Client Case Study · Retail Site Selection & Market Intelligence

When the portfolio undersells the site

A national home-improvement retailer's own market curve could mathematically never predict a R30 million store, no matter how strong the site. Rebuilding the model found out why.

17
stores modelled across 36 months
7 of 24
stores clearing the R30m sales target
R29.15m
average store sales across the set
1
new power-regression curve built
The problem

A model built on underperformers keeps predicting underperformance.

Before committing capital to a new site, the client reads expected sales off a market curve, fitted on its own store portfolio. But when that portfolio includes a lot of underperforming stores, the curve inherits their weakness and passes it on to every future forecast.

7 of 24 stores
In the banner under review actually cleared management's R30 million annual turnover target.
R10.85m spread
Standard deviation across the modelled stores, too much variation for market size to explain alone.
0 ceiling
The original curve could mathematically never output a R30 million forecast, for any site, however strong.
What the data showed

Market size alone does not predict store performance

Stores were grouped into six market-size bands so each could be judged against genuine like-for-like peers. The top-performing store in each band tells its own story.

Top store's annual sales in each market-size band
Above R550 million
Largest market band · 2 stores
R53.2m
R350m – R450m
3 stores
R44.1m
R450m – R550m
2 stores
R41.9m
R150m – R250m
5 stores · out-earns the band above it
R36.3m
R250m – R350m
1 store · smaller market band, weaker result
R19.9m
R50m – R150m
Smallest market band · 4 stores
R18.0m
The R150m–R250m band's top store out-earned the entire R250m–R350m band, despite sitting in a smaller market, direct evidence that market size alone cannot explain performance.
What the evidence showed

Several stores scored well above normal statistical variation, evidence of a genuinely higher-performing tier within the portfolio, not just noise to be averaged away.

The structural flaw

With most of the portfolio below the R30 million target, the original curve was mathematically incapable of ever forecasting a R30 million store.

Proof of the problem

Same-sized markets, wildly different results

Within a single market-size band, the gap between the weakest and strongest store often rivals the R30 million target itself, a spread market size cannot explain on its own.

R350m–R450m band · widest spread
R24.8m
Weakest store
R44.1m
Strongest store
R19.3m spread
Above R550m band
R31.2m
Weakest store
R53.2m
Strongest store
R22.0m spread
R150m–R250m band
R25.7m
Weakest store
R36.3m
Strongest store
R10.6m spread
How we get there

Our market-sizing method

The same four-step process behind this result, adapted to your footprint, your banners, and your growth plans.

1

Segment

Group stores into like-for-like market-size bands so performance comparisons are genuinely fair.

2

Detect

Run outlier detection to separate genuine high performers from statistical noise.

3

Refit

Rebuild the sales curve with a model shaped to match real portfolio performance, not just the average.

4

Disclose

Flag the model's real limitations honestly, so decisions get made with eyes open, not false certainty.

Is your site-selection model quietly underselling your best locations?

If your market curve is built on a portfolio that includes underperformers, it may be doing exactly the same thing. Let's find out.

Start a conversation: flowvisualpatterns.co.za
Flowing Visual Patterns · Market intelligence & site-feasibility modelling