, so it may not be the person you mean.
| Business | By hand on purpose | On systems | Number |
|---|---|---|---|
| Sharla Hinskens | Her videos, travel content, a stationery shop | Sponsor deals, shop, Airbnb listing | 790K subscribers; Hyogo sponsorship |
| Hobonichi | A daily essay by the founder | Manufacturing partners, global sales | ¥8.677bn sales (screenshot); 52.5% of Techo sales overseas (screenshot) |
| Chewy | Pet portraits, handwritten notes | Autoship | 1,000+ paintings a week (screenshot) |
| Trader Joe's | Staff-lettered signs | Standard retail systems | About 3,000 products need signs (screenshot) |
| Le Labo | Blended at purchase, hand-labelled | Global brand operations | Hand-labelled for each buyer (screenshot) |
| Patagonia Worn Wear | Repair in stores and at Reno | Resale and logistics | 100,000+ repairs a year (2019) (screenshot) |
| Bokksu | Curation, maker outreach, culture articles | Subscriptions, support, email | $22M Series A in 2022. Founder card later printed (screenshot) |
| Sakuraco / ICHIGO | Themed boxes, small makers | Shipping to about 180 countries | About ¥4bn sales by 2022 (screenshot); makers often referred by local governments (screenshot) |
| HOSOO | Nishijin weaving by hand | A new 150 cm loom | Weave left kimono for Chanel and Dior clients (screenshot) |
| Monocle | Correspondents, print, shops | Sister agency, sponsored content | Valued at $47m in 2017. |
| Kumano Travel | Local storytelling, English-speaking staff | A simple booking system | ¥40M to ¥521M, 2011 to 2019 (screenshot) |
In these examples, people handle the creative and personal work, and systems handle routine tasks. Sharla sells products and also works with public-sector sponsors.
Possible link between the two companies: Sakuraco finds makers partly through local governments and banks. Discover Japan works with regions, and POJ works with makers.
Ideas I do not recommend
| Idea | Why not |
|---|---|
| Robot handwriting | It imitates human handwriting. Machine notes performed like no note (screenshot) in one study. |
| AI support agents for customers | Automatic replies would remove human contact, despite low prices such as Fin at $0.99 per outcome. Drafts for staff are fine. |
| AI articles, stories or images | Writing and images are the core product. AI use also creates copyright and contract risks. |
| Personalized homepages | Human selection is part of the value of both sites. |
| An AI travel concierge for DJ | GuideGeek exists for DMOs, but it would replace the editors' voice. |
| Airbnb as POJ's main booking channel | Airbnb itself narrowed its approach. |
| Returns software such as Loop | From $340 a month, designed for clothing-style exchanges, not fragile handmade goods. |
Held for later: a yearly accessibility check for sites DJ builds (only 52.1% of bodies have a policy (screenshot), but the check is only recommended), and maker stories for ふるさと納税 after the point ban, which is still a hypothesis.
What I could not verify
- Internal numbers: revenue, margins, order counts and team workload at both companies. For this reason, I did not estimate the financial size of the ideas. Third-party POJ revenue estimates looked unreliable, so I left them out.
- POJ: seed funding appears only in a PitchBook search snippet, with no amount. Waitlist size, repair volume and corporate clients are unknown. Hana Tsukamoto is missing from the About page, so check her role before naming her.
- Discover Japan: the revenue split. Current circulation is reported as 100,000 in FY2024 government reports. Most PR sponsors are unnamed.
- Press quotes: Wallpaper and some other articles came through a summarizing tool. Read the originals before quoting them.
- Tools: I found no independent proof of return for Flow, Sidekick, forecasting, booking or support tools.
- Legal: tariff and tax points are research, not advice. Both need a professional.
Terms used in this report
| Term | Meaning |
|---|---|
| API | A way for software to request data directly from another service. |
| Autoship | Chewy's automatic repeat delivery service. |
| CRM | Customer relationship management. Software that stores customer records and contact history. |
| DMO | Destination management organisation. A regional body that manages and promotes tourism for an area. |
| Duty-paid label | A shipping label showing that import duty has been paid before delivery. |
| HBS | Harvard Business School. |
| ICF | International Coaching Federation, which accredits coaching courses. |
| JNTO | Japan National Tourism Organization, which publishes visitor numbers. |
| KAP | POJ's Kintsugi Apprentice Program. |
| KPI | Key performance indicator. A number used to judge results. |
| Klaviyo | Email marketing software that also collects product reviews. |
| Landed cost | The full cost of an imported item, including shipping, duty and fees. |
| PR | Public relations. In this report, a PR label marks content paid for by a client. |
| Prefiling | Sending customs data before a parcel ships. |
| Proof of concept | A small test that shows whether an idea can work. |
| Section 127 | The US tax rule that lets an employer pay for education tax-free, up to a yearly limit. |
| Section 301 | A US trade law that allows tariffs on goods from specific countries. |
| SHRM | Society for Human Resource Management, a US human resources association. |
| Shopify Flow | Shopify's free tool for automating store tasks. |
| Sidekick | Shopify's built-in AI assistant. |
| Stocky | Shopify's inventory app. |
| Tender | A public request for companies to bid or submit proposals for a contract. |
| Tie-up | Paid editorial content made with a client. |
| USITC | US International Trade Commission, which publishes US tariff rates. |
| Wabi-sabi | The Japanese idea of finding beauty in imperfect and aged things. |
| Win-loss book | A record of which proposals won or lost, and why. |
| Yakimono | Japanese pottery. Here, POJ's pottery apprenticeship. |