FlowsBook a callSign inCreate free accountSign up
Business and places data

How to sample a market when a city is too big to list in full

In a city of millions, a full list of every shop is slow and often unnecessary. How to pick sample areas, count inside them and scale up to a fair estimate.

Part of the guide Location intelligence automation

By the xMap team · · 5 min read

When a city is too big to list every business of a kind, you sample it: you choose a set of small areas spread across the whole city, count the businesses inside each one, work out how dense they are, and scale that density up to the city's area, district by district. Done carefully, twenty or thirty well-placed sample circles give an estimate good enough to compare cities, size a market or set targets, in a fraction of the time a full list takes.

This article is for analysts, strategy teams and consultants who need a market size for a very large city, and for anyone who has run a search, hit a limit and wondered what the result really covers. It explains why one big search is not a sample, how to design a fair one, a worked example with made-up counts, the traps, and how to have an AI employee run the counts for you.

Why one big search is not a sample

Most searches for places return the nearest ones first and stop at a limit. That is the right behaviour for finding what is around a point, but it means a single wide search from the middle of a city describes the middle, not the city.

xMap's own market figures show the effect clearly. Its places database lists 3,792 restaurants within 10 km of the centre of London (November 2026), but the chains, independents, ratings and openings on that page are worked out from the 326 nearest the centre. Of those 326, 50 belong to a chain and 276 are independent. That is a fair description of central London. It would be a mistake to assume the same share holds in the outer boroughs, where the mix of high-street chains and family restaurants can be quite different.

How to design a fair sample

  1. Split the city into zones. Use districts, or a simple ring model: centre, inner ring, outer ring. Zones should differ from each other in how built up they are.
  2. Place sample points inside each zone. Spread them evenly, not on the busiest street. A grid of points, or points at random, both work; what matters is that you do not choose them by eye.
  3. Use small circles. A circle of 1 km around each point (about 3.14 km²) is small enough that a search rarely reaches its limit, so the count is complete inside it.
  4. Count, do not list. For each circle you need the number of places, and perhaps how many belong to a chain and the average rating. You do not need every name.
  5. Work out the density per zone. Average the counts of the circles in a zone and divide by the circle's area.
  6. Scale up. Multiply each zone's density by the zone's area, and add the zones together.

A worked example (made up)

Say an analyst wants the number of pharmacies across the whole of Jakarta, a city far too large for one search. They split it into three rings and place 10 sample points in the centre, 12 in the inner ring and 14 in the outer ring, each with a 1 km circle. The counts below are invented to show the arithmetic.

ZoneSample circlesAverage count per circleDensity per km²Zone area km²Estimate
Centre10227.060420
Inner ring12134.12501,025
Outer ring1451.6350560
Whole city36660about 2,000

Made-up counts for an imaginary sample. Density is the average count divided by 3.14 km²; the estimate is density times zone area, rounded.

Notice what would have happened with a single search from the centre: the first 500 pharmacies would all have come from the centre and the inner ring, and the outer ring, which holds over a quarter of this made-up total, would barely appear.

Telling a good estimate from a lucky one

  • Look at the spread within each zone. If one circle has 40 and its neighbour 2, the zone is mixed and needs more points or a split.
  • Repeat with a second set of points. If the total moves a lot, you need more circles.
  • Report a range, not a single number. "Between 1,800 and 2,200" is more honest than "2,013".
  • Check one zone in full. List every place in one district and compare with what the sample predicted for it.

Mistakes to avoid

  • Putting sample points on main roads. Shops line main roads; circles there overstate density for the whole zone.
  • Circles that hit the limit. If a circle returns 500, the true count is higher and the circle is too big for that zone.
  • Counting water and parks as zone area. A river or a large park inflates the area you multiply by. Leave them out, or draw zones around them.
  • Applying one city's mix to another. The share of chains in central London says nothing about central Lagos or central Lima.
  • Mixing types. Count each type of place separately; "shops" as one type hides the answer.

How to do it with an AI employee

An AI employee can run the counts on a timetable, so the estimate refreshes each quarter without anyone redrawing circles. The template Market size of a category across cities gives the comparison across cities; for a single city sample, the flow looks like this:

  1. My data: a table of sample points, each with a zone name and its position.
  2. Nearby places: the type you are sizing, around each incoming place, within 1 km, passing on each incoming place with counts.
  3. Calculated field: density = nearby_count / 3.14.
  4. Data analyst: average the density by zone, multiply by each zone's area and give the total with a range.
  5. Email to the strategy team, with a map of the sample circles.

What lands in the inbox is a short table per zone (circles, average count, density, estimate), the city total with a range, and a map that shows where the circles were. Running the same flow on a schedule shows whether the market is growing, zone by zone.

Questions

How many sample circles do I need?

Enough that each zone has several and a second set of points gives a similar total. For a large city, a few dozen 1 km circles across three or four zones is a reasonable start.

Why 1 km circles?

They are small enough that a search rarely reaches its limit of the 500 nearest places, so each count is complete, and large enough to hold a fair number of places in most built-up areas.

Can I just list the whole city instead?

Sometimes. Nearby places searches each city or area of the Area of interest from its middle to its edge, up to 25 km, and Map business scan can cover an area with a grid of searches. For very large or very dense cities, sampling is faster and the estimate is usually good enough.

Does sampling work for comparing two cities?

Yes, as long as both are sampled the same way: the same zone rules, circle size and type of place.

Templates that do this

Read next

More on business and places data · Every article · Templates

Put it to work

An AI employee in xMap Flows does this on live data, on its own timetable. Start from a template or describe it in plain words.