Skip to main content
Locatalyze
Sign in

Data and Methodology

How we observe, estimate and model.

Locatalyze separates external evidences, customer assumptions and modelled outputs. A rules-based engine calculates the score and recommendation and AI helps explain what the result means and what to verify next.

Start a free location analysisView sample report
Methodology
Engine 7.4.0
Last reviewed
18 September 2026
Typical report
typically 5–10 minutes

Method at a glance

A Traceable path from Address to Report.

Locatalyze has two related surfaces. Address assessments research your specific site and return a Proceed / Verify / Avoid decision with labelled estimates. Suburb and city analyse pages use a separate rules-based Location Score. In both cases, evidence gathering, calculation and written interpretation stay separate, so the written narrative never quietly replaces a number.

  1. 01

    Define the decision

    You provide the address, business type, proposed rent and any commercial assumptions you know.

  2. 02

    Resolve the location

    The address is geocoded and nearby providers are queried around the confirmed pin.

  3. 03

    Classify the evidence

    Each input is treated as observed, user-provided, estimated, proxy-based or unavailable.

  4. 04

    Score this business

    The research agent names the components that decide this business, gives each a weight and a score, and sets the overall score to their weighted average. The weights add to 100. A fixed rent, competition, demand, profitability and location-quality split is not used.

  5. 05

    Apply decision gates

    Economics, coverage, contradictions and missing commercial inputs can make the recommendation more conservative.

  6. 06

    Explain the result

    The same run writes the risks, conditions and verification actions. A report is rejected when the component weights do not add to 100, or when their weighted average is more than 2 points away from the overall score.

  7. 07

    Support due diligence

    The report shows what to validate before signing rather than presenting the output as professional advice.

Evidence language

One label for every kind of input.

These terms distinguish where a value came from. A source label describes provenance; confidence describes how much the report can rely on it.

User-provided input
Entered by the customer and accepted as an assumption; not independently verified by Locatalyze.
Observed external data
Returned by a named provider or public dataset for the location or surrounding area.
Benchmark / estimate
A category, area or listing-based estimate used when a site-specific value is unavailable.
Area estimate (proxy)
An indirect signal used in place of a direct measurement, such as nearby activity standing in for footfall. It is not a people count at this door.
Modelled scenario
Calculated from inputs, benchmarks and deterministic formulas; it is not an observed trading result.
AI-written
Narrative generated from structured results, risks and validation checks.
Missing / unavailable
The system could not resolve the data reliably enough to present it as evidence.

What goes into a report

Every input has a Source and a Limit.

Site-specific does not always mean site-measured. This matrix shows the normal primary source, fallback, geographic precision and decision effect for each major input category.

InputSource and fallbackPrecision and limitationReport effect

Address and coordinates

User-provided inputObserved external data

Typed address plus latitude and longitude returned after confirmation.

PrimaryLocationIQ geocoding and autocomplete; Mapbox renders the map.

FallbackExisting submitted coordinates or pipeline geocoding where available.

Typical precisionStreet-level when the geocoder resolves a valid pin; otherwise area-level context is disclosed.

Main limitationA geocoded pin is not a cadastral survey and can be wrong for suites, arcades or new developments.

All radius searches, the map and location-quality signals.

Competition

Observed external dataArea estimate (proxy)

Nearby businesses, category match, rating, review volume, distance and listing status where returned.

PrimaryGoogle Places, with Geoapify/OpenStreetMap support; Foursquare is used for fitness categories.

FallbackCategory-specific provider fallbacks, including OpenStreetMap coverage for some fitness searches.

Typical precisionThe scored count is anchored to 500 m; provider discovery can search a wider area.

Main limitationListings can be missing, duplicated, miscategorised, closed or not yet published.

Competition score, saturation, opportunity notes, and named competitors with distances. The assessment does not include a competitor pin map.

Demographics

Observed external dataBenchmark / estimateMissing / unavailable

Population, household and income context when the relevant Australian public-data lookup resolves.

PrimaryABS Census 2021 data and stored Australian geographic mappings.

FallbackBroader area context or an explicit unavailable state.

Typical precisionSA1, postcode or suburb context depending on what resolves for the report.

Main limitationArea demographics do not describe the exact frontage, daypart or customer catchment.

Demand context, customer fit and confidence.

Demand and street activity

Observed external dataArea estimate (proxy)Benchmark / estimate

Demand trend, nearby anchors, transit, local activity and category-fit signals.

PrimaryPipeline market signals plus pin-anchored place, anchor and transit data.

FallbackCompetitor-density and category proxies; the demand weight is reduced when the signal is proxied.

Typical precisionDirectional trade-area context, not a verified pedestrian count at the door.

Main limitationMost reports do not contain a measured, continuous foot-traffic series.

Demand score, location quality, opportunity notes and confidence.

Rent pressure

User-provided inputBenchmark / estimateModelled scenario

Submitted monthly rent, estimated comparison context and rent as a share of modelled revenue.

PrimaryThe customer’s rent quote plus commercial-listing and area/category signals returned by the pipeline.

FallbackCategory cost-stack benchmarks when current comparable evidence is thin.

Typical precisionThe quote is site-specific; the comparison context is usually suburb or corridor level.

Main limitationPublic listings are not a complete rent database and rarely capture incentives, outgoings or lease condition.

Rent score, rent-burden status, recommendation gates and financial scenarios.

Revenue and costs

User-provided inputBenchmark / estimateModelled scenario

Ticket size, customers, COGS, labour, rent, other operating costs, setup budget and working-capital assumptions.

PrimaryUser inputs plus location-agent estimates checked against Australian category benchmarks.

FallbackCategory defaults, clearly labelled and subject to more conservative recommendation rules.

Typical precisionMonthly model outputs; low-confidence values are displayed as wider bands.

Main limitationThese are commercial estimates, not audited accounts, quotes or guaranteed performance.

Profitability score, break-even, payback, scenarios and recommendation.

Data confidence

Modelled scenario

Feed presence, provider coverage, benchmark reliance and material corrections made by the engine.

PrimaryDeterministic confidence and validation logic.

FallbackA lower confidence tier, wider display range or missing-data state.

Typical precisionA report-level percentage plus topic-level confidence labels.

Main limitationConfidence describes evidence coverage; it is not an accuracy guarantee.

How figures are shown and whether a positive recommendation is allowed.

Location Score

Five dimensions. One rules based composite.

Each dimension is scored from 0–100, multiplied by its current weight and added. The base weights total 100%; proxy demand uses a deliberately smaller contribution.

Printed on the report

The address score is the weighted average of the components on that report. The research agent chooses the names and the weights for that business. The weights add to 100. If they do not, or if the weighted average is more than 2 points away from the overall score, the report is rejected. There is no fixed rent, competition, demand, profitability and location-quality split.

How to read a score

The components on the page are the model.

Each component has a name, a weight and a score from 0 to 100. Multiply score by weight, add those products, and divide by 100. That is the overall score, within 2 points. A workshop and a cafe are not scored on the same list. Two nearby sites can differ because the pin, the competitors, the rent and the evidence differ.

Score versus recommendation

Three outputs answer three questions.

The score describes the weighted location read. The recommendation asks whether the current commercial case supports action. Confidence describes the evidence behind both.

01

Location Score

How the five dimensions combine.

A weighted 0–100 composite. It is rounded after the axis contributions are added.

02

Recommendation

What the current case supports.

A Proceed / Verify / Avoid decision after evidence quality, commercial conditions and trust gates are applied. Address assessments are research-led; suburb Location Scores remain rules-based.

03

Data confidence

How much evidence backs the read.

A separate coverage and model-quality signal. It is not added to the Location Score.

PROCEED60–100 base score

The score, economics and evidence gates support progressing, subject to normal lease and professional due diligence.

VERIFY40–59 base score

The site may work, but one or more commercial, evidence or competition checks must be resolved first.

AVOID0–39 base score

The current inputs show a material mismatch in economics, competition, location or the operating model.

INSUFFICIENT DATANo decision-grade base

Core commercial inputs or critical report evidence are missing, so the product withholds a normal recommendation.

Why the recommendation can be harsher

A good composite cannot hide a decision-critical weakness.

The score band is a ceiling on how favourable the recommendation can be. The confirmed rules below only keep or lower that result; they do not turn a weak score into PROCEED.

  • Benchmark-only revenue or missing ticket/setup inputs block a PROCEED recommendation.
  • If both location/rent and demand feeds are missing, the result cannot read PROCEED.
  • Data completeness below 35% cannot support a positive display verdict; the engine’s stricter hard gate requires more than 45% for PROCEED.
  • Unknown or zero-result competitor coverage is treated as unverified, not as evidence of a blue ocean.
  • A location-grounded loss cannot be turned into PROCEED solely because a category benchmark is more optimistic.
  • A user budget below 75% of a credible recommended setup blocks PROCEED; below 50% forces AVOID.
  • Negative base-case profit, serious rent pressure, fatal competition/profitability reads or validation contradictions can make the final recommendation harsher than the score band.
  • Where no demand reading is measured at all, the demand axis is excluded from the composite rather than filled with a neutral placeholder, and the result cannot read PROCEED.
  • Capital that cannot be recovered inside the committed lease term, including a site that never recovers it at the modelled net, blocks PROCEED. Where the fit-out figure is a category estimate rather than your own, the result is capped at VERIFY rather than issuing an outright AVOID.
  • Business categories with no category benchmark (for example medical or professional services) return INSUFFICIENT DATA. No commercial recommendation is issued on invented category economics.

Financial methodology

A modelled Estimate, not a Promise.

The paid financial view combines customer assumptions, category benchmarks and validated pipeline estimates. It calculates revenue, costs, break-even, sensitivity and payback, but it is not a full set of accounts.

Base revenue

daily customers × hours multiplier × access multiplier × average ticket × 30

User ticket and calibration inputs take priority; category values fill gaps. Location-agent estimates are plausibility-checked and may be blended with benchmarks.

Gross margin and COGS

revenue × category gross-margin / COGS rate

Category rates are defaults, not supplier quotes. Implausible upstream values are corrected or rejected.

Monthly net

modelled revenue − COGS − labour − rent − other operating costs

This is a commercial estimate rather than a full accounting P&L. Tax, finance structure and owner circumstances are not fully modelled.

Break-even customers per day

fixed monthly costs ÷ contribution per transaction ÷ 30

Contribution per transaction is average ticket after variable COGS. The value is withheld if the required inputs are not usable.

Payback period

capital at risk ÷ positive monthly net profit

Capital at risk is your fit-out budget less any landlord contribution. The figure is shown only when you supplied a budget and monthly net profit is positive. Where the budget is a category estimate, the number is withheld to avoid false precision, but payback is still assessed against the estimate, so a site that could not repay a typical fit-out is flagged rather than passed.

Occupancy cost

face rent + outgoings

Face rent alone understates the commitment. Where outgoings are not disclosed, a conservative category allowance is imputed and labelled, so withholding the figure never produces a cheaper-looking site.

Lease liability (total and NPV)

occupancy cost escalated annually over the committed term, discounted at 8% p.a. pre-tax

This is the cost of occupying the premises over the term. It is a liability, not a return. Escalation defaults to 3.5% p.a. where the deal terms are not supplied. Both the discount rate and the default escalation are modelling conventions, not readings from your lease.

Scenario range

worst 65% revenue · base 100% · best 140%

Variable costs move with revenue. Fixed costs stay fixed, with a small stress/efficiency adjustment. These scenarios are sensitivities, not forecasts.

Workable rent band8–18%

Engine estimate from the category cost stack, not a universal lease rule.

Healthy
Up to 14%
Watch
14–18%
Risky
Above 22%

Free location decision

What you receive before an unlock

  • Rent as % of revenue vs an estimated category range
  • Rent per sqm vs estimated asking comps
  • Timeline to lease expiry or review
  • Up to 3 findings from rent and revenue only
  • Not a stay / leave recommendation

Paid report · $29

What the financial unlock adds

  • Verdict and the highest rent you can afford
  • P&L, break-even and safety margin from your costs
  • Proposed / midpoint / walk-away rent outcomes
  • Shareable landlord evidence page (held until comps are sourced)
  • 90-day plan with dollar ranges

What AI does

AI explains the model. It does not own the decision.

The language layer is useful for turning structured outputs into plain-English risks and actions. Keeping its boundary visible prevents readable prose from being mistaken for numeric evidence.

AI may

Explain structured evidence.

  • Turns structured scores, risks and scenarios into readable narrative.
  • Summarises competitive positioning and SWOT-style interpretation where supported.
  • Drafts conditions, failure modes and actions to verify next.
  • May contribute upstream estimates that are labelled and checked against deterministic bounds before use.

AI does not

Control the decision rules.

  • Use a fixed five-way weight table. The components and weights printed on the report are the score for that run, and they must add to 100.
  • Publish a score whose weighted average is more than 2 points away from those components.
  • Treat a cited rental listing as proof the premises are still for lease, or a Census 2021 count as a live population.
  • Convert an estimate into a measured fact or guarantee a trading outcome.

Confidence and fallbacks

Uncertainty changes the output.

Missing data is not silently treated as a good result. Depending on the gap, the engine uses a labelled fallback, reduces a weight, widens a range, lowers confidence or withholds the normal recommendation.

A

Data completeness

Data completeness measures how much usable external and commercial evidence resolved, then subtracts for rejected values, operational corrections and fallback-heavy inputs.

B

Model confidence

Model confidence is a High / Medium / Low-style classification of how much the commercial model depends on observed feeds versus estimates and corrections.

C

Location Score

A score describes the modelled location read. Confidence describes how much reliable evidence supports that read. They answer different questions and are never interchangeable.

If a source fails

When a source fails, the engine may use a labelled benchmark or proxy, widen the displayed range, lower confidence, reduce the demand weight, downgrade the recommendation or return INSUFFICIENT DATA.

Observed sourceLabelled fallbackLower confidenceConservative output

Limits and due diligence

What you should still verify yourself.

Locatalyze is an early decision-support layer. A responsible lease decision still requires physical observation, current quotes, professional advice and inspection of the property and lease.

Known limitations

What the report cannot know reliably.

  • Public and third-party data can be incomplete, delayed or unavailable.
  • Competitor listings can be missing, duplicated, miscategorised or out of date.
  • A cited rental listing shows what that page said when it was opened. It is not proof the premises are still for lease.
  • Census population figures used in reports are from the 2021 Census, not a live count.
  • Street activity is often proxy-based; it must not be read as a measured pedestrian count.
  • Commercial rent depends on incentives, outgoings, lease terms, fit-out and property condition.
  • Demographic context may describe a wider area than the site’s true customer catchment.
  • Financial results are highly sensitive to the rent, ticket, volume, staffing and setup assumptions.
  • The model cannot reliably predict future competitors, roadworks, developments or demand shocks.
  • A report does not inspect the property, lease, licences, planning position or physical condition.

Before signing a lease

Turn the report into a verification plan.

  1. 01Count foot traffic and stopping behaviour at the actual frontage across weekday, weekend and relevant dayparts.
  2. 02Verify base rent, incentives, outgoings, review clauses, permitted use and make-good obligations.
  3. 03Walk the catchment and confirm which competitors are open, relevant and trading strongly.
  4. 04Check planning, zoning, food, liquor, signage, accessibility and other licence requirements.
  5. 05Replace category labour, supplier, fit-out and utility assumptions with current quotes.
  6. 06Stress-test revenue using your capacity, conversion rate, ticket and realistic ramp-up.
  7. 07Ask an accountant to review the commercial model and working-capital allowance.
  8. 08Ask a commercial leasing lawyer to review the lease before signing.
  9. 09Inspect the property and check the local development, vacancy and construction pipeline.

Useful next steps: opening checklist, break-even foot-traffic tool and Expert Business Review.

Sources and reference

Named providers, update policy and plain language.

“Current” means the provider was queried for the report, not that every record is complete or real-time. Source responses may be cached, and model versions change only when the decision logic is deliberately updated.

Provider and categoryReference and refreshCoverage and limitation

LocationIQ opens in a new tab

Address autocomplete and geocoding

ReferenceQueried when the address is confirmed

RefreshPer confirmation, with application caching

CoverageAddress or area, depending on match quality

LimitationNot survey-grade; complex premises can resolve imperfectly

Google Places opens in a new tab

Competitor and place listings

ReferenceCurrent provider response

RefreshQueried for a report; responses may be cached

CoveragePin-centred; competition scored within 500 m

LimitationA listing service is not a complete register of active businesses

Geoapify / OpenStreetMap opens in a new tab

Places, anchors, transport and fallback coverage

ReferenceCurrent provider/community dataset

RefreshQueried for a report; responses may be cached

CoveragePin-centred radii

LimitationCommunity-edited records may be incomplete or stale

Foursquare opens in a new tab

Fitness-category place discovery

ReferenceCurrent provider response

RefreshUsed for applicable categories

CoveragePin-centred radius

LimitationCategory coverage varies by market

Reference2021 Census reference data

RefreshUpdated when the underlying public-data integration is revised

CoverageSA1, postcode or suburb context where resolved

LimitationHistorical area data is not an exact-site customer survey

Commercial-listing and category signals

Rent comparison context

ReferenceSignals available to the report pipeline

RefreshPer report where available

CoverageUsually suburb or corridor level

LimitationNot a complete market database or a broker valuation

Locatalyze compute engine

Score, financial model, confidence and recommendation

ReferenceEngine 7.4.0

RefreshVersioned when decision logic changes

CoverageDeterministic calculation from the report inputs

LimitationModel quality remains constrained by input quality

Glossary

Terms used across the product.

Location Score

The weighted average of the components printed on that report. The weights are chosen for that business and add to 100.

Recommendation

The action label PROCEED, VERIFY, AVOID or INSUFFICIENT DATA. It can be more conservative than the score band.

Data completeness

The share and quality of usable evidence after fallback and correction penalties.

Model confidence

A qualitative read of how strongly the financial and location model is supported.

Rent burden

Monthly rent divided by modelled monthly revenue.

Workable rent band

The engine’s 8–18% rent-to-revenue screening range; it is a model benchmark, not a universal rule.

Competition intensity

Pressure from relevant nearby competitors, considering count, strength and proximity.

Demand signal

A directional read from market, demographic, anchor and activity evidence.

Location quality

Access, transit, nearby anchors and local activity around the pin.

Break-even

The transaction volume required for contribution margin to cover fixed modelled costs.

Contribution margin

Average ticket less variable cost of goods for that transaction.

Modelled estimate

A calculated value produced from inputs and assumptions rather than observed accounts.

Observed external data

Information returned by a named third-party or public source.

Proxy

An indirect signal used when the direct measurement is unavailable.

Failure condition

A defined change or threshold that would make the current case no longer viable.

Common questions

The shortest version of the method.

Does AI calculate the Location Score?

The research agent chooses the components and weights for that business and sets the overall score to their weighted average. The weights must add to 100, and the weighted average must land within 2 points of the overall score, or the report is rejected. It is not a fixed five-way table.

Why can a 60+ score still receive VERIFY?

The score is a weighted composite, while the recommendation also checks commercial inputs, economics, competition coverage, confidence and contradictions. A positive-looking composite cannot override a missing or serious decision risk.

Is foot traffic measured at the shopfront?

Usually not. Most reports use nearby activity, anchors, transit, provider signals or other proxies. The report should label these as proxy-based and users should conduct physical counts before signing.

Are the financial outputs forecasts?

They are modelled scenarios built from user inputs, provider signals and category benchmarks. They are decision-support estimates, not audited forecasts, guaranteed revenue or professional financial advice.

Apply the method to an address

Start with the evidence. Decide what to verify.

Your first location check includes the score, PROCEED / VERIFY / AVOID recommendation, data confidence and competitor map. No card is required to begin.

Start a free location analysisView the full sample report