Industries / Construction and property / Real estate / Appraisal
Businesses
Business valuation is estimating what a company, a block of its shares or an intangible asset such as a brand or patent is worth, for taxes, estates, divorces, lawsuits, deals and company accounts.
5
Milestones achieved, from 1959 to 2017.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1959
IRS lists what to weigh
Revenue Ruling 59-60 told appraisers how to value shares in closely held companies, firms owned by a few people whose shares rarely trade, for US estate and gift tax. It defined fair market value as the price between a willing buyer and a willing seller who both know the relevant facts, listed eight factors to consider, among them earning capacity, dividend-paying capacity, goodwill and the prices of similar listed companies, and said that "No formula can be devised that will be generally applicable".
LessonThe ruling says outright that "valuation is not an exact science", so the appraiser's judgment has to be explained.
Credit
US Internal Revenue Service
1973
Options get a formula
Fischer Black and Myron Scholes published a formula for the value of an option, a right to buy or sell something later at a fixed price, which Robert Merton generalized; Merton and Scholes received the 1997 economics prize for it. US accounting rules later named the "Black-Scholes-Merton formula" among the accepted ways to value an asset from its expected income.
Credit
Fischer Black, Myron Scholes and Robert Merton
1983
Courts admit modern methods
In an appraisal case, shareholders who reject a merger ask a court to set the fair value of their shares. In Weinberger v. UOP, decided on 1 February 1983, Delaware's Supreme Court called the old "Delaware block" method, which added up fixed weights of asset value, market price and earnings, "clearly outmoded", and allowed "any techniques or methods which are generally considered acceptable in the financial community". The lower court had rejected a discounted cash flow analysis, and the ruling opened Delaware's courts to it.
Credit
Delaware Supreme Court
2001
Buyers must price intangibles
US accounting Statements 141 and 142, issued by the Financial Accounting Standards Board (FASB) in 2001, ended "pooling", which let merging firms simply add their books together. Every deal became a purchase, in which the buyer must list identifiable intangible assets apart from goodwill, the premium left over, and goodwill is no longer written down a little each year but tested for impairment, a fall in value. The rules made purchase-price allocation, splitting a deal's price among the assets acquired, routine work for appraisers.
Credit
FASB
2017
Deal price beats the model
In DFC Global (1 August 2017) and Dell (14 December 2017), Delaware's Supreme Court reversed trial rulings that had leaned on experts' cash flow models; in Dell, the trial court had set $17.62 a share against a $13.75 deal price. In Dell the court said the deal price deserved "heavy, if not overriding, probative value"; in DFC it had stated that "there is no presumption in favor of the deal price". Hedge funds had been buying shares after deals were announced in order to sue for more, a strategy called appraisal arbitrage; Cornerstone Research counts appraisal petitions in Delaware falling from 76 in 2016 to 26 in 2018.
LessonAn expert's model that lands far from the price an open sale produced needs strong evidence behind it.
Credit
Delaware Supreme Court
Watch for
Claude builds cash flow models
Anthropic released ready-made skills with which its Claude models build discounted cash flow models, with free cash flow forecasts, cost-of-capital calculations, scenarios and sensitivity tables, along with comparable-company analyses, and opened a test version of Claude for Excel to 1,000 users. It cited 55.3% accuracy for Claude Sonnet 4.5 on Vals AI's Finance Agent benchmark. An outside test published in April 2026, BankerToolBench, built with 502 investment bankers, found that the bankers rated 0% of the best model's work ready for a client.
Achieved whenAn outside test scores AI-built valuations against practitioners' work or later market prices, and they pass at a stated rate.
Claimed by
Anthropic, October 2025
Sixty-second small-business values
Unbroker says its ValuBot produced estimates for "over $13 billion" of small businesses in nine months, based on more than 10,000 real sales, with a fuller report costing $745.
Achieved whenSomeone outside Unbroker compares its estimates with the prices the businesses later sold for and publishes the error.
Claimed by
Unbroker, as reported by Business Valuation Resources, August 2026
A shared performance framework
The American Institute of CPAs (AICPA), the American Society of Appraisers (ASA) and RICS released a draft Valuation Performance Framework, which builds on the earlier Mandatory Performance Framework, and are taking comments until 22 December 2026.
Achieved whenThe final framework is issued and valuers or auditors are required to follow it.
Claimed by
AICPA, ASA and RICS, October 2026
Open
01
Sealed pre-deal valuation test
A public test in which AI agents value companies that were later bought for cash. Each agent sees only filings and share prices from before the deal was announced, and locks a value per share and a range, which are hashed and time-stamped. The deal price is revealed only afterward.
The briefCredit · unclaimed02
Fairness opinions, rebuilt
A fairness opinion is a banker's letter telling a company's board that a deal price is fair, and the merger proxy sent to shareholders summarizes the banker's analysis. Build an extractor that pulls each banker's discounted cash flow inputs from merger proxies, such as the forecast years, the discount-rate range, the long-run growth rate or exit multiple and the implied value range per share, and a calculator that rebuilds the range from those inputs and flags gaps.
The briefCredit · unclaimed03
An open marketability discount dataset
A discount for lack of marketability (DLOM) is the cut in value for shares that cannot be sold quickly. Build a free dataset, one row per deal, of public companies selling unregistered shares, which buyers cannot resell at once: the price paid, the market price before the sale, the limits on resale and the implied discount, each linked to its filing.
The briefCredit · unclaimed
Open 01 · the brief
Sealed pre-deal valuation test
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test in which AI agents value companies that were later bought for cash. Each agent sees only filings and share prices from before the deal was announced, and locks a value per share and a range, which are hashed and time-stamped. The deal price is revealed only afterward.
Why it is still open
Current tests of finance agents grade their work against bankers' scoring rules, not against what a buyer actually paid; in BankerToolBench, bankers rated none of the best model's outputs ready for a client.
Built with what exists
SEC EDGAR full-text search and its financial-data interfaces; Aswath Damodaran's free cost-of-capital and industry multiples data at NYU Stern, last fully updated in January 2026; agents built on models such as Claude Fable 5.2 (see the Note "Three working devices"), each case run many times (see the Note "All twenty times"); deals announced after each model's training cutoff, or with names and dates masked, so no model can recall the answer.
First step
List the cash mergers of 2026 from EDGAR merger-proxy filings and build 20 frozen data packets.
Done looks like
At least 100 deals and a public table of each agent's median absolute percentage error against the deal price, beside two baselines, the share price before the announcement and the midpoint of the bankers' discounted cash flow range in the merger proxy, with each case run 20 times to show the spread.
Open 02 · the brief
Fairness opinions, rebuilt
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A fairness opinion is a banker's letter telling a company's board that a deal price is fair, and the merger proxy sent to shareholders summarizes the banker's analysis. Build an extractor that pulls each banker's discounted cash flow inputs from merger proxies, such as the forecast years, the discount-rate range, the long-run growth rate or exit multiple and the implied value range per share, and a calculator that rebuilds the range from those inputs and flags gaps.
Why it is still open
A 2024 academic study of fairness opinions still worked from "hand-collected" valuations for mergers completed from 2000 to 2015.
Built with what exists
EDGAR full-text search; a small extraction model for long filings (see the Note "The jobs agents repeat"); a second model as reviewer (see the Note "Seven more flagged"); Python.
First step
Hand-code 20 recent merger proxies to fix the fields and the answer key.
Done looks like
200 proxies extracted, at least 95% of fields matching on a 50-proxy set coded by hand, and, for each opinion, the gap in percent between the rebuilt midpoint and the banker's, with page references for every mismatch.
Open 03 · the brief
An open marketability discount dataset
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A discount for lack of marketability (DLOM) is the cut in value for shares that cannot be sold quickly. Build a free dataset, one row per deal, of public companies selling unregistered shares, which buyers cannot resell at once: the price paid, the market price before the sale, the limits on resale and the implied discount, each linked to its filing.
Why it is still open
The reference set many appraisers cite, the Stout Restricted Stock Study, is a paid database of more than 780 transactions, and one use of its discount calculator costs $369.
Built with what exists
Form 8-K Item 3.02 ("Unregistered Sales of Equity Securities") filings, found with EDGAR full-text search, and the purchase agreements filed with them; a licensed source of daily share prices; Python.
First step
Pull the Item 3.02 filings for 2025 and code 50 by hand.
Done looks like
At least 500 transactions, each linked to its filing, with a second person re-checking a random 10% and the agreement rate published.