The zone where orders pile up but no store reaches within 15 minutes shows up on the map

Three spreadsheets: the address column full, the coordinate column blank

This company's data is already well organized. Stores and customers, orders and competitors sit as connected entities inside the workspace knowledge graph. Which customer bought how much from which store, which competitor sits in which trade area — all of it tied together by relationships. Just one thing is missing: where each of those places actually is on a map. The coordinates.

• One store-list spreadsheet. The address column is filled in and the coordinate column is blank
• One spreadsheet of a year of customer orders. It holds which customer bought how much from which store
• One competitor spreadsheet. It records which competitor sits in which trade area
• The stores, customers, orders, and competitors in those three sheets already sit as connected entities in the knowledge graph. There's no new data to gather
• There are hundreds of addresses to put on the map. All of them address strings, not one coordinate

Without coordinates, questions asked by location stall. "Which neighborhood has the most orders with no store of ours within 2 km?" "How far does each store reach in 15 minutes, and where are the holes between those reaches?" The answers to these aren't written in any single row.

Drop pins one by one, estimate by straight-line distance

The old way was simple but grueling. Paste each address into a map service one at a time, drop a pin, screenshot the screen, and paste the image into the deck. Move hundreds of addresses like that and time melts away — and one typo lands a pin in the wrong neighborhood entirely.

Distance was a guess, too. People judged "near" and "far" by the straight line between two points, but real people don't move in straight lines. If there's a river to cross or a main road to go around, two kilometers as the crow flies is a long drive. The moment you draw reach as a straight-line circle, the coverage judgment is already off.

The most frustrating part is that a map made this way is cut off from the data. Click a pin in the screenshot and you can't get back to who that customer is or which order it came from. The map ended as an image pasted into the deck, and you couldn't ask it anything more.

Don't draw it — ask for the map in chat

Setup isn't gathering new data; it's giving coordinates to the places already in the graph. There's no map button or sidebar icon anywhere in the app. A map only exists once the agent draws it during a chat turn.

Ask on the map, trace back through the graph

"How far does each store reach in 15 minutes, and where are the holes between those reaches?"
→ Each store's 15-minute reach is drawn, with the order hotspots overlaid on top.
   Zones remain where the hotspot is clear but no store's reach covers it.
   Empty trade areas: demand exists, but we don't reach it.

"Which neighborhood has the most orders with no store of ours within 2 km?"
→ A zone rows could never show surfaces as one spot on the map.
   Click that zone and a "Selected: {label}" pill appears,
   and if the mark is tied to an entity, a "View in graph" button comes with it.
   One click and you cross straight to which customers those orders came from.

"Which competitors are within 2 km of this candidate site?"
→ The answer comes right there, with no re-pinning.
   A place put on the map remembers its coordinates back in the graph.

The reason this answers is a question of who does the computing. The model decides what to draw; deterministic services and libraries compute the coordinates and the reach. So an empty zone on the map is not a spot the model guessed at — it's what fell out of real travel times. In a meeting choosing where to open next, that zone goes on the candidate list with its evidence.

Here the map doesn't end as a picture. Click a mark and an @mention chip drops into the chat composer, and your next message carries that feature's properties and coordinates along with it. You never have to describe which dot you meant. You find the "where" on the map and follow it to the "who and what" in the graph. Computed results like hexbin cells and isochrones, though, have no entity behind them, so no "View in graph" button is attached.

What changes

Before, you scanned rows of addresses by eye and guessed "probably around here," dropping pins by hand and measuring distance in straight lines. Now the reach and the hotspots are computed on the map, and the person looks at that result and judges where to open.