How to Build a Submarket Intelligence Report with AI

To build a submarket intelligence report with AI, you pair two things most reads keep apart: clean backward-looking fundamentals and the forward signals that lead them. You pull the rent comps, absorption, supply pipeline, and cap-rate context with a source and date on every number, normalize them into one shape so the submarket is actually comparable, then layer the leading indicators (permit clustering, businesses opening versus closing, how brokers describe the district now versus six months ago) clearly labeled as signals, not facts. AI does the sourcing, the reconciliation, the signal-scanning, and the first draft of the read. A person writes the thesis and owns the bid. The result is a report your investment committee can act on, one that says where the submarket is going before the comps catch up.

This is written for the acquisitions team about to bid in a submarket it does not live in every day. The problem with the standard read is not that it is wrong. It is that it is late. A signed lease takes one to two quarters to show up as a comp, so a report built only on comps is a photograph of where the market was, not where it is going. In a flat market that lag is harmless. In a market that is moving, and the biggest office-leasing force right now, AI and tech tenants, are signing in places the historical model has almost no data on, that lag is the whole game. Below is how to build the read that has both halves, and where a human stays firmly in charge.

Step 1: Pull the Fundamentals, Sourced and Dated

Start with the backward-looking half, because it has to be clean before anything else matters. Pull the rent comps, the absorption trend, the supply pipeline, and the cap-rate context for the submarket, and attach a source and a date to every single number. Rents from a named comp set, not "the market is around fifty a foot." Absorption from the actual quarterly figure, not a vibe. This is the part everyone already does, and AI mostly speeds up the gathering rather than changing it.

The discipline that matters here is provenance. A submarket number you cannot trace to a source and a date is a number you cannot defend in front of your investment committee, so the workflow refuses to carry one. That sounds obvious, and it is exactly what gets skipped when an analyst is assembling a read under deadline. Note also that clean comps are commoditizing: new platforms now centralize public rental data into real-time, unit-level comparisons of rents, fees, and concessions, the manual pull that used to eat an afternoon. When the inputs become table stakes, the edge moves to the read you build on top of them, which is the rest of this report.

Step 2: Normalize the Comps Into One Shape

Comps arrive from five sources in five formats, and most submarket reads quietly die in the reconciliation. You cannot compare across a submarket while you are squinting at five different PDFs, so the next step is to normalize everything into a single schema: the same fields, the same units, the same date convention, for every comp. This is tedious, structured, repeatable work, which is precisely what AI is good at and precisely where a person should not spend an evening.

Getting to one shape is what turns a pile of documents into something you can actually reason about. Once the comps sit in a single structure, you can see the spread, spot the outlier, and compare the subject against the set instead of against your memory of the last deal. The AI does the parsing and the reconciliation at speed and without transcription errors. A human spot-checks the normalized set against the sources, because a clean-looking table built on a misread field is more dangerous than a messy one, and this is the moment to catch it.

Step 3: Layer the Forward Signals, and Label Them as Signals

Now the part that separates a real read from a data dump. On top of the clean fundamentals, layer the leading indicators: permit activity clustering on a corridor, business licenses opening versus closing, and the way tenants and brokers describe the district today compared with two quarters ago. None of these are comps. All of them lead comps. Recent reporting has shown AI can turn exactly these soft signals into testable ones, by scanning public meetings for recurring infrastructure themes, tracking whether local businesses are opening or closing, and flagging when brokers describe a district differently than they did six months ago.

The load-bearing discipline is in how you present them. These are leading indicators to test, not facts to underwrite, and the report has to say so on the page. A forward signal labeled as a signal makes your read earlier and honest at the same time. A forward signal quietly mixed in with the comps as if it were settled fact makes your read a liability. This is where AI helps and also where it can hurt: it will happily surface a pattern, so the workflow forces every signal to be tagged as directional and tied to what it would take to confirm it. That is the difference between reading the windshield and inventing a road.

Step 4: Write the Read, With the Thesis on Top

A submarket report an acquisitions team can use is not forty tabs. It is a read: where this submarket is, where it is going, and what would change your mind, in two paragraphs a principal will actually finish. So the report opens with the thesis, not buries it. Where is this submarket headed, what is the evidence from the fundamentals and the signals, and what specifically would flip the call.

AI drafts that read from the structured inputs you assembled, because the narrative is built on the numbers and signals already in front of it. But the thesis is a judgment call, and it stays with the team that will live with the bid. The AI turns the clean data and labeled signals into a coherent first draft in minutes, which is the slow part. A person pressure-tests the thesis, cuts the evidence that does not hold, and decides what the read actually recommends. The upstream discipline that makes this trustworthy, sourcing every number and screening hard, is the same one we lay out for inbound deals in the CRE deal-inbox playbook.

Step 5: Add the Honest Caveats

The last section is the one that keeps the report from getting you in trouble: the caveats, stated plainly. The soft signals are directional, not predictive, so you never price off them alone. Every number is either sourced or flagged to confirm. And the whole report is decision-support, not a forecast: it tells you where the weight of the evidence points and where you are still guessing.

This is not lawyerly hedging, it is what makes the read usable. A report that admits what it does not know is one an investment committee can trust, because it is not asking them to underwrite a guess dressed up as a fact. The forward signals give you the early call. The caveats keep the early call honest. Together they let you move before the comps catch up without pretending you can see the future, which is the entire point of building the report this way instead of just averaging last quarter's rents.

Why This Beats the Standard Submarket Read

The standard read is a backward mirror: clean comps, averaged rents, noted absorption, all true and all a quarter or two old. In a moving market that is not enough, because the tenants redrawing the submarket have already left a trail the comps will not show for months. AI and tech firms are taking a fast-growing share of US office leasing, and they are clustering in mixed-use pockets and converted industrial space rather than the towers the historical comps are built around. By the time that demand prints as a comp, the good space is gone and you paid last quarter's price for it.

There is a broader lesson here worth stating. MIT's 2025 State of AI in Business found that 95% of organizations are getting zero return on generative AI despite tens of billions in spending, and the ones seeing returns are not the ones using AI, they are the ones who redesigned a workflow around it. A submarket read is a perfect candidate for exactly that, if you rebuild the process (fundamentals, normalization, labeled signals, thesis, caveats) instead of bolting a chatbot onto the old one. If you are deciding where AI actually fits across acquisitions, our guide to the best CRE market data and comps platforms covers the input layer, and AI site selection for development covers the forward-looking read for ground-up. The submarket you can see in the comps is the one everyone else can see too. The edge is reading the one that is forming.

Frequently Asked Questions

How do you build a submarket intelligence report with AI?

You pair clean fundamentals with forward signals. First pull the rent comps, absorption, supply pipeline, and cap-rate context with a source and date on every number, then normalize them into one schema so the submarket is actually comparable. Then layer leading indicators, permit clustering, business openings versus closings, and shifts in how brokers describe the district, clearly labeled as signals to test rather than facts to underwrite. AI does the sourcing, the reconciliation, the signal-scanning, and the first draft of the read; a person writes the thesis, adds the honest caveats, and owns the bid. The output is a two-paragraph read your IC can act on, not a forty-tab data dump.

Why are traditional submarket comps not enough anymore?

Because comps lag. A signed lease takes one to two quarters to show up as a comp, so a read built only on comps describes where the submarket was, not where it is going. In a flat market that is fine, but in a moving one it is the whole problem, especially now that AI and tech tenants, who are taking a fast-growing share of US office leasing, are signing in mixed-use and converted-industrial pockets the historical model barely tracks. By the time that demand prints as a comp, the space is gone and you paid last quarter's price. Adding labeled forward signals to the clean fundamentals is how you see the shift before it is priced in.

What are forward signals, and can you underwrite off them?

Forward signals are leading indicators that move before comps: permit activity clustering on a corridor, local businesses opening versus closing, and brokers describing a district differently than they did six months ago. You do not underwrite off them alone, and the report must label them as directional, not predictive. Their job is to tell you where to look and what to test, not to set a price. Used that way they make your read earlier and still honest. Mixed in with the comps as if they were settled facts, they make your read a liability. The discipline of tagging every signal as a signal, tied to what would confirm it, is what keeps the report trustworthy.

Does the AI decide which submarket to buy in?

No. The AI assembles the fundamentals, normalizes the comps, scans and surfaces the forward signals, and drafts the read. The thesis, where the submarket is headed and whether to bid, stays with the acquisitions team that will live with the decision. The value is that your team starts from a clean, sourced, signal-layered draft instead of a blank page and five PDFs, so their judgment goes to the call that matters rather than to reconciliation. The report is decision-support, not a forecast: it tells you where the evidence points and where you are still guessing, and a human owns what to do about it.

How is this different from a market report generator or a data platform?

A data platform gives you cleaner inputs, and clean comps are becoming table stakes that everyone can buy. This report is the read you build on top of those inputs: the normalization into one comparable shape, the forward signals layered and labeled, the thesis, and the caveats. It is a redesigned workflow, not a dashboard. That distinction matters because, as recent research on AI adoption shows, the firms getting a return are the ones who rebuilt a process around AI rather than bolting a tool onto the old one. The platform commoditizes the fundamentals; the report is where the edge actually lives.