Transformation · Banking

Breaking district deposit data down to the branch catchment.

A private bank could see deposit balances by district and nothing below it. We estimated the deposit base around each branch, removed what competitors already hold, and drew territories with goals and market share attached.

Private bank

Branch deposits

Urban and semi-urban

The model, in four numbers

Accuracy is carried from prior deposit engagements. The overlap test is the gate this engagement agreed to before scale-up.

90%
Validated, prior engagements
Accuracy for top three decile deposit clusters.
>70%
Target
Blind overlap test against the bank's actual deposit-dense locations.
~8,900
Scope
Hexagons per district, each 0.7 km².
33+
Scope
Signal columns delivered per hexagon.

The case

01 · No branch-level share

Two branches, one number

Branches in the same district appear identical in published deposit data, whether they hold a fifth of the market or a twentieth.

02 · Invisible boundaries

Nobody could size the catchment

Without spatial decomposition a branch manager cannot tell whether untapped deposit sits two kilometres away or twelve.

03 · Uneven competition

Rivals do not spread evenly

Competitor density varies block by block, and smaller markets sit furthest from the district average.

Four steps from a district total to a branch catchment

The published district figure is the anchor. Everything below it is estimated, then reduced by what rivals absorb.

01

Ingest

Published district deposit balances across savings, current and term set the addressable pool.

→

02

Decompose

The total is distributed across hexagons using population, vibrance, business density and formalisation.

→

03

Assign

Hexagons join to branch catchment polygons on the spatial index the bank already uses.

→

04

Net off

Competitor branches are modelled as gravity wells. What remains is net addressable potential per hexagon.

What arrives per hexagon

One row per hexagon, delivered as a flat file that joins to the bank's own branch table on the spatial index.

Field

What it carries

How it is used

Deposit potential

Estimated total addressable deposit in the hexagon.

Aggregate to catchment for branch-level addressable market

Savings, current, term split

The estimate broken into its three components.

Point commercial branches at current account density

Catchment ring

Distance band from the nearest branch counter.

Separate walk-in potential from outreach potential

Nearest competitor tier

Classification of the closest rival branch.

Set the competitive response for the zone

Vibrance and population scores

Economic activity and density indices.

Filter out hexagons too thin to justify field cost

Market tier flag

Whether the hexagon runs the semi-urban and ruralmodel.

Apply the scoring weights that market needs

Semi-urban and rural hexagons run a separate model. Formalisation and road access carry higher weights there, because deposit acquisition works differently in those markets.

Every hexagon carries a class

Potential and competition together decide the instruction.

Class

What it means

Action

Blue Ocean

Strong potential, weak competitor absorption.

Deploy first

Battleground

Strong potential, strong rivals present.

Contest on service and rate

Fortify

Existing share concentrated here.

Defend against leakage

Caution

Potential present, structural risk elevated.

Enter with limits

Graveyard

Neither potential nor headroom.

Remove from field plans

What each team received

One output per role, and a file the data team can join without a new system.

Leadership

Like-for-like branch comparison

Urban and semi-urban branches compared on estimated potential against actual performance, so a small branch in a thin market stops looking like an underperformer.

Branch heads

A catchment insight pack

Total addressable deposit in the catchment, how much sits in each distance ring, and which product the composition calls for.

Data team

A joinable flat file

One row per hexagon keyed on the spatial index, delivered as parquet or CSV so it lands in the bank's existing platform rather than a separate tool.

How the workflow changed

Before

Deposit potential known only at district level

Branch performance judged without a market denominator

Catchment boundaries assumed rather than measured

Competitor pressure treated as uniform across a district

Semi-urban branches scored on urban assumptions

After

Potential estimated for every hexagon in the catchment

Performance read against estimated addressable market

Catchments drawn from distance rings and competitor pull

Competitor absorption modelled branch by branch

Semi-urban and rural hexagons run their own model

How the model gets checked

A blind test, designed before the pilot ran.

Pilot

The model runs across five to ten branches and produces ranked hexagons.

Blind

Test

EPIC's top-decile hexagons are compared against the bank's own deposit-dense customer locations, which EPIC has not seen.

GATE

Overlap between the two sets.

>70%

Decision

Clearing the gate opens the scale conversation. Missing it ends the engagement.

Branch performance is varied drastically. Are we doing badly in a good market? How should we staff each branch? What should be their goals? Are we missing some areas? How do we increase Current Accounts?

Bank Leadership
During scoping

Test it blind on five branches.

Share branch locations and aggregated deposit performance.
We will rank the hexagons, and you can check our top decile against customer locations we never saw.

EPIC Intelligence · Client identity withheld