How it works

A valuation here is a story about a business, turned into a few numbers, run through a discounted cash flow. The story is the judgment; the arithmetic is code. Every number carries its source and its date.

Valuing ADP, start to finish

Every number from a filing, dated and checked.

AI Pulling ADP's filings, its industry's data, the Treasury rate and the price.

AI ADP holds about $37-40 billion of client money between collecting and paying wages and taxes. Is any of it counted as ADP's own?

AI Checked against the balance sheet: only $195.2 million of pledged securities is added, about $0.49 a share.

Source: 10-K, fiscal 2026 (SEC EDGAR 0000008670-26-000030)

Source: 10-Q to 2026-09-30 (SEC EDGAR 0000008670-26-000044)

Source: Industry data, Damodaran Online, January 2026

Source: Equity risk premium, Damodaran Online, October 2026

Source: US 10-year Treasury (FRED), 2026-10-29

Source: Price $305.44, close of 2026-11-10 (Qualuation's records)

The AI reads the filings and drafts the story.

AI Interest on client funds was $1,355 million in fiscal 2026, about 6% of revenue and nearly all of it profit: interest rates move ADP's margin, and payroll volumes move with employment.

AI Two stories, drafted at 20:24: “ADP keeps doing what it does” and “The worry comes true”.

Source: 10-K: business and results

Every story gets challenged.

Reviewer A second AI session reads both stories and the filings, and asks questions: the latest quarter's margin gain, PEO pass-throughs, debt maturity.

Source: The reviewer's questions, answered in the revised cases

The stories are yours to change.

You Story 1: after year 10, new investment earns the industry's 28.3%.

You Story 2: long-run growth 3%; the margin still ends at 24%, a severe ending.

AI Revised at 20:36, each change with its reason.

Source: Revised cases: every change, what it was, what it is, and why

You decide. Nothing is valued until you approve.

You “approved”

Source: Approval recorded at 20:37 UTC

The AI never does the math.

Engine Story 1: $324.12 a share. Story 2: $152.69 a share. The price: $305.44.

Engine The growth the price needs in years 2 to 5: 4.7560%, against a riskfree rate of 4.7614% and a record of 6.68%.

Source: Engine run: the solver inside its bracket of -50% to 100%

Weak spots are written down, not hidden.

AI A second look at the long-run cost of capital, saved at 20:42: at 8.46% the price needs 4.76% growth; held at 7.94%, it needs 2.32%. Added to the weak spots.

Source: Report, second save

AI At today's price, ADP grows from year 2 at about the rate it grows forever.

Condensed from a real valuation session on November 11, 2026. Figures as recorded. The post it became.

1. The data

Company filings from SEC EDGAR: annual and quarterly reports and their tagged figures, adjusted the way a valuation needs them, with R&D and leases treated as investments, options and restricted stock counted, and revenue split by business and geography where the filings allow. Industry data, equity risk premiums, country risk and rating tables come from Damodaran Online, each by the edition published by the valuation's date. The riskfree rate is the US 10-year Treasury yield from FRED. Nothing from after the valuation date is used.

2. Where every number comes from

Much of the financial data online is copied, rounded or simply made up, and AI models repeat it. Here, every figure comes through tested tools from a named source with a date: a filing and its accession number, a dataset and its edition, a price and its close. The AI never supplies a number from memory or the web. Where a number is a judgment rather than a fact, such as a driver chosen for a story or the AI's own estimate of a market's size, it is marked as one, with its reason.

3. The method

Every valuation starts as a story: what the business is, how it grows, what it earns, what it needs to reinvest, how risky it is. The story becomes a few drivers (revenue growth, operating margin, sales to capital, cost of capital), and a discounted cash flow turns them into a value per share. There are always two stories, and the question is what today's price assumes: which story it sits closer to, and what one driver would have to be for the price to hold.

4. The AI's role, plainly

It reads the filings, drafts the stories and finds precedents in Damodaran's published writing.
It never does the math. Every number comes from tested code: the valuation engine, the data layer, the solver.
A person approves the story and the drivers before anything is valued. The stories are theirs to change.
A second AI session challenges the stories, in either direction, before approval.

5. How it's tested

The engine reproduces Aswath Damodaran's published valuation spreadsheets line by line, every year of every case within a millionth. The AI's own judgment is measured against 14 of his published valuations, at their dates, on the data available then; the gap is reported on Under the hood, not hidden.

Run your own story See what prices assume