Retail Site Selection Software: 12 Platforms Compared, and How to Test Them on Your Stores

You have a broker deck on one screen, an analyst spreadsheet on the other, and an expiring letter of intent. The question that matters right now is not who licenses the most map layers, but whether a platform can evaluate a real corner, explain its math, and hold up when tested against stores you already run.

This guide puts twelve retail site selection platforms side by side, looking at what each is built to do and where its answers originate. It also lays out a practical vetting method: let the model calibrate on your stores, hold a cohort back, and see whether it ranks them properly before you sign anything. That will tell you far more than a polished slide deck or an opaque score.

The category is having a noisy year. Vendors have rushed to paste chat interfaces across their products, and real estate leaders are rightly skeptical. "If I'm a 500-unit coffee shop, 500 observations is not enough data for AI to do anything intelligently," JLL's Paul Sill told Modern Retail. At ICSC in Las Vegas, a CBRE broker described a client's AI-curated property list as "absolutely terrible" (Commercial Observer). Both critiques are fair warnings against treating software like an oracle. The goal is to buy a system your team can inspect and question, keeping the final recommendation firmly in your hands.

I'm Clyde Christian Anderson, CEO of GrowthFactor. I started evaluating sites at 15 in my family's business, Books-A-Million, which became our first customer. We are one of the twelve platforms covered here, and I will be straightforward about where another option is the better fit.

The short answer: which platform fits which team

• A real estate team at a 20 to 500 location brand that opens stores every year and has to defend every pick: GrowthFactor. Type an address and the full site report loads in about 10 seconds, matched to the stores you already run and checked for cannibalization. Evaluators your team writes rate each site and show their reasoning, and the Agent works through a whole batch of broker sites at once.
• Restaurant, QSR and franchise expansion: Shortlist SiteZeus, Kalibrate and GrowthFactor, and test all three on your own restaurants.
• You mainly need visit trends and competitor traffic: Placer.ai, often running alongside a site selection platform rather than instead of one.
• You have GIS analysts and want to build your own models: Esri ArcGIS Business Analyst (what it costs and the alternatives), or CARTO if your data already lives in Snowflake or BigQuery.
• You want a forecast model co-built with you: SiteWise refreshes it once a year; GrowthFactor Labs co-builds too, refreshes quarterly, and lets your team rewrite evaluators between refreshes.
• You want analysis and the lease negotiation from one firm: Locate, which operates as a brokerage.
• Your leases and facilities already run in a real estate suite: Accruent's Lx Markets & Sites or Tango.

How the top retail site selection platforms compare in 2026

A few names that frequently surface in searches are underlying data feeds rather than decision platforms. StreetLight sells vehicle and pedestrian counts by road segment, Spatial.ai sells consumer segments from social and spending data ($425 to $1,500 a month, published), and Unacast sells mobility data by API. They feed an analysis, but they do not structure the decision. For foot traffic sources specifically, see our foot traffic provider comparison.

Plenty of teams run more than one tool. Many GrowthFactor customers keep a Placer.ai or Esri contract active and use us to evaluate every candidate site against their own stores before walking into committee. Before shopping, count how many tools one site decision takes today and decide which handoffs are worth collapsing. If you are reviewing an existing stack, we have written detailed breakdowns of alternatives to Placer.ai, SiteZeus, Kalibrate, Buxton and SiteWise.

What site selection software actually costs

Only two vendors in this group publish platform pricing openly. Esri sells Business Analyst starting at $4,200 a year in its online store, and Spatial.ai lists monthly self-serve packages. Every other provider, including GrowthFactor, scopes annual contracts through an enterprise sales conversation. Public contracts put Placer.ai at about $12,000 to $126,000 a year, while Buxton listed Audiense In-Person packages starting at $18,000 a year until it retired that pricing page in mid-2026.

Software license fees rarely represent the full operational cost. User access structures diverge: a 2023 municipal procurement record lists Placer.ai at $31,000 annually for unlimited seats, while Buxton stood at $40,000 for 12 seats with export functionality billed as a separate add-on (City of Bloomington). Data sets are frequently stacked; United Franchise Group's development leadership has discussed running SiteZeus, CoStar, Kalibrate, and Placer.ai in parallel. GIS-centric platforms also demand trained specialists to operate them.

When evaluated against physical store capital, software expenses are minor relative to the cost of opening an underperforming box. Chipotle invested roughly $1.5 million to build each new restaurant in 2025 (10-K). Retail fit-out costs averaged $157 per square foot this year, meaning a standard 3,000 square foot in-line unit requires about $471,000 before doors open (Cushman & Wakefield). Passing on problematic locations carries real financial impact as well: Portillo's recorded $5.1 million in dead site costs during fiscal 2025 while recalibrating its development pipeline (earnings release). The real return comes from opening fewer missteps and speeding up the process of defending strong sites.

How to test site selection software before you sign

Every vendor claims high accuracy. The relevant follow-up is asking how that accuracy was measured. A review of retail forecasting research by Fildes, Ma and Kolassa found little published evidence on new-store models "beyond untested claims as to the model based forecasts being highly accurate" (Lancaster working paper). You can cut through sales claims by running a controlled test in the exact sequence a sound model gets built.

First, let the model learn your stores

A vendor that scores your fleet cold is serving up an industry average. That shows how a generic brand performs near an interchange, not how your specific box behaves. Even traditional gravity models require local tuning: Esri's Huff calibration tool explicitly asks for customer data from existing and competing locations (Esri docs). Hand the vendor your sales history and let it identify what actually drives your locations, which variables move the needle, and how far your customers travel.

This calibration forms the foundation of Discovery, a 30-day analysis of your sales history that maps your customer profile, isolates the variables that predict volume, and defines your true trade area. A mutual NDA protects the work upfront. Your data is never pooled with another retailer's or used to train another customer's model, and we delete it permanently upon request.

Then, backtest on stores the model never saw

Once the model studies your fleet, withhold a group of units to see how it performs on sites it has never encountered. A 1989 regression study explained about 80% of store turnover yet never reported performance on subsequent openings. A later grocery model explained existing units well, then failed on new stores. Fildes and his co-authors point to overfitting on small datasets as one likely culprit. A hold-out backtest is the standard remedy.

Here is how to structure a clean backtest:

• Hold back 10 to 15 stores. Include top performers, struggling units, average performers, recent openings, and at least one location that defied internal expectations. Training on stores opened before 2024 and testing against units opened since provides a clean split.
• Withhold their sales. Provide street addresses and your site criteria, nothing more.
• Compare rank order first. If the model places the hold-out stores in the right relative sequence, the underlying logic is working even if absolute dollar figures need calibration. If it places your worst location in the top quartile, cross that platform off your list.
• Check the dollar forecasts and define the target year. A restaurant can take around 36 months to hit mature sales (Chipotle notes this pattern in its 10-K), while some specialty retail concepts experience an opening honeymoon surge above their long-term run rate. Require the vendor to state whether the forecast targets year one or year three.
• Watch for data leakage. If the model pulls mobile foot traffic generated by the hold-out store after its opening date, it is reading answers from the back of the book.
• Ask about the sites you passed on. You only have actuals for approved leases, and candidate locations with the highest raw estimates are statistically vulnerable to regression to the mean. A thoughtful vendor will raise this selection bias before you do.

One benchmark you'll hear is replicating existing revenues within 10% of actual, 90% of the time (Beckers et al., 2022), which only measures fit on open units rather than predictive power on new sites. Insist on seeing hold-out validation instead.

Then, read its reasoning on a live site

A backtest proves whether the ranking logic holds up mathematically. It does not show whether your team can make sense of the output on a rushed afternoon with a broker waiting on an answer. To judge day-to-day usability, run an active candidate parcel through each finalist and read the verdict.

Custom Evaluators became generally available in October 2026. In a GrowthFactor demo workspace, a team set up criteria the same way they would instruct an associate: prioritize weekday morning foot traffic, favor parcels within a five-minute walk of offices and transit, penalize proximity to independent coffee shops on the same block, cap the rating if the trade area overlaps an existing store by more than 25%, and flag whichever input carried the least confidence.

The evaluator gave 222 Third St in Cambridge a 4 out of 5. The report cited the closest analog store along with its similarity score, identified a Tatte Bakery 0.17 miles away as the primary competitive threat, confirmed trade-area cannibalization stayed below 2%, and noted that the provided foot traffic sample matched a dog park nine miles away. It highlighted that visit count as the least reliable input and advised gathering on-site morning counts before finalizing a lease. You can take that analysis straight to your real estate committee because the operational drivers, trade-offs, and data limitations are spelled out in clear language.

Demand that level of transparency on an active address from every vendor on your shortlist. A composite score without clear reasoning is just an expensive black box.

How many stores you need for each kind of model

Software directories rarely address data maturity, yet portfolio size dictates which analytical approach will actually work.

• A handful of stores: A predictive revenue model lacks sufficient data to train on. Rely instead on explicit site criteria, analog comparisons against peer brands, and physical market tours. Lil Sweet Treat expanded from 2 to 8 locations in a year with two founders and no analysts, reducing site review times from three weeks to two days.
• Dozens of mature stores: Analog store matching and multi-variable regression become practical options. Leave-one-out validation, where the system trains across all locations except one in repeated cycles, serves as a solid proxy for a larger hold-out cohort.
• 40 or more mature stores with revenue data: Calibrating a custom revenue model on your historical sales becomes viable. Our Labs team works with your team to build this model, and once calibrated, it generates new site forecasts in under 10 seconds.

Paul Sill's point about sample size holds: a model is only as sound as the fleet data backing it up. That is a compelling reason to pick software that matches your portfolio scale today, rather than holding off on structured analysis.

Forty stores in? Find out why your best ones win.

Thirty days on your own sales. Normally $5,000, free for qualifying brands with 40+ stores.

• Who your customer is
• What moves your sales
• Your true trade area

Start Discovery

Mutual NDA. Data never pooled.

Can developers use site selection software?

Partly. Retailers hold their brand concept constant and search for viable real estate. Landowners and developers hold a parcel constant and search for matching tenants. Because most site selection platforms are designed around brand expansion, catchment data translates well to developer needs while sales forecasting does not.

This distinction comes down to model training. Revenue forecasting models calibrate against a brand's historical performance across existing stores. When evaluating a vacant parcel without an operating tenant attached, the model has no baseline to learn from.

The location data remains valuable: demographic profiles, drive-time contours, pedestrian volumes, and local retail co-tenancy. A developer can use these metrics to identify tenant categories best supported by local demand, packaging the findings into materials prospective tenants can drop directly into their committee decks. Backing an available pad with local trade-area data creates a more productive conversation than leading with rent per square foot.

What site selection platforms do not handle is the technical development workflow. Zoning overlays indicate broad allowable uses, but managing entitlements, site density calculations, parcel assembly, and sourcing off-market land represent entirely different disciplines. For those processes, consult our guide to AI in real estate development. On greenfield sites lacking adjacent commercial density, these platforms gauge existing demand rather than the new demand a project is meant to create, leaving absorption modeling to experienced development teams.

Questions to ask in every demo

A prepared vendor should have direct answers for each of these items:

• Is my sales data pooled with other customers', or used to train their models? (GrowthFactor never pools your data or uses it to train another customer's model.)
• Who builds the model, and can I see its inputs on any single site?
• Are you forecasting year-one sales or mature sales?
• How does your foot traffic handle a store in a 50-story building, a thin rural sample, or a big neighbor in the same parking lot? A Ripco broker highlighted this vertical high-rise challenge directly (Commercial Observer).
• What do seats, exports and data licenses cost beyond the subscription?
• What happens to my data and my model if we cancel?
• Can someone in my seat drive this demo, not your solutions engineer?

Before sitting through software presentations, compile your internal baseline using our guide to store site selection criteria.

AI site selection tools: what the AI actually does

Most platforms now highlight AI capabilities, but the label generally masks three distinct functions. Some systems train predictive models on sales figures. Others place a chat interface across spatial datasets so users can search using natural language. Still others allow teams to set up evaluators based on specific criteria and review the logic behind every score.

The major development across 2026 has been conversational querying. Kalibrate introduced natural-language geospatial search via Microsoft Azure AI Foundry in April 2025. SiteZeus restructured Locate around an interactive assistant during its Atlas release in May. Placer.ai launched an MCP server enabling enterprise clients to query visit metrics through Claude or ChatGPT. GrowthFactor combines Custom Evaluators with an Agent that pins and rates batches of broker submissions, investigates the live web with a linked source for every claim, and ranks every block in a market against rules your team writes, with every tool call visible.

Teams are simultaneously experimenting with general-purpose LLMs on internal records, seeing both promise and blind spots. A franchise development head noted to Franchise Times that a Claude analysis credited roughly 22% of a clinic's ramp-up velocity to site selection. In that same article, an Avison Young executive cautioned, "You have to fact check everything in AI." The approach we're betting on pairs automated analytical legwork with completely visible inputs, ensuring your team can audit the reasoning before making an investment call.

One straightforward test clarifies any presentation: ask the system why a location received its score, and observe whether it details the underlying inputs or merely repeats the grade.

What teams are doing with it

• Books-A-Million expanded from 6 openings in 2024 to 19 in 2025 with the same internal headcount, reviews over 3,000 potential sites annually (up from roughly 5 to 10 a week by hand), and achieved a 14.1% increase in sales per square foot across new locations.
• Cavender's grew from 9 store openings in 2024 to 27 in 2025, with every new location performing at or above underwriting projections after reviewing 2,000+ candidates. Identifying and avoiding three flawed sites protected roughly $2 million in capital.
• TNT Fireworks stood up 153 retail locations over six months (150 seasonal operations and 3 permanent units) without missing a deadline. Its committee reviewed 10 times more locations per cycle, every opening hit or exceeded revenue targets, and avoiding late-stage site reconfigurations saved an estimated $500,000.
• Lil Sweet Treat evaluates over 120 properties monthly with two founders and no dedicated analysts, maintaining sales at or above underwriting across all operating stores.
• Ivy Kids, an early education franchise, uses the platform to supply pre-vetted locations to franchisees. Jessica C., head of real estate, notes: "I find it very reliable. When I turn a site over to a franchisee and say I feel really good about this one, they usually drive by and go, 'Oh my gosh, it's fantastic.'"

A multi-location veterinary group tested its own assumptions the same way we recommend testing any vendor platform:

"We thought we understood what made a good site. GrowthFactor showed us our primary KPI was hiding the real story, and the variables we'd been ignoring were the ones that actually predicted revenue. Now every forecast is built on what drives our business, not assumptions we never tested."

VP of Real Estate, multi-location veterinary group

Explore additional case studies across specialty retail, dining, and seasonal concepts in our customer stories.

Frequently asked questions about retail site selection software

Put the shortlist to work

Select software designed for your actual organizational capacity and store count. Require every platform under consideration to calibrate against your historical data, rank stores from your fleet it has never seen, and walk through the analysis of an active parcel you are reviewing this week. The right software is the one that equips your team to defend every location in committee with clear math across your entire network, rather than relying on an opaque score nobody can explain.

To run this evaluation against your own portfolio, book a demo and bring a live street address. In 30 minutes we will review the full site report, identify the closest comparable stores in your network, calculate local cannibalization, and run an evaluator configured to your criteria.