A few months ago, a hearing-aid retail chain in Germany was opening a new branch in a major city. The incoming store manager wanted to know which ENT practices (known as HNO in Germany) in his catchment would be worth visiting first.
The question sounded like a thirty-minute task. Search Google Places, sort by distance, hand over a spreadsheet.
We started there. The first query returned about 75 candidates. After clicking through ten of them, it became clear we were looking at the easy half of the problem.
This article walks through what came next: a six-layer validation pipeline that took the raw 75 down to 43 verified ENT practices, the failure modes that each layer caught, and the methodology we now use whenever a client needs a defensible doctor list in Germany.
Why the naive query fails
Three problems show up when you treat Google Places as the source of truth for medical data in Germany.
First, Google over-reports. Pharmacies, hearing-aid retailers, general practitioners, and the occasional occupational health doctor all surface in ENT keyword searches. Roughly a third of the returns are not what you wanted.
Second, Google under-reports. Some real ENT practices are tagged as medical_office rather than doctor in Google’s internal categorization. They never appear in the right keyword filter, even when their listed name contains "HNO Praxis" (the German abbreviation for ENT).
Third, Google has no concept of regulatory standing. In Germany, ENT doctors fall into three categories: panel-licensed contract doctors (Vertragsärzte, the ones who can bill statutory insurance), hospital-employed specialists, and pure private practices. A retail customer cares about all three. The federal directory only lists the first.
For a hearing-aid retailer choosing which doctors to invest relationship time in, getting any of these three categories wrong has real cost. A missing high-volume practice is a missed referral channel. A false-positive Hausarzt visit is a wasted Tuesday morning.
The six validation layers
Each layer catches a different category of error. Here is what each one does and what it caught in this project.
Layer 1: Broad recall via Google Places
We started with the obvious tool. Google Places Nearby Search and Text Search across the city boundary, using ENT-specific German keywords (HNO, Hals-Nasen-Ohren, Ohrenarzt, Otologie). We pulled the actual administrative polygon from OpenStreetMap rather than using a viewport rectangle, so the search did not bleed into neighboring municipalities.
This returned 53 candidates. The goal at this stage is recall, not precision. Every plausible ENT practice must be captured, with later layers trusted to filter.
Layer 2: Rule-based classification
A classifier scored each candidate on positive vocabulary (HNO, Otologie, Phoniatrie) and disqualifying vocabulary (Hausarzt, Apotheke, Zahnarzt, Frauenheilkunde, Akustiker, Betriebsarzt — all German medical terms). Four ratings: HIGH, MEDIUM, LOW, UNKNOWN.
This layer caught the obvious failures, including one Hörakustiker that Google had mis-tagged as a medical practice. Cheap to run, removes maybe half of the false positives.
Layer 3: Google Place Details
For ambiguous candidates from Layer 2 (the MEDIUM and UNKNOWN buckets), we pulled full Place Details responses. These contain business descriptions, website URLs, and more detailed type classifications than the basic Nearby Search returns.
One representative case: an entry whose name suggested ENT turned out, on Place Details inspection, to be a different category of medical service entirely. Place Details costs more per call but resolves most of the genuinely ambiguous candidates.
Layer 4: Cross-reference against the federal directory
This is the layer that separates a hobbyist analysis from a defensible one.
Germany publishes a public directory of every panel-licensed physician at arztsuche.116117.de, maintained by the Kassenärztliche Bundesvereinigung. For this city’s catchment, the directory listed 58 ENT doctors across 35 distinct practices.
We parsed the PDF export into a structured dataset and matched it against the Google results using a two-pass algorithm: first by street address with German compound-word normalization (so Straße and Str. both match the same physical street), then by doctor surname as a tiebreaker for practices with multiple physicians at one address.
The cross-reference produced four outcomes:
- 37 confirmed matches: Google listing maps cleanly to a federally-registered practice
- One surname collision: two unrelated doctors sharing a family name but working 15 km apart, who matched on name overlap until the address column revealed the discrepancy
- 13 Google-only candidates: present in Google but not in the federal directory. Each one needed manual review.
- 10 federal-only practices: present in the regulatory directory but missing from Google
Layer 5: Specialty filter on address-matched entries
Address-first matching has a known failure mode: two specialties can share a building. Layer 5 re-checks every confirmed match for disqualifying specialty markers in the practice name itself.
Two examples from this project. In one multi-floor medical building, a gynecology practice occupied the 6th floor and the real ENT practice occupied the 5th. The address-first match confirmed the gynecology practice as ENT because the street address matched. The specialty filter caught it.
In another case, a general practitioner sat four buildings down from a real ENT practice. Same street name, house numbers close enough that the address matcher considered them the same practice. The specialty filter caught that one too.
Layer 6: Manual additions for structural directory gaps
The federal directory only lists contract doctors. It excludes two categories of practicing ENT physicians who matter for hearing-aid referrals.
Hospital-employed ENT specialists. In any major German city, the university hospital’s ENT clinic is one of the largest single referral sources in the region. None of those doctors are in the federal directory.
Pure private practices. These doctors do not bill statutory insurance and therefore carry no federal-directory registration. They still see patients. They still write hearing-aid prescriptions.
We added four such practitioners after manual verification. We also added two of the 10 federal-only entries from Layer 4 that turned out to be real ENT practices in the city: one of them was tagged by Google as medical_office rather than doctor, despite the practice name containing "HNO Praxis." That single misclassification had hidden a high-volume practice from a hundred different keyword searches.
What the audit revealed
After all six layers, 43 verified ENT practices remained. That number, on its own, is a small story. The shape of the verification work is the bigger one.
Of the 43 verified practices, only 37 came through both Google and the federal directory cleanly. The other 6, about 14% of the final list, would have been missing or incorrectly characterized if we had relied on Google alone. For a retailer choosing where to invest first-year relationship effort, missing 14% of the real practices in your catchment is a non-trivial blind spot.
The new branch was the closest hearing-aid retailer for zero of the 43 practices. Every ENT doctor in the catchment was already closer to an existing competitor.
The strategic finding from the analysis is sharper than the methodology one. When we mapped each verified ENT practice to its geographically nearest hearing-aid retailer (including the new branch the client was opening), the new branch was the closest hearing-aid retailer for zero of the 43 practices. Every ENT doctor in the catchment was already closer to an existing competitor.
That finding changed the strategic framing of the launch. It moved from "capture the natural catchment" (which there was none of) to "identify and displace specific incumbents at named practices" (which is a different operating playbook). And the finding only holds if you trust the underlying list. If 14% of your data is wrong, "zero of the practices" is a noise statement. With the validation work done first, the zero became a defensible operational claim worth taking to the regional manager.
What generalizes beyond ENT
This six-layer pattern applies to any healthcare specialty in Germany that has a regulatory directory and a Google Places presence. Dentists, ophthalmologists, orthopedists, veterinarians. The scaffolding is the same. Only the keyword vocabulary in Layer 2 changes.
It also extends beyond healthcare to any vertical with a regulatory or industry registry that can serve as the cross-reference layer. Certified accountants with professional body lists, licensed engineers, building survey firms with chamber registrations, licensed pharmacists. Wherever a public directory exists, Layer 4 has an analog.
The cost of doing this right, in our experience, is three to five business days per metropolitan area. The cost of skipping it is committing your operational effort to a list that’s missing the practices you most needed to find.
43
Verified ENT practices
14%
Would be missing with Google alone
0
Practices with the new branch as nearest retailer