Skip to content
Shyenova

Area 05

How many of your decisions are being settled by a circle drawn on a map?

A straight-line radius around a location is not a catchment. Almost every site decision made in India uses one anyway.

15–40%

Catchment overstatement

Typical error from using a radius rather than a drive-time isochrone in a barrier-heavy geography.

10–25%

Travel time

Reduction from a properly optimised territory and route structure over an inherited one.

10–30%

Revenue cannibalised

Share of a new site's revenue commonly transferred from sites you already own.

These are the improvement ranges these levers are known to produce in industry where the underlying practice has not been examined recently. They are targets to test against your own baseline, not claims about named past clients.

There is a class of business question where the answer is determined almost entirely by geography: where to put the depot, which territory each rep should cover, which customers a new store will take from an existing one, which route the vehicle should run. These get decided constantly, and they get decided badly, because the analysis that would settle them properly is thought to be harder than it is.

The usual failure is substituting a circle for the real thing. A ten-kilometre radius is drawn around a candidate site and treated as its catchment, when the actual catchment is bounded by a river with no crossing, a highway nobody can turn across, and a competitor two kilometres closer to half the population inside the circle. Doing this properly requires road networks, travel times and a demand model, none of which is exotic, all of which changes the answer.

What this is not

We do not resell Esri or any other platform, and we are not a mapping bureau producing cartography to order. The work is the analysis and the decision it settles; the map is how it gets communicated.

Artificial intelligence

Where machine learning earns its place in spatial work

Less often than in the neighbouring practices, and it is worth being straight about that. Site selection and territory design are optimisation and demand-allocation problems with well-understood methods that outperform learned approaches on the data volumes most businesses have. Where learning does earn its place is in geocoding messy Indian addresses, in imputing demand where you have no observation, and in extracting features from imagery, which is really the remote sensing practice.

How we approach applied AI

What an engagement looks like

A representative shape.

Illustrative. This is the shape of work of this type, not an account of a specific client. We do not publish named results, and we would be wary of a firm that publishes them without the client in the room.

10 weeks

  1. 01Situation

    A distribution business was assessing eight candidate locations using ten-kilometre radius catchments, and had recently opened two sites after which nearby existing sites declined without anyone having forecast it.

  2. 02What we did

    Geocoded three years of transaction data to customer origin, built drive-time catchments calibrated against where customers demonstrably came from, and fitted a gravity model to allocate demand between competing sites.

  3. 03Where it landed

    Modelled catchments were substantially smaller and differently shaped than the radii implied. Two candidates that had ranked highly on gross population fell sharply once cannibalisation was subtracted, and the ranking changed at the top.

Questions

About gis.

Our addresses are messy. Can you still geocode our customers?

Usually, and it is one of the harder parts of this work in India. Unstructured addresses, inconsistent locality names and large PIN code areas all reduce match rates. Seventy to ninety per cent at usable precision is typical, with the remainder aggregated to PIN or ward level rather than discarded. We report the match rate and its spatial bias, because unmatched addresses are rarely randomly distributed.

Do we need to buy a GIS platform?

No. Everything we build runs on open source and open standards, and we sell no licences. If you already have enterprise GIS, keep it. We would change the storage and interchange layer so the authoritative data is portable, with your existing platform as one client among several.

How accurate can a revenue forecast for a new site be?

Plus or minus ten to twenty per cent on first-year revenue for a calibrated model in a format you have many existing sites in. For a new format or a new region with nothing to calibrate against, considerably worse, and we say so. The model is far better at ranking eight candidates than at predicting what the best one will earn.

Will our field team accept a territory redesign?

Only if it is designed with their resistance in mind, which is usually rational, someone is losing accounts they built. We model the disruption of each individual move, drop the ones costing a lot of goodwill for little gain, and give management the numbers for the ones worth doing anyway.

Start a conversation

Talk to us about gis.

Most useful first conversations are thirty minutes and involve no proposal. Describe the problem you are looking at and we will tell you honestly whether it is one we can help with.

Or write directly to hello@shyenova.com