A comp is a comparable — a property similar enough to the one you are pricing that what it sold or leased for tells you something real about what yours is worth. “Comps” is just the trade’s shorthand for comparables, and they are the backbone of how commercial real estate gets valued: you find recent transactions on buildings like yours, in a market like yours, and you read your property’s likely price or rent off theirs. The concept is simple. The work is not. For a small firm, the hours vanish into finding the right comps, cleaning the data, and adjusting for the fact that no two buildings are ever the same. This piece explains what comps are, the two kinds that matter, what makes one truly comparable, and where AI genuinely takes work off your plate versus where it will hand you a confident, wrong number.
What “comps” means
A comp is a recently transacted property you use as a yardstick for the one in front of you. If you are trying to price a 20,000-square-foot industrial building, you look for other industrial buildings of a similar size, age, and location that recently sold or leased, and you use their numbers as evidence for yours. Comparables is the full word; comps is what everyone says.
The logic is the same one behind pricing anything in a market. A house appraisal leans on the three homes down the street that sold last quarter; a commercial deal leans on the warehouses, strip centers, or apartment buildings that changed hands nearby. You are not guessing at value from theory — you are reading it off the market’s own recent decisions.
What comps produce is not a single number but a set of market signals: price per square foot, price per unit, cap rate, market rent, typical concessions. Those signals become the inputs to everything downstream — pricing an offer, projecting income, and sanity-checking whatever a seller is asking. Get the comps wrong and every number built on them is wrong too.
Sales comps vs. lease comps
The single most useful distinction is that “comps” covers two different jobs, and a small firm needs both for different reasons.
Sales comps are recent sales of similar properties. They tell you what buildings like yours are trading for — price per square foot, price per unit, and the cap rate the sale implied. Sales comps are how you pressure-test an asking price and how you set your own. If three comparable retail centers sold at cap rates between 6.5% and 7%, and a broker is pitching you one at a 5.5% cap, the comps just told you the price is rich for the market.
Lease comps, sometimes called rent comps, are recent leases on similar space. They tell you what the market pays in rent, on what terms, with what concessions — free months, tenant-improvement allowances, escalations. Lease comps drive the income side of a deal. Before you can trust a property’s projected rent, you need to know whether that rent is what the market bears or a number the seller wrote down and hoped for.
The two feed each other. Lease comps establish what a building can earn; sales comps establish what that earning stream is worth. Buyer-side, you usually need both — the rent the space can command and the price similar income streams fetch. Lease comps are also harder to source than sales comps, because lease terms are private and the real economics hide in the concessions, not the headline rate. A quoted rent with three months free and a fat improvement allowance is a different deal entirely from the same rent with none.
What makes a comp comparable
A comp is only worth as much as its similarity to your property. Reach too far for comparables and you are no longer comparing — you are guessing with extra steps. Five tests separate a real comp from a misleading one.
Property type. Office comps do not price industrial; multifamily does not price retail. The comp has to be the same asset class, and ideally the same subtype — a Class B suburban office is not a comp for a downtown trophy tower.
Location. Same submarket, or as close as you can get. Value in real estate is local to a degree that surprises people from other industries; two identical buildings ten minutes apart can carry materially different rents and cap rates because one sits in a stronger submarket.
Recency. A sale or lease from the last six to twelve months. Markets move, and a comp from two years ago prices a different world — interest rates, rents, and cap rates all shift. Old comps are not evidence; they are history.
Physical similarity. Comparable size, age, condition, and class. A renovated building is not a clean comp for a tired one; a 100,000-square-foot warehouse does not price a 15,000-square-foot flex space. Clear height, loading docks, parking ratios, unit mix — the physical details that drive value have to line up.
Transaction quality. A closed, arm’s-length transaction — not a listing price, not a distressed fire sale, not a sale between related parties. A listing is what someone hoped to get; a comp is what the market paid. Confusing the two is one of the most common ways a comp set gets poisoned.
Where comps come from
Comps come from a handful of paid databases, public records, and the relationships in your own network — and for a small firm, the mix you can afford shapes how good your comps are.
The data platforms. CoStar is the largest and most widely used verified-comp database in the US market, with a research team that confirms transaction details; it is also the most expensive, which puts full access out of reach for many small shops. LoopNet, which CoStar owns, is strong for browsing active listings but thinner on historical comps. Crexi is a more accessible platform for listings and some transaction data. Reonomy focuses on property and owner intelligence, useful for off-market and ownership records. CompStak crowdsources lease comps specifically, filling the hardest-to-source gap. Verify what each platform covers in your markets before you buy — coverage shifts, and a national tool can be thin in a secondary metro.
Public records. County assessor and recorder data captures sale prices and transfer dates and costs nothing but time. It is the fallback when a database is out of budget, though it lacks the lease terms, physical detail, and verification that make a comp trustworthy without extra legwork.
Your network. For a small firm, brokers and appraisers in your market are often the best comp source you have. The person who closed the deal knows the real price, the real terms, and the concessions that never make it into a database — comp intelligence that money alone cannot buy.
Most small firms end up with a patchwork: one paid platform they can justify, public records for the rest, and a network that fills the gaps. Choosing which platform to pay for is its own decision, and the tradeoffs behind AI-enabled research tools are worth understanding before you commit — a subject taken up in the rundown of AI market-research tools for brokers.
The adjustment problem
Here is the part the tidy definitions skip: no two properties are ever identical, so a raw comp is never the answer. It is a starting point you then adjust.
Adjustment is the appraiser’s discipline of correcting a comp’s price for every material way it differs from your property. The rule is mechanical in direction: if the comp is inferior to your building in some respect, you adjust its price up, because your better building is worth more than what the weaker one fetched. If the comp is superior, you adjust down. A comp that sold with covered parking your building lacks gets adjusted downward; a comp in worse condition than yours gets adjusted upward. You do this across every meaningful difference — location, size, age, condition, amenities, date of sale — until the comp is expressed as what it would have sold for if it were your property.
The trouble is that adjustment is judgment, not arithmetic. How much is covered parking worth in this submarket? What is the right adjustment for a building fifteen years older? The appraisal profession codifies the categories, but the size of each adjustment comes from market knowledge, and reasonable experts disagree. This is why comps are evidence, not proof: a comp set is a range and an argument, not a single certain number, and the quality of your adjustments is the quality of your conclusion.
That judgment is also where comps connect to the rest of the deal math. The cap rate a comp implies only means something once you have adjusted for how the properties differ, and pressure-testing that rate is its own step — walked through in the guide to cap rates and sanity-checking them.
Where the hours go for a lean firm
For a 4-to-20-person firm with no analyst, the cost of comps is not the thinking — it is the grind of assembling them. The judgment about which comps fit and how to adjust takes a principal a focused stretch of attention. The hours disappear into everything before that: searching databases, pulling records, keying transaction details into a spreadsheet, chasing a broker to confirm a price, reconciling three sources that disagree on the same building’s square footage.
Industry estimates put traditional comp gathering at several hours of manual data work per property before anyone reaches the analysis. Multiply that across a pipeline and the arithmetic gets grim: a lean firm either analyzes fewer deals than it should, or it analyzes them on thin, half-assembled comp sets because building a full one costs more time than the deal flow allows. Principals end up pricing deals on three comps they could find quickly rather than the eight that would truly triangulate the value.
That imbalance — cheap judgment, expensive assembly — is the specific place a small firm bleeds capacity, and it is also the place technology has the most room to help. Understanding comps as two separable layers, the mechanical gathering and the human judgment, is what makes the next question answerable: which layer can you hand off, and which has to stay with you?
Where AI helps with comps — and where it must not
AI is genuinely useful with comps, but only on one side of that line, and the whole value depends on keeping the two straight.
On the mechanical side, AI-assisted tools are good at the gathering and matching that eats a lean firm’s hours. Purpose-built platforms surface candidate comps by scoring similarity across submarket, distance, size, and physical attributes; general assistants like ChatGPT, Claude, or Microsoft Copilot can read a stack of transaction PDFs and pull the prices, dates, and square footage into a clean table, or summarize a set of leases into a rent-comp grid. Vendors report real time savings here — HelloData, for instance, publishes claims of saving several hours a week on multifamily market surveys and matching property-manager comp selections around nine times in ten. Treat vendor numbers as marketing until you test them on your own markets, but the direction is right: the assembly layer is where the hours come back. A small shop that automates gathering can build eight-comp sets in the time it used to spend finding three, which is a direct lift in how many deals it can price well — the pattern that lets small firms out-operate larger ones, argued in the small-firm playbook for out-operating institutional competitors.
On the judgment side, the line is firm: AI does not decide which comps are truly comparable, and it does not own your adjustments. A model asked whether a comp fits will answer confidently with no feel for the submarket, and asked to adjust for a condition difference it will invent a plausible dollar figure it cannot defend. In a valuation, a confidently wrong comp or a fabricated adjustment prices a bad offer. The rule is the one that governs every safe use of these tools in deal work: the model gathers and reconciles, a human decides which comps count, how to adjust them, and what number to trust. And every figure a model extracts from a document gets verified against the source — models misread a table cell often enough that unverified extraction is its own risk.
AI comps also have honest failure modes worth naming up front. In thin markets with few recent transactions, or for genuinely unique assets with no real peers, there simply are not enough comparables for any tool to reason from, and a model will paper over the gap with weak matches rather than admit it. Knowing when your comp set is too thin to trust is a skill the tools do not have — a limit examined in where AI comps stop working.
One guardrail for a firm with no IT department: transaction records, rent rolls, and lease terms are confidential financial data. Before you put deal documents into any AI tool, get in writing where the data is stored, whether your inputs train shared models, and how you delete your history. Major providers state that business-tier and API data is not used for training by default, but the contract is your safeguard, not the marketing page.
Comps rarely stand alone — they are one input in a larger sequence. They feed underwriting, where the market rent and cap rate they establish become the assumptions behind a value, and they inform the screening pass that decides which deals earn a full look in the first place. For how comps, screening, underwriting, and the market view fit into one deal-analysis workflow, the deal-analysis playbook for lean teams sequences the whole stack.
FAQ
What are comps in commercial real estate?
Comps, short for comparables, are recently transacted properties similar enough to the one you are pricing that their sale or lease figures serve as evidence for its value. If you are pricing an industrial building, your comps are other industrial buildings of similar size, age, and location that recently sold or leased. They produce market signals — price per square foot, price per unit, cap rate, market rent — that become the inputs to valuing, pricing, and underwriting a deal. The best evidence of what something is worth is what similar things just traded for.
What is the difference between sales comps and lease comps?
Sales comps are recent sales of similar properties; lease comps are recent leases on similar space. Sales comps tell you what buildings like yours trade for — price per foot, price per unit, implied cap rate — so you can pressure-test an asking price or set your own. Lease comps tell you what the market pays in rent, on what terms, with what concessions, which drives the income side of a deal. Buyers usually need both: lease comps establish what a building can earn, and sales comps establish what that income stream is worth. Lease comps are harder to find because terms are private and the real economics hide in concessions.
What makes a good comp?
A good comp matches your property on five things: property type and subtype, submarket location, recency (ideally a transaction from the last six to twelve months), physical similarity in size, age, condition, and class, and transaction quality — a closed, arm’s-length deal rather than a listing or a distressed sale. The closer the match, the stronger the evidence. A listing price is not a comp; it is what someone hoped to get. A sale between related parties or a fire sale is not a comp either. Reach too far on any of these and the comp starts misleading you instead of informing you.
Where do commercial real estate comps come from?
From paid data platforms, public records, and your own network. CoStar is the largest verified US comp database but expensive; LoopNet and Crexi are more accessible for active listings; Reonomy focuses on owner and property intelligence; CompStak crowdsources lease comps specifically. County assessor and recorder records provide sale prices for free but lack lease terms and verification. For a small firm, brokers and appraisers in your market are often the best source, because they know the real price, the real terms, and the concessions that never make it into a database. Most small firms use a patchwork of one paid tool, public records, and relationships.
Why do you have to adjust comps?
Because no two properties are identical, so a raw comp is a starting point, not the answer. Adjustment corrects a comp’s price for every material difference from your property. The direction is mechanical: if the comp is inferior to yours in some respect, you adjust its price up; if it is superior, you adjust down, until the comp reflects what it would have sold for as your property. The size of each adjustment, though, is judgment drawn from market knowledge, not arithmetic, and experts disagree. That is why a comp set is a range and an argument rather than a single certain number.
Can AI find and analyze comps?
AI can do the gathering and matching well and should not be trusted with the judgment. Purpose-built tools score candidate comps for similarity across submarket, distance, size, and attributes, and general assistants can pull transaction details out of PDFs into a clean table — the work that consumes a lean firm’s hours. What AI cannot do is decide which comps are truly comparable, own the adjustments, or know when a market is too thin to trust. A model will produce confident, undefendable answers on all three. The safe pattern is AI for gathering and reconciling, a human for deciding which comps count and what number to trust, with every extracted figure verified against its source.
How long does it take to pull comps for a deal?
Manually, several hours of data work per property before any analysis begins — searching databases, pulling records, keying details into a spreadsheet, and confirming prices with brokers. That assembly time, not the judgment, is what limits how many deals a lean firm can analyze. AI-assisted gathering can compress it substantially, letting a small firm build fuller comp sets in the time it used to spend on partial ones. The judgment layer — deciding fit and adjustments — stays roughly constant, because it depends on market knowledge a tool does not have.
Is a listing price a comp?
No. A listing price is what a seller hopes to get; a comp is what the market paid in a closed, arm’s-length transaction. Treating an active listing as a comp is one of the fastest ways to poison a valuation, because listings systematically reflect aspiration rather than achieved value, and they may sit unsold precisely because the price is wrong. Distressed sales and transactions between related parties are similarly unreliable as comps. When you build a comp set, weight closed, market-rate transactions and treat listings only as a loose signal of where asking prices sit, not as evidence of value.
Is it safe to put deal documents into an AI tool?
Only under the right contract. Transaction records, rent rolls, and lease terms are confidential financial data, so before uploading them to any AI tool, get three answers in writing: where the data is stored, whether your inputs train shared models, and how you delete and export your history. Major providers state that business-tier and API data is not used for training by default, but for a firm with no IT department the contract is the safeguard, not the marketing copy.
Key takeaways
- A comp is a comparable — a recently transacted property similar enough to yours that its price or rent is evidence of your property’s value. Comps produce the market signals (price per foot, cap rate, market rent) that feed pricing and underwriting.
- Sales comps tell you what similar buildings trade for; lease comps tell you what the market pays in rent and on what terms. Buyers usually need both, and lease comps are the harder set to source.
- A comp is only as good as its similarity: same property type, submarket, recency, physical profile, and a closed arm’s-length transaction. A listing price is not a comp.
- No two properties match, so comps get adjusted — up for an inferior comp, down for a superior one. The direction is mechanical; the size is judgment, which is why a comp set is a range and an argument, not a single number.
- AI is genuinely useful for the mechanical layer — gathering, extracting, and matching comps — and dangerous on the judgment layer. Let it assemble; keep the decisions about fit, adjustment, and trust with a human, and verify every extracted figure against its source.
Want to know where your firm’s hours go in deal analysis, and which parts AI can safely take off your plate? A short assessment maps that against your deal flow, your documents, and where AI belongs without introducing risk. Book your free AI-readiness assessment →
Arthur Wandzel
