Every problem has a path we worked out together.

Not just the numbers: what the problem was and how we went about it. Our work on patient acquisition, operational improvement, and AI in healthcare.

How to read these results

The four patient acquisition cases are all from the same clinic. They are not results from four different clinics. The clinic’s name and specialty are not disclosed for confidentiality reasons.

Patient acquisition figures are based on GA4 and Google Ads, and working hours on actual measurement of the tasks involved (data checked: August 31, 2026). Conversions, which include clicks on web booking, LINE, and phone, differ from the actual number of completed bookings or new patients. These are results under specific conditions and do not guarantee similar outcomes.

Clinics can check their own website for free. Website check (Japanese; no sign-up, about 1 minute)

From 500 users a week to over 25,000 a week in six months

Weekly users

About 500 / week25,000+ / week

Duration
6 months

The starting problem

The clinic had a website, but it was reaching only people who searched for the clinic by name. It was barely reaching people searching with words for their symptoms or worries. When ads were paused, traffic almost stopped too.

What we did together

  • Add articles that answer the words patients search for, checking that each one does not overlap in intent with existing pages
  • Connect articles to information on visiting the clinic
  • Review the search numbers every week, build on what works, and drop topics that miss
  • Make a wording review based on Japan’s Medical Advertising Guidelines a fixed step before publishing

Carrying the extra traffic through to bookings

Monthly conversions (web booking + LINE + phone clicks)

1,451 / month2,281 / month

Duration
1 month

The starting problem

Traffic had doubled, but bookings and calls had not grown in step. To begin with, there was no way to measure where people were dropping off.

What we did together

  • Make web booking, LINE, and phone actions measurable, so we could see where people were leaving
  • Improve the entry points to booking and the path from articles to information on visiting the clinic
  • Answer, in advance, what first-time patients want to know before a visit

Because the added traffic came from people gathering information, the conversion rate itself fell, but the number of actions toward the measured booking flows rose. These are not actual completed bookings.

Going after traffic from people who ask AI

Monthly sessions via AI assistants

About 3.1xvs. before

Duration
1 month

The starting problem

Few visits were coming via AI assistants such as ChatGPT, and there was room to organize the clinic’s services and information so they were easier to read.

What we did together

  • Reorganize the clinic’s services and information so both AI and people can read them easily
  • Present the clinic’s information and common questions in a form that search engines and AI can also read
  • Rewrite the body text so the key points and their basis are quick to grasp

The conversion rate for this channel was 26.7% (including clicks on booking flows, etc.). This compares periods in which several improvements were made; it does not show the effect of any single measure or guarantee being featured by AI.

Cutting the cost per acquisition for ads in half

Cost per acquisition (CPA)

¥2,879¥1,326

Duration
1 month

The starting problem

The ad budget was being spent in full, yet the cost per acquisition kept rising. Clicks that did not match intent were draining the budget, and the main keywords were running out of budget.

What we did together

  • Revisit keyword match types to stop spend leaking to clicks that do not match intent
  • Shift the budget back toward the areas of care that were producing results
  • Use negative keywords so ads do not show for unrelated searches

Over the same period, acquisitions rose from 35/month to 75/month (2.1x).

Seven hours of paperwork a day, down to one

Time spent on paperwork

7 hours / day1 hour / day

Duration
3 weeks

The starting problem

Paperwork split between paper and Excel was taking seven hours a day of a staff member’s time. Adding people did not reduce the hours.

What we did together

  • Spend a full day on site, write down the actual procedure exactly as it was, and break it into steps
  • Automate transcription, formatting, and output, leaving to people only the points that need judgment
  • Stay alongside the team after go-live, ironing out exception patterns until it ran stably

Shortening insurance-claims work on a cloud-based EHR

Monthly processing time

60 hours / month2 hours / month

Duration
4 weeks

The starting problem

Monthly work related to insurance claims was taken up by manually exporting data from the electronic health record (EHR) and reconciling it.

What we did together

  • Test both the EHR’s public API and its CSV export, and choose a realistic integration method
  • Implement the reconciliation logic and reorganize the work so that people review only the differences
  • Release in time for the monthly close, and run the first cycle side by side with the staff

Letting AI draft clinical records

Time to write one record

20 min / record5 min / record

Duration
3 weeks

The starting problem

Writing each record took 20 minutes, squeezing the time after consultations. The PoC worked, but it was not making it into daily operations.

What we did together

  • Redesign the data flow so that personally identifiable information is anonymized or pseudonymized before processing
  • Treat the output strictly as a draft, with a UI that assumes the physician will edit it
  • Settle the API’s data-retention policy and a cost ceiling before starting implementation

Let’s figure out the next step together.

No spec required. Tell us about the problem or the idea — we’ll take it from there.

Talk to us