New Store Site Selection Model

A new store site selection model scores candidate locations before you sign a lease or commit capital. It combines footfall, catchment demographics, competitor density, mobility patterns, rent, and expected sales transfer from nearby stores into a ranked shortlist. The model does not pick the site. Real estate and planning teams still decide which deals to pursue, which to renegotiate, and which to walk away from.

Planners use the score to compare like-for-like options across a market, surface sites that look strong on rent but weak on demand, and flag locations where a new store would mainly pull sales from existing ones. The output is a ranked list with drivers, not a binary approve or reject.

The model only runs when catchment and competitor inputs are present. If either is missing, it returns empty output rather than a partial score. That rule keeps weak candidates from looking stronger than they are because a field was blank.

Core inputs typically include:

• Footfall and mobility: Pedestrian counts, dwell patterns, origin-destination flows, and time-of-day profiles for the trade area.
• Catchment demographics: Population, income bands, household composition, and demand proxies for your category within a defined drive or walk radius.
• Competitor density: Count, proximity, and format of competing stores that already serve the same catchment.
• Rent and occupancy cost: Asking rent, estimated CAM or rates, and lease structure where known.
• Network context: Distance and overlap with your existing stores, plus historical sales transfer when similar openings have occurred.

Optional enrichments help once the baseline is solid: parking access, transit adjacency, visibility from major roads, and local employment density. They refine ranking; they do not replace catchment or competitor coverage.

When a data source is stale or incomplete for one candidate, exclude that site from scoring until the gap is closed. Ranking incomplete records against complete ones creates false precision.

Most site models produce a composite score that balances demand potential against cost and cannibalization risk. Demand-side features estimate how much category spend the catchment can support. Cost-side features normalize rent against expected volume. Network-side features estimate how much of that volume would come from your own nearby stores rather than from competitors or new demand.

Sales transfer is the piece that often changes the decision. A site can show strong independent demand and still be a poor network move if it sits inside the primary catchment of a high-performing store. Transfer estimates usually draw on distance decay, overlapping catchments, and historical opening outcomes. Treat them as planning ranges, not guarantees. Local conditions, assortment differences, and competitor reactions all shift the realized transfer after opening.

Human review focuses on the drivers behind the rank. A top-ranked site with thin competitor data, an aggressive rent assumption, or an untested mobility source should be challenged before it enters the deal pipeline. A mid-ranked site with clearer evidence and cleaner transfer risk may be the better commitment.

Use the ranked list at three points in the planning cycle.

Market screening. When several candidates appear in the same metro or district, score them on a common feature set so the shortlist reflects comparable evidence, not the loudest broker pitch.

Deal negotiation. When rent or term structure is still open, re-score after each material change. A rent reduction can lift a borderline site; a longer exclusivity clause can change the competitor picture around it.

Network sequencing. When capital is limited, rank openings by contribution after transfer, not by headline catchment size alone. That keeps early openings from starving later ones in the same trade area.

The model also supports post-opening review. Compare predicted demand and transfer against early sales. Large, consistent misses point to feature gaps (for example, under-counting lunchtime office traffic) that should update the next scoring run.

Keep a clear handoff between model output and real estate judgment. The model ranks. The team selects.

Practical controls:

1. Gate incomplete sites. Empty output when catchment or competitor inputs are missing. Do not backfill with market averages just to produce a number.
2. Require driver transparency. Every ranked site should show which features lifted or lowered the score, including estimated sales transfer from named nearby stores.
3. Separate score from approval. Site committees review rank, drivers, lease terms, construction risk, and brand fit. A top score without an acceptable lease is still a pass.
4. Version the feature set. When mobility providers, demographic vintages, or competitor panels change, re-score open candidates so older ranks do not linger as stale truth.
5. Document overrides. If planners select a lower-ranked site for strategic reasons (anchor adjacency, landlord relationship, format test), record the reason so later reviews do not treat the override as model failure.

This workflow keeps machine ranking useful without pretending location strategy is an automated lease decision.

A healthy site selection process produces fewer surprises after opening and fewer deals advanced on incomplete catchments. Teams spend less time debating gut feel on every broker package and more time pressure-testing the few sites that clear the evidence bar.

Watch for these failure modes:

• Scoring sites with missing competitor maps and calling the result “directional.”
• Optimizing for raw footfall while ignoring category-relevant demographics.
• Ignoring transfer and then blaming the model when an opening cannibalizes a strong neighbor.
• Freezing ranks for months while rents, openings, and traffic patterns move.

When catchment and competitor coverage are complete, transfer assumptions are explicit, and real estate still owns the final choice, the model becomes a durable planning tool rather than a one-off spreadsheet exercise.