AI broker opinion of value: a first-pass BOV workflow
How a brokerage team has AI assemble the first pass of a BOV in its own format, and which calls on comps, cap rate, and price stay with the broker.
King & Company
In short
AI can assemble most of a broker opinion of value: the property facts, in-place income, the comp table, the valuation math in your own workbook, and a narrative draft in your team's voice. Which comps belong, what cap rate range is defensible, and what price to recommend remain the broker's decisions, and a well-built workflow stops and asks at each one. Before building, confirm that your comp data license permits the use and put your state's required not-an-appraisal language into the template.
AI can do the assembly work on a broker opinion of value, and the opinion itself should stay with the broker. Split the BOV into those two kinds of work, hand the assembly to AI, and have the workflow stop and ask whenever it reaches a judgment call.
That split matters because of how a BOV wins a listing. The owner is hiring the broker's read on which comps apply and how the property should be positioned, and no draft produced without that read is worth sending. Everything around it (pulling facts out of the owner's documents, building the income picture, formatting the comp table, writing the first version of the narrative) is assembly, and assembly is where an analyst's hours go.
What goes into a BOV, and where do the hours go?
Formats vary by team and property type, but most commercial BOVs carry the same parts: a property description, in-place income and expenses, a comp table, one or more valuation approaches that produce a range, a recommended price or pricing strategy, and a narrative that makes the case for the team.
How long a BOV takes depends on the property and on the state of the documents the owner sent, and we have not seen a published figure worth quoting. The useful exercise is to time your own. Have the analyst log one BOV from first document to final draft, by section. The log tells you how much of the time went to re-keying a rent roll, reconciling it to the operating statement, and formatting, and how much went to the valuation judgment itself.
Assembly versus opinion: where is the line AI should not cross?
| Part of the BOV | Who owns it | What the workflow does |
|---|---|---|
| Property facts | Workflow | Extracts from the owner's documents, cites the source page, flags gaps |
| In-place income | Workflow | Builds NOI from the rent roll and operating statement, flags anything that does not reconcile |
| Comp selection | Broker | Stops and asks for the comp set |
| Comp table and adjustments | Workflow, then broker | Formats the chosen comps, leaves adjustments blank or marks them as proposed |
| Cap rate range | Broker | Stops and asks, shows what the chosen comps imply |
| Valuation math | Workflow | Runs the team's existing approaches in the team's workbook |
| Recommended price | Broker | Stops and asks |
| Narrative | Workflow, then broker | Drafts from prior BOVs in the team's voice |
| Required statement | Template | Included by default, never generated fresh |
The rule we build to is that a number expressing an opinion is entered by a person, and the document shows a visible blank until it is.
How do you build in-place income from the owner's documents?
Owners send what they have: a rent roll export, a trailing twelve month operating statement, sometimes leases, sometimes a tax bill. The workflow reads each one and fills the income section of your template, tenant by tenant, with the page or cell each figure came from.
The most useful output is the list of exceptions it produces alongside the numbers. A good first pass tells the analyst that the rent roll total does not match the operating statement's rental income, that a suite appears on one document and not the other, that a lease expires inside the next twelve months, or that an expense line looks like a one-time item. The analyst then spends their time on those questions and on the call back to the owner. When the leases themselves are in the package, the same method used for lease abstraction applies.
Write down how your team defines its terms before you build. Whether NOI is stated before or after reserves, and how vacancy is treated, should be an instruction the workflow follows every time. If nothing is written down, a standard glossary such as Commercial Real Estate Terms and Definitions, a terminology reference begun in 2004 and reissued in a 2024 edition, is a reasonable place to start.
The comp table: the broker picks, the workflow formats and flags
Comp selection is part of the opinion, so the workflow asks the broker for the comp set and does not go looking for one.
Once the broker has chosen, the workflow lays the comps into the firm's table, computes the derived fields (price per square foot, price per unit, implied cap rate where income is known), and flags what a careful analyst would flag: a sale that is older than the rest, a comp well outside the subject's size range, a missing field. Adjustments stay blank, or appear as proposals clearly marked as proposals with the reasoning written beside them, for the broker to accept or overwrite.
Running your existing valuation approaches in your own workbook
The team already has a workbook, and the BOV should keep coming out of it. Rebuilding the math in a new tool gives the broker a number they cannot trace.
Anthropic's documentation for Claude for Excel says the add-in can populate an existing template, update values while keeping formula relationships intact, and explain how a number was derived with cell-level citations. The same page lists "final client deliverables without human review" under uses it is not recommended for, and warns that files from outside sources can contain hidden instructions, which is relevant when the inputs are spreadsheets an owner emailed you. Both points argue for the same design: inputs flow into your workbook, the cap rate cells wait for the broker, and a person reviews the result. The broader version of this approach is covered in underwriting faster without giving up your own model.
Drafting the narrative from your prior BOVs
A narrative written from a blank prompt reads like every other AI narrative. One written from ten of your own past BOVs, with your section order and the way your team describes a submarket, reads like your team.
This is a good use of a Claude skill, which Anthropic describes as a directory of instructions, scripts, and resources that Claude loads for a specific task. The skill holds the template, the prior examples, the definitions, and the stop points, so every analyst gets the same first pass. We cover which skills are worth building in Claude skills for commercial real estate. The draft should make no claim about the market that is not supported by the comp table or by something the broker supplied.
Check your data license before comps go into an AI tool
Comp data often comes from a subscription service, and the subscription comes with a license agreement. Before building anything, read what that agreement says about putting the data into outside software, AI tools included, and ask your account representative and your counsel if it is unclear. We are not characterizing any provider's terms here, because they differ and they change.
A workflow built only on your firm's own closed-deal records keeps licensed data out of it. Comps the broker copies in from a subscription service are still that provider's data, so the license question applies to them as well.
A BOV is not an appraisal: what disclosure language should you build in?
An appraisal is developed by a licensed or certified appraiser under USPAP, which The Appraisal Foundation describes as the national standards for real estate appraisal. A BOV is a broker's estimate, and state law governs when a broker may give one and what it has to say.
Two examples show how much the rules vary:
- Texas. Texas REALTORS explains that a broker may charge a fee for a broker price opinion, and TREC's rules (Rule 535.17) require a prescribed statement, reproduced verbatim in at least 12-point font, that the estimate is not the same as the opinion of value in an appraisal developed by a licensed appraiser under USPAP.
- North Carolina. G.S. 93A-83 lists what a broker price opinion must contain, including a statement that it is not an appraisal and may not be used in lieu of one. The rule has also moved over time: a March 2011 commission bulletin described a narrower rule tied to a present or prospective brokerage client.
Those two states are only examples, and other states have their own rules. Confirm the current rule in your state with your managing broker or counsel, then put the required language in the template as fixed text. An automated draft should carry it by default so that omitting it takes a deliberate act.
The review pass before it goes to the owner
The broker's review is a designed step with a short list:
- Every figure in the income section traces to a source document.
- Every exception the workflow raised has been resolved or noted.
- The comp set, adjustments, cap rate range, and recommended price were entered by the broker.
- The narrative says nothing the numbers do not support.
- The required statement is present and unaltered.
More on building that step is in how to design the human review step in an AI workflow.
What does a first-pass BOV workflow look like once it is built?
The analyst drops the owner's documents into a folder and starts the skill. It returns the team's workbook with the income section filled and sourced, an exceptions list, and a request for the comp set. The broker supplies comps, the table and derived fields fill in, and the workflow asks for the cap rate range and the price. The narrative draft arrives in the team's format with the state language already in place, and the broker edits it the way they would edit an analyst's draft.
A BOV first pass is one of the workflows we build with brokerage teams against their own template and prior work, and you can get in touch to see how it would fit yours.
Common questions
Can AI write a broker opinion of value?
AI can assemble the first pass: property facts, in-place income from the rent roll and operating statement, the comp table in your format, the valuation math in your workbook, and a narrative draft. The choice of comps, the cap rate range, and the recommended price are the broker's opinion and should be entered by the broker, with the workflow pausing to ask for each.
What is the difference between a BOV and an appraisal?
An appraisal is developed by a licensed or certified appraiser under the Uniform Standards of Professional Appraisal Practice. A BOV is a broker's estimate of likely price, usually prepared for an owner considering a sale or lease, and state law sets when a broker may give one and what statement it must carry. Confirm the current rule in your state with your managing broker or counsel.
Can I use CoStar or other licensed comp data with AI tools?
That depends on your license agreement, so read it or ask your account representative and your counsel before any licensed comp data goes into an outside AI tool. A workflow that draws only on your firm's own closed-deal records keeps licensed data out of it, while comps copied in from a subscription service remain subject to that license.
Does a BOV need a disclaimer that it is not an appraisal?
In some states, yes, and the wording is prescribed. Texas requires a specific statement reproduced verbatim in at least 12-point font, and North Carolina's statute requires a statement that the opinion is not an appraisal. Check your own state's rule and build the required language into the template as fixed text so every draft carries it by default.
What should a commercial BOV include?
Most include a property description, in-place income and expenses, a table of comparable sales or leases, one or more valuation approaches with a resulting range, a recommended price or pricing strategy, a marketing narrative, and any statement your state requires. Some states also prescribe contents, so compare your template against your state's rule.