An AI agent for site selection runs the whole site-evaluation loop on its own — geocoding an address, pulling its trade area, scoring it against your portfolio, and flagging cannibalization — instead of answering one prompt at a time. The trait that makes it an agent is tool use: it calls live data and models, then chains the results into a finished, sourced read on the site.
Retail teams now see far more candidate sites than they can work through by hand, and analyst hours have not scaled with them. Retail real estate teams evaluate roughly 2.5x more candidate sites per opening than they did a decade ago, while decisions still run on the same scarce analyst time. That gap is what site-selection agents are built to close. This guide covers what an agent does in the site-selection workflow, the evaluation loop it runs step by step, how to point your own AI assistant at it through MCP, and what to verify before a site score goes in front of a committee.
What makes it an agent, not a chatbot
Ask a general assistant about an address and you get prose: plausible, undated, impossible to check. Ask an agent connected to real estate data and it geocodes the address, pulls the trade area, runs demographics and foot traffic, scores the site, and lines it up against comparable stores in your own portfolio. One returns an answer. The other returns work product.
The difference is multi-step execution. An agent decomposes a goal into steps, runs each step with a tool, and carries context between them. AI agents across commercial real estate do this for deal triage and market planning too; this piece stays on the site-evaluation job, where most retail teams feel the bottleneck first. For the broader category — including residential uses like lead routing and listing copy — the hub guide to AI real estate agents covers the wider map.
The evaluation loop an agent runs
When an agent evaluates a site, it runs the same sequence a careful analyst would — just in seconds instead of an afternoon. The steps matter because each one is a place the work can be wrong, and each one is a place a good agent shows you what it did.
1. Geocode the address and draw the trade area
The loop starts by turning an address into a location and a trade area — the geography the store will actually draw from, shaped by drive times and barriers, not an arbitrary ring. Everything downstream depends on getting this boundary right, which is why a real agent draws it from data rather than dropping a fixed radius on a map.
2. Pull demographics, foot traffic, and competition
With the trade area set, the agent gathers the inputs: who lives and moves through the area, how much foot traffic the location sees, and which competitors and complements sit nearby. Done by hand, this is the part that eats the afternoon — five tabs, three logins, a spreadsheet. An agent calls each data source and returns a full site report in about ten seconds.
3. Score the site against your own stores
A score is only as honest as what it is calibrated against. This is the step that separates a useful agent from a confident one. A score against generic benchmarks tells you a site is good for somebody. A score calibrated on your own portfolio's analogs — the five stores whose trade areas most resemble this candidate, and how they perform — tells you whether it is good for you. When an AI handles the first pass of site selection, the question to keep asking is whose data taught it what "good" looks like.
4. Check cannibalization before you fall for the site
A strong site that steals sales from the store two miles away is not a win. The agent overlays the candidate's trade area against your existing locations and estimates the overlap, so the cannibalization question gets answered before the lease is signed instead of after the second quarter of flat comps. This is the check that turns a promising address into a defensible one.
5. Assemble a read you can hand to committee
The last step is the one chatbots skip: packaging the work so a person can stand behind it. A good agent returns the score with its inputs attached — each scoring lens, its grade, and a written justification — alongside the map, the analogs, and the cannibalization read. That is the difference between a number and a repeatable site-selection process the team can defend.
Bring your own agent: MCP for site selection
Until recently, "AI agent" meant whatever agent your software vendor built. The Model Context Protocol changed that. MCP — launched by Anthropic in November 2024 and since adopted across the major AI assistants — is an open standard that lets any assistant call external tools and data, and the ecosystem passed 10,000 active servers within its first year.
For site selection, the implication is concrete: the assistant your team already uses can run the evaluation loop, provided someone publishes the tools. GrowthFactor shipped the first MCP integration in commercial real estate in April 2026 — scoring, demographics, foot traffic, cannibalization, and analogs, callable from Claude, ChatGPT, or any MCP-compatible client.
The working session that shows it off: a broker email lands with five addresses. Inside one conversation, the agent geocodes all five, scores each against your portfolio analogs, and saves the top three as deals — three sites scored from a single prompt in about thirty seconds. The full walkthrough of the MCP workflow shows each step. Two honest caveats: MCP access needs authenticated customer credentials, so it is a customer capability rather than a demo you can run from a marketing page, and a general assistant without those tools stays a poor substitute — no foot traffic, no trade-area math, no portfolio memory, and answers that shift between runs.
Before there is a site: market planning
The loop above evaluates a known address. Earlier in the work sits a different question — where the next five stores should go at all. An agent handles that sweep too: comparing sub-markets on demographics, foot traffic, and competitive density, then ranking where the open demand actually is and when a trade area is already saturated. The output is a ranked shortlist with reasons attached, which is a more useful artifact than a heat map someone still has to interpret.
What to verify before you trust a site score
Every agent demo looks impressive. The vetting question is what happens after the demo, when a VP of Real Estate has to stand behind the output in a room where someone asks, "Where did this number come from?" Four checks, in order.
Can you open the score? Click into any recommendation and see the variables that moved it — demographics fit, foot traffic, competition, visibility. If the inputs are hidden, the output is a black box with a chat interface, which is the exact thing buyers already distrust in older scoring tools. Mike Cavender, Co-Owner and Head of Real Estate at Cavender's, set the standard: "Other services hide behind black-box models that are hard to trust. The beauty of GrowthFactor is they make site selection incredibly simple, and give us clear unbiased recommendations."
Can you change the weights? A defensible agent lets your team adjust the scoring lenses to match how your concept actually works, and shows how the score moves when you do. An agent you cannot argue with is an agent you cannot learn from.
Is it calibrated on your data? Ask what the agent compares candidates against. If the answer is not "your own stores," the recommendation carries someone else's assumptions.
Does it know its job is the first pass, not the final call? Teams that get value run the agent as a second opinion: it compresses the analysis, the team owns the recommendation. Customers running that loop report meaningful portfolio outcomes — roughly 80% fewer underperforming locations once the full workflow is in place — precisely because the agent surfaces the evidence and people make the call.
Where this goes
Within a couple of years, "agent" will stop being a feature word in site selection and start being assumed, the way "cloud" did. The question that will still matter is not whether your tools include an agent, but whether that agent reaches data worth reasoning over and produces a read you can defend.
Start with one bounded job. Hand an agent your next broker list and see whether the first pass holds up against your team's read. If it does, you have found the second opinion. If it cannot show you why, you have found the next black box — and you already know how that story ends. See how the GrowthFactor platform runs the loop end to end.