LOKLU is working…
Every question goes to one chat. The model decides whether it needs LOKLU's Berlin evidence. If the question touches a place, a market, premises, competition, audience, rent, a budget or a location decision in Berlin, it calls LOKLU tools. The numbers in those answers come from the tools, not from the model.
The tools compute on a dated snapshot of LOKLU's production database: 448 planning areas (Planungsräume), the Zensus 2022 100 m grid, 47,711 Google Maps places, OpenStreetMap, 486,731 IHK business-lifespan records from 37 monthly register snapshots, VBB transit timetables and LOKLU's category benchmarks. Results say which date they are from. Nothing is real-time.
Questions that need no local evidence — a letter, a VAT question, a business-plan outline — get a general answer. Each answer says which kind it is:
LOKLU evidence covers Berlin only. For other places it answers general questions and states no local numbers.
Where it matters, a statement carries one of these labels. A signal, scenario, assumption or general statement is never presented as a fact.
The charts in LOKLU answers follow a few fixed rules, so one chart reads like the next.
Half of the Berlin planning areas lie below it, half above (areas with at least 3,000 residents). Differences are written as a ratio (“1.8× Berlin”) or in percentage points (“−2.8 pp”), never as “good” or “bad”.
● FACT and ◐ SIGNAL are drawn solid; ◇ SCENARIO and ~ ASSUMPTION outlined or hatched, because they are calculated from premises; ○ UNKNOWN is the hatched no-data pattern.
Where LOKLU does not know a value, the chart shows diagonal hatching, ○ UNKNOWN and the reason. A zero is only drawn when the source really says zero.
UNKNOWN · no measured footfall in this snapshot
Below five places (or thirty register entries) a share would mislead, so the chart prints the counts instead of bars.
3 places: 1 × €€, 2 × no price
| Measure | this area |
|---|---|
| € | 0 |
| €€ | 1 |
| €€€+ | 0 |
| no price | 2 |
“Table” next to the title shows the exact numbers; screen readers get the table and a one-sentence summary; printouts include both.
Up to four numbers, each with the Berlin median and ▲ / ▼ in words; colour never carries the meaning alone.
The dot is this area, the dashed line the Berlin median, the grey band the typical Berlin range (p10–p90).
Google, OpenStreetMap and the IHK register count separately and are never added up.
each source counts separately, never added up
| Measure | Value | Label |
|---|---|---|
| 23 | ● FACT | |
| OpenStreetMap | 12 | ● FACT |
| IHK | 18 | ● FACT |
Two to five areas on the same measures, told apart by ① – ⑤, never by colour; each row has its own scale.
| Measure | ① Helmholtzplatz | ② Arnimplatz | ③ Kollwitzplatz | Berlin median |
|---|---|---|---|---|
| Cafés per 1,000 residents | 0.62 | 0.21 | 0.48 | 0.28 |
What each component adds to the screening score; a component without data is hatched and its weight goes to the others.
The line is how many are registered now; bars show what appeared (up) and left (down) each month.
Parts of one whole in one bar, the largest part darkest (lightest on black), at most four parts plus “other”.
| Measure | Share |
|---|---|
| Helmholtzplatz | 71.3 % |
| Kollwitzplatz | 18.9 % |
| Arnimplatz | 6.2 % |
| Humannplatz | 3.6 % |
Shares of € / €€ / €€€+ for the area and for Berlin; “no price” is hatched and always last because it is unknown, not cheap.
| Measure | this area | Berlin |
|---|---|---|
| € | 18.4 % (7) | 26 % (1,225) |
| €€ | 42.1 % (16) | 37 % (1,743) |
| €€€+ | 21.1 % (8) | 8 % (378) |
| no price | 18.4 % (7) | 29 % (1,366) |
The dot is this area, the dashed tick Berlin; the line between them is the gap, not a recommendation.
| Measure | n of 29 | this area | Berlin |
|---|---|---|---|
| closed Sunday | 3 | 10.3 % | 12.4 % |
| closed Monday | 7 | 24.1 % | 18.2 % |
| opens before 8:00 | 12 | 41.4 % | 33.1 % |
| open after 19:00 | 6 | 20.7 % | 22.9 % |
Four days of the year: when the sun can reach the facade (upper bound, no buildings modelled) and how much is left on an Altbau street.
Facade SW as you gave itYOUR INPUT· expected in June after typical cloud ~4.3 h
| Day | Sun on the facade | direct sun (open horizon) | over an Altbau street | 17:00 terrace evening |
|---|---|---|---|---|
| 21 Mar | 09:41–17:43 | 8.0 h | 1.5 h | 0.7 h |
| 21 Jun | 11:23–20:45 | 9.4 h | 5.7 h | 3.8 h |
| 21 Sep | 10:28–18:30 | 8.0 h | 1.7 h | 1.5 h |
| 21 Dec | 09:10–15:00 | 5.8 h | 0.0 h | 0.0 h |
Fit-out, deposit and reserve per size on one scale; the solid line is your budget and each size says “fits” or “short by” in words.
| Size | fit-out | deposit (3 mo.) | reserve (6 mo.) | capital needed | Status |
|---|---|---|---|---|---|
| 60 m² | €150,000 | €6,840 | €14,636 | €171,476 | ✓ fits · €88,524 left |
| 120 m² | €300,000 | €13,680 | €29,270 | €342,950 | ✕ short by €82,950 |
250 m and 500 m straight-line rings, not walking distance, and where the count sits among Berlin addresses.
compared with 386,000 Berlin addresses
Every map says in words which way the shading runs, where the class bounds are, and which areas have no data.
Local numbers — rents, residents, competitor counts, closure rates, footfall, revenue of places — may only come from a LOKLU tool result in the conversation or from you. After each answer, the interface checks every material number and sorts it:
The check deliberately over-flags numbers tied to a place: a false warning is cheap, an invented local number is not. General numbers such as a tax rate are allowed only as GENERAL knowledge, with a note on where to verify them.
The screening ranks Berlin's planning areas for a café concept. Each of six components becomes a 0–1 percentile across the screened areas; the score (0–100) is their weighted mean. By default, areas with fewer than 3,000 residents are left out. When a component has no data for an area, the area is scored on the others and the missing part is reported.
| Component | Weight | What it measures |
|---|---|---|
| Audience Premium-audience proxy: ESIX social-status class of the planning area (1 = highest status of 7). Sozialstrukturatlas Berlin 2022. Label SIGNAL — status index, not measured income. | 25 % | Premium-audience proxy: ESIX social-status class of the planning area (1 = highest status of 7). Sozialstrukturatlas Berlin 2022. Label SIGNAL — status index, not measured income. |
| Café demand Observed café activity: total Google review count of cafés in the area (Apify Google Maps scrape). Label SIGNAL — reviews track visitor traffic, not revenue. | 20 % | Observed café activity: total Google review count of cafés in the area (Apify Google Maps scrape). Label SIGNAL — reviews track visitor traffic, not revenue. |
| Demand per café Saturation signal: Google review count per existing café in the area (needs ≥3 cafés). High = existing cafés attract many visitors each. Label SIGNAL. | 15 % | Saturation signal: Google review count per existing café in the area (needs ≥3 cafés). High = existing cafés attract many visitors each. Label SIGNAL. |
| Residents 18–49 Residents aged 18–49 in the area, Zensus 2022 100 m grid. Label FACT. | 15 % | Residents aged 18–49 in the area, Zensus 2022 100 m grid. Label FACT. |
| Stability Gastronomy business stability: share of gastronomy businesses in the IHK register that closed between 2023-07 and 2026-07 (inverse). Label FACT for the rate; it is gastronomy overall, not cafés specifically. | 15 % | Gastronomy business stability: share of gastronomy businesses in the IHK register that closed between 2023-07 and 2026-07 (inverse). Label FACT for the rate; it is gastronomy overall, not cafés specifically. |
| Transit Public-transport departures per weekday at stops inside the area (GTFS in LOKLU v3_transport_frequency). Label FACT. | 10 % | Public-transport departures per weekday at stops inside the area (GTFS in LOKLU v3_transport_frequency). Label FACT. |
You can shift the weights by telling LOKLU what matters most, and limit the screening to districts.
A transparent way to decide where to look first. The shortlist is a RECOMMENDATION; you decide.
Not a prediction of success, and not LOKLU's 20-dimension address score from Classic. Suitability, not a success forecast.
LOKLU has only about 47 Berlin retail and gastronomy rent listings — too thin for rents per area. Residential rent (Zensus) is shown as a proxy only, never as commercial rent.
There are no measured pedestrian counts in this snapshot. Café review volume is used as an activity SIGNAL instead.
Closure rates cover gastronomy overall in the IHK register, window 2023-07 to 2026-07 — not cafés specifically.
Google sets a premium price level for only 17 Berlin cafés, so premium demand is approximated through the social-status class (ESIX).
The screening covers 24 business categories. Each carries a data-quality grade (good, medium or thin) from how well Google, OpenStreetMap and the IHK register agree, and the café keeps the exact weights described above; other categories use weights from their profile. Budget benchmarks exist for cafés only. Categories
Every data source behind the answers, with its freshness and the status of each collector, is on the Sources page.