How My Analysis AI values a company

Principles

• Data first. Numbers come from filings and licensed data providers. A language model is never used as a source of numbers.
• Deterministic calculations. Every formula is code with automated tests, so the same inputs always produce the same outputs.
• Ranges, not false precision. Each method produces a range, and the summary is a weighted range with the weights shown.
• Transparency. Each figure is labelled as Data, Calculated, Assumption, Your input or AI analysis.
• Honest gaps. When a figure cannot be retrieved it is shown as "Data unavailable", and methods that need it are left out with the reason.

Data and source hierarchy

For each figure, My Analysis AI uses the most authoritative source available and records which one was used:

1. Company filings: annual (10-K) and quarterly (10-Q) XBRL data from SEC EDGAR, as filed, with the filing date and accession number.
2. Official statistics: the 10-year U.S. Treasury par yield from the U.S. Department of the Treasury.
3. Licensed market data: prices, price history, exchange rates and reference data from Twelve Data, and other providers when configured.
4. Last good value: if a provider is unreachable, the last successfully retrieved value is shown with "Last updated N minutes ago".
5. Otherwise "Data unavailable". Nothing is estimated to fill the gap.

Validation checks run on every dataset: statements that do not reconcile, restated periods (the latest filed value wins), stock splits (share counts and per-share data are adjusted), currency mismatches between reporting and trading currency, stale financials, unusually large acquisitions or disposals, and earnings that are not backed by operating cash flow. Each problem appears as a warning next to the result. See Data sources for every provider and its limits.

Industry classification

The company is assigned to one of 19 business models using structured data, in this order: whether the security is a fund or ETF; the provider's industry; the provider's sector; the official SEC Standard Industrial Classification (SIC) code. A known sector never falls through to "General". The company's own numbers then refine the result: a software company with low growth and high margins is mature technology rather than high-growth SaaS, a fast-growing internet company with thin margins is treated as high growth, and a company with no meaningful revenue (under 10% of operating costs, with a loss) is routed to the pre-revenue path. When the classification systems disagree, the disagreement is shown. In advanced mode you can override the classification.

Sub-types. Within each business model a sub-type is assigned from the industry and SIC code and the company's own figures, because the same model can need different measures: commercial versus investment banks, insurance brokers, asset managers and market infrastructure; property and casualty, life, reinsurance and diversified insurers (an "insurer" whose premiums are under half of revenue is treated as a conglomerate); integrated oil, exploration and production, midstream, refining and services; digital advertising, hardware, semiconductors, mature software and high-growth SaaS; airlines, aerospace and defence and railroads; regulated utilities and power producers; commercial biotech, pharma, medical technology, managed care and clinical-stage biotech. Equity and mortgage REITs, conglomerates and funds are separate business models.

Classification confidence. High when the provider industry, SEC SIC code and the company's figures agree; Medium when one source is missing or they partly disagree; Low when the result rests on one weak signal; User when you overrode it. The reasons are listed with the result, and a Low classification lowers the Model Reliability measure.

Method selection

Each business model has a predefined set of primary and secondary methods, plus methods that are excluded with a reason. The set is then adjusted by the company's data: P/E is excluded when net income was negative in most reported years, EV/EBITDA when EBITDA was mostly negative, and DCF is demoted when free cash flow was mostly negative.

Every included method shows why it was chosen using the company's figures (for example, "free cash flow was positive in 5 of 5 years"), and every excluded method shows why it was left out. In advanced mode any method can be added or removed.

Historical and forecast periods

Periods are chosen automatically from the company's reported history. The reason is shown with the result, and you can change either period.

• History. Cyclical industries (oil and gas, mining, semiconductors) and companies whose revenue fell by more than 5% in at least two years use 10 years, to cover a full cycle. Companies growing over 20% a year, or that made a very large acquisition, use 3 years because older years describe a different business. Established companies use 5 years. With fewer than 3 years available, all years are used.
• Forecast. 10 years when growth is above 15% or free cash flow has mostly been negative, so growth can fade to a steady state before the terminal value; 7 years for growth between 7% and 15%; 5 years for mature, cyclical and financial companies. When analyst consensus is available, the forecast is extended (up to 10 years) so that at least three years follow the last consensus year and growth fades to the long-run rate instead of stepping down.
• Fiscal periods. Periods follow the company's own fiscal year. Only figures that had been filed by the valuation date are used, so a valuation dated in the past reproduces what was knowable then. LTM figures are built from the latest 10-Q year-to-date values.

Discounted cash flow

The default DCF ties growth to the capital needed to produce it (an alternative "accounting" mode forecasts EBITDA, CapEx and working capital separately):

EBIT = Revenue × operating margin
Taxes = tax rate × max(0, EBIT − loss carry-forward still available)
Reinvestment = max(ΔRevenue ÷ sales-to-capital, 0)
FCFF = EBIT − Taxes − Reinvestment

• Revenue growth uses analyst consensus where a connected provider supplies it, otherwise the historical growth shaped by the company's lifecycle stage, and then fades linearly to terminal growth.
• Operating margin starts at today's margin and moves part of the way back toward the company's own history when it is more than 5 points away; a loss-making company converges to a typical mature margin for its business type, labelled as the key assumption.
• Reinvestment is net of depreciation, so its floor of zero means maintenance spending only. When revenue falls, capital is not treated as released to shareholders.
• Tax. The effective rate moves to the marginal rate. Losses reported in the history become a loss carry-forward that shelters later profits. REITs pay no entity-level tax on distributed income.
• Leases. Operating lease liabilities count as debt when reported. When the provider does not supply them they are shown as unavailable with a warning, not as zero.
• Cost of capital moves from today's WACC to a mature-company WACC over the second half of the forecast.

Terminal value (perpetuity) = NOPAT(next year) × (1 − g ÷ terminal ROIC) ÷ (stable WACC − g)
Terminal value (exit multiple) = EBITDA(final year) × exit EV/EBITDA

Terminal growth defaults to 2.5%, capped 0.5 percentage points below the risk-free rate. The exit multiple defaults to the peer median EV/EBITDA, or the company's own 5-year median, and is kept between half and twice a typical mature multiple for the business type, because the exit year describes a mature company. Any value that had to be limited this way is listed as a warning. The model refuses to run when WACC is not above terminal growth, and warns when the terminal value exceeds 75% of enterprise value or implies an extreme multiple.

A DCF is not built for banks, insurers, REITs, funds or pre-revenue companies unless you add it under Methods, and no enterprise value is shown for banks, insurers or mortgage REITs.

Equity value = Enterprise value − Net debt − Lease liabilities − Minority interest − Preferred stock
Value per share = Equity value ÷ Diluted shares

Net debt is total debt (long-term including current portion, short-term borrowings and commercial paper) minus cash and short-term investments, from the latest balance sheet.

Discount rate (WACC)

Cost of equity = Risk-free rate + β × Equity risk premium (+ country and other premiums)
WACC = E/(D+E) × Cost of equity + D/(D+E) × Pre-tax cost of debt × (1 − tax rate)

• Risk-free rate: the latest 10-year U.S. Treasury par yield.
• Beta: a bottom-up beta from peers (unlevered, relevered at the company's debt-to-equity) when enough peers have price history. Otherwise the provider or regression beta is blended 50/50 with an industry beta for the business type, because one company's regression has a large estimation error. The result is kept between 0.6 and 2 (0.8–2 for financial companies); a beta outside that range lowers confidence.
• Equity risk premium: 5.0% unless the operator has set a dated implied premium. Advanced mode also computes an implied premium from the index's cash yield and expected growth, shows the cost of equity under both, and lets you adopt it.
• Cost of debt: the risk-free rate plus a default spread from a synthetic rating based on interest coverage, with the interest embedded in lease payments added to both operating profit and interest. Spreads come from a dated table (January 2026: AAA 0.40% to D 19.0%). Insurers and other financials use the lower coverage thresholds of the financial-company table. A company with more cash than debt whose coverage is low only because profits are small is not treated as near default. For banks and mortgage REITs borrowing is an operating input, so the figure is shown for reference and not used.
• Tax rate: the average effective tax rate over the window when it is between 5% and 40%, otherwise the statutory rate; zero for REITs.
• Weights: market value of equity and book value of debt, including lease liabilities.

Comparable companies

Peers are suggested from companies with the same business model (for REITs, the same property type), ranked by industry code and similarity in size, growth and margins, and the reason for each is shown. For a U.S. company, U.S.-listed peers come first; candidates in a different line of business, or more than five times larger or smaller, are marked secondary or dropped. A peer listed in another currency is converted at the current exchange rate, or excluded when no rate is available. You can add or remove peers.

Screening record. Every candidate considered is listed with the reason it was kept or left out: a different line of business, ranked below the top peers, a foreign or over-the-counter listing, a more distant industry code, revenue at least 20 times larger or smaller (never a core peer), or a very different capital intensity (capital spending as a share of revenue). Each kept peer shows its trading and reporting currency, the exchange rate used with its date and source, and the end date of the period its figures come from.

Negative enterprise value. A peer whose enterprise value is negative because it holds more cash than its market value plus debt keeps its price-based multiples (P/E, P/B) and loses only its EV multiples, with the reason shown. A negative enterprise value that net cash does not explain marks the whole peer as invalid.

Each peer multiple must pass validity gates before it counts: the peer must have a positive market value and enterprise value, the multiple must be positive, and it must lie within 0.2x to 5x of the peer median. Excluded cells are marked with the reason and left out of the statistics. For each multiple the minimum, 25th percentile, median, mean, 75th percentile and maximum are then calculated. A peer median P/E more than twice the long-run norm for the business type (a sign that peers are at trough earnings) is shown but not weighted; a normalised P/E (market value over average net income for up to five years, for peers and the company alike) takes its place.

Implied value applies the peer median (or another statistic you choose) to the company's metric for the same period (LTM, current fiscal year or a forward year), bridging enterprise-value multiples to equity value through net debt. The range uses the 25th to 75th percentile and needs at least four valid peers.

Historical multiples

The company's own P/E, EV/EBITDA, EV/Revenue, P/B, P/TBV and (for REITs) P/FFO are rebuilt month by month from price history and the financial statements that had been filed at each date. Revenue and net income roll forward each quarter (the last fiscal year plus the quarters filed since, minus the same quarters a year earlier), so P/E and EV/Revenue use trailing-twelve-month figures where quarterly filings exist; other metrics use the latest fiscal year. EV multiples are not shown for banks and insurers. The current multiple is compared with its 1, 3, 5 and 10-year medians. The implied range takes the 25th percentile, median and 75th percentile of one window (5 years by default, or all history when 5 years has fewer than 24 months) and applies them to the latest metric, so the low, central and high values come from the same distribution. Months with zero or negative enterprise value are left out.

Bank, insurance and dividend models

Justified P/B = (ROE − g) ÷ (Cost of equity − g)
Residual income value = Book value per share + Σ PV[(ROE − Cost of equity) × Book value]
Dividend discount value = Σ PV(dividends) + PV[Dividend × (1 + g) ÷ (Cost of equity − g)]

Stock-based compensation is treated as an operating cost: it is inside operating income and never added back to cash flow, and existing awards are in the diluted share count. Banks, insurers and mortgage REITs are valued on the equity side only: deposits and float are operating items, so enterprise value and WACC are not meaningful. Justified P/B is two-stage: book value grows at the same near-term rate the dividend model uses, then at the terminal rate, so the two equity models share one set of assumptions. Its range comes from the uncertainty in its inputs: cost of equity ±1 point and ROE ± one standard deviation of the company's own history. Excess returns in the residual income model fade at 20% a year after the forecast. Dividend growth is ROE × retention, except for REITs, where it follows the growth of AFFO (or FFO), capped at 6%; mortgage REITs assume no more than 2%. AFFO is only shown when a capital-spending deduction is reported. Bank capital ratios and non-performing loans are not in standard tagged filings and are shown as "Data unavailable" with the reason; insurance combined ratios are computed where the expense lines are tagged (see Industry analysis).

Industry analysis

Each industry gets the measures an analyst in that industry checks first. They explain whether the method values are credible; they do not produce a separate fair value. Measures that the connected data cannot supply are shown as "Data unavailable" with the reason.

• Banks. Net interest margin (net interest income ÷ total assets, a proxy because interest-earning assets are not tagged), efficiency ratio (non-interest expense ÷ revenue), cost of risk (provision ÷ loans), loan-to-deposit ratio, return on tangible equity and ROE minus the cost of equity. CET1 and non-performing loans are not in tagged filings.
• Insurers. Loss ratio = claims ÷ premiums earned; expense ratio = (policy acquisition costs + other underwriting expenses) ÷ premiums earned; combined ratio = loss ratio + expense ratio, computed only when an expense line is tagged so a loss ratio is never shown as a combined ratio; investment yield = net investment income ÷ average investments. The combined ratio is not shown for life insurers, whose profit comes from the investment spread.
• Mortgage REITs. Price ÷ book value, ROE minus the cost of equity, leverage (assets ÷ equity), dividend coverage (net income ÷ dividends) and payout. No fair value is published because book value moves daily with rates and spreads and is only reported quarterly.
• Oil and gas. EV ÷ current EBITDA beside EV ÷ mid-cycle EBITDA (average of up to seven years), and the share price's sensitivity to WTI crude (regression of monthly share returns on monthly WTI returns over up to five years, FRED series DCOILWTICO). Reserves and production are in the ASC 932 supplement, which is not connected.
• Airlines. EBITDAR = EBITDA + operating lease cost; lease-adjusted leverage = (debt + lease liabilities) ÷ EBITDAR; fuel as a share of revenue; the EBIT effect of a 10% fuel price rise before hedging; and the share price's sensitivity to U.S. Gulf Coast jet fuel (FRED DJFUELUSGULF).
• Utilities. Capital spending ÷ depreciation (rate-base growth), net debt ÷ EBITDA, payout and ROE minus the cost of equity. Rate base and allowed ROE come from regulatory orders, which are not connected.
• Technology. Stock-based compensation as a share of revenue, free cash flow after stock compensation, the Rule of 40, dilution, and for semiconductors the cycle-average margin.
• Development-stage biotech. A risk-adjusted NPV from pipeline assets you enter (phase, peak sales, launch year, margin, remaining development cost): rNPV = Σ probability of approval × after-tax cash flow ÷ (1 + cost of equity)^t, plus net cash. Default probabilities are the BIO, Informa and QLS 2011–2020 averages: Phase I 7.9%, Phase II 15.1%, Phase III 57.8%, filed 90.6%. With no assets entered the result is withheld; it is never blended into the published fair value because every input is yours.
• Holding companies. The sum-of-the-parts accepts an optional holding-company discount, applied to equity value as its own visible step. There is no default.

ETF analysis

An ETF's value is the market value of what it holds, so no intrinsic value, fair value or undervalued or overvalued label is calculated for a fund. ETFs open their own analysis under ETF analysis, in Quick and Advanced modes.

Classification. Each fund is identified from the provider's instrument type, the issuer profile and its name, and given an asset class (equity, fixed income, commodity, multi-asset), a primary label (equity, bond, gold, silver, oil futures, leveraged, inverse, multi-asset) and style tags (index, sector, international, dividend, growth, value, thematic, option income, ETN). Every tag lists the evidence it came from. Bond funds are also recognised from their holdings when the name does not say so. The structural category decides which measurements apply.

Price return = P_end ÷ P_start − 1            (split-adjusted closes)
Total return = A_end ÷ A_start − 1            (closes adjusted for splits and distributions)
Annualised (over 1 year) = (1 + R)^(1 ÷ years) − 1
Volatility = stdev(daily log total returns) × √252
Sharpe = (annualised return − average 3-month T-bill yield) ÷ volatility
Sortino = (annualised return − T-bill yield) ÷ downside deviation below the T-bill rate
Maximum drawdown = min over t of A_t ÷ max(A_≤t) − 1
Beta = slope of fund daily return on benchmark-fund daily return;  correlation = Pearson ρ
Tracking error = stdev(r_fund − r_benchmark) × √252;  tracking difference = annualised return gap
Distribution yield = distributions with ex-dates in the last 12 months ÷ price
Concentration: top-5/10/20 weight; HHI = Σ w²; effective number of holdings = 1 ÷ HHI
Holdings P/E = 1 ÷ Σ wᵢ·(Eᵢ ÷ Pᵢ)             (weights rescaled to the covered holdings)
Holdings P/B, EV/EBITDA: same harmonic form on B/P and EBITDA/EV;  ROE = aggregate E/P ÷ aggregate B/P
Leveraged fund: realised leverage = slope on index daily return;  drag ≈ ½ × L × (L − 1) × σ²
Bond fund duration = −slope of daily return on the daily change in the matching Treasury yield
Roll yield (annualised) = (F1 ÷ F2 − 1) × 12   (negative = contango, a cost)
Holdings overlap (A, B) = Σ min(wA, wB) over shared holdings

• Holdings-based valuation: company data for each fund's 25 largest holdings with a US ticker is aggregated with the harmonic (earnings-yield) average, so loss-making companies count instead of being dropped and the result is the fund's price over its share of the holdings' earnings. Coverage, the share of the fund the result represents, is always shown. Forward P/E and EV/EBITDA use companies with positive values only, with their own coverage. Country, company-size and industry exposure come from the same company data.
• Benchmarks and peers: index levels are licensed, so each fund is measured against the most liquid fund on its benchmark (for example SPY for the S&P 500, AGG for US aggregate bonds). Peers are curated groups of funds giving the same exposure (same index family, maturity band, sector or commodity); every peer figure is retrieved live.
• Quick-mode label: Higher Risk applies first (leveraged or inverse funds, exchange-traded notes, futures roll costs of 10% a year or more, three-year volatility of 40% or more or 1.5 times the peer median, or a top-10 weight of 65% or more). Otherwise the fund is scored against the peer median on cost (expense ratio), holdings valuation (weighted P/E, equity funds), risk-adjusted return (three-year Sharpe) and size and liquidity, each +1, 0 or −1 with the thresholds shown on the page. A total of +2 or more is Attractive Relative to Peers, −2 or less Expensive Relative to Peers, and anything between Reasonably Positioned. Fewer than three scored criteria (two for bond and commodity funds), less than a year of prices or fewer than two peers with data gives Insufficient Data. The label compares the fund with similar funds; it is not a forecast or a recommendation to buy or sell.
• Bond funds: measured duration against the 2, 5, 10 or 20-year Treasury yield (FRED) that matches the fund's maturity, with its 95% range and the price effect of ±50 and ±100 basis points, plus the market credit spread for corporate funds.
• Commodity funds: futures funds show curve shape and roll yield from the first four EIA contracts; physically backed gold shows the metal held, whose changes are the fund's investor flows.
• Leveraged and inverse funds: stated versus realised daily leverage, the expected volatility drag, and how far returns over rolling 1-month, 3-month and 1-year periods have differed from the multiple of the index.

NAV and premium or discount, bid-ask spreads, yield to maturity, SEC yield and credit-quality breakdowns are published by issuers but not supplied by any connected provider, so they are shown as DATA UNAVAILABLE, never estimated.

Scenarios and sensitivity

The bear and bull cases shift revenue growth by ±3 percentage points a year, EBITDA margin by ±2 points, WACC by ±1 point, terminal growth by ±0.5 points, CapEx by ±1 point and the exit multiple by ±2x. These are symmetric model assumptions, not forecasts, and no scenario is labelled as most likely. Every shift can be edited.

Sensitivity tables show value per share across WACC and terminal growth, WACC and exit multiple, and any other pair of drivers. The DCF range in the summary uses the inner cells of the WACC table (WACC ±0.5 points with terminal growth ±0.25 points, or exit multiple ±1x).

Weighting and the valuation range

Methods are grouped into four families of independent evidence: intrinsic (DCF, dividend discount, residual income, justified P/B), peer multiples, the company's own history, and asset value (NAV, sum-of-the-parts). Each family has a fixed share by business type, and a family with no valid method hands its share to the others:

Within a family, primary methods count 2, secondary 1 and supporting 0.5, with reliability cuts: a DCF whose terminal value exceeds 75% of enterprise value is halved, peer multiples with fewer than six valid peers and historical multiples with less than three years of monthly data are cut by a quarter. No single method may carry more than 35% of the total. A method is shown but gets no weight when its central value is zero or negative, lies outside its own range, or comes from a DCF whose terminal value is over 90% of enterprise value.

Weighted low = Σ weight × method low   (and likewise for central and high)

Publish gate. A fair value is published only when at least two families have weight and every family's central value is within ±35% of the median family value. Otherwise the result is Insufficient Evidence and the reason is shown. Every weight and its reason is shown, and you can replace any of them (your weights are rescaled to 100%). The weighted central value is what Quick mode shows as the estimated fair value. It summarises the methods shown; it is an estimate, not a price target.

Data quality grade

Each valuation is graded High, Medium or Low from objective checks: at least five years of reported statements, a recent quarter for LTM (the age of the latest annual report is shown), a current market price, a retrieved risk-free rate, reported share count and net debt, at least four valid peers with no more than a quarter of the selected peers excluded, and no high-severity model warnings. High means at most one check fails, and is not given when the latest annual report is over six months old with no quarter since, or when the company has no meaningful revenue. Low means a core input (statements, price, share count or net debt) is missing or fewer than three years are reported. Everything else is Medium. No confidence percentage is shown, because none could be calculated honestly.

Freshness. Every result states the basis of its figures (LTM to a quarter end, or a fiscal year), the period end, the filing date and when the data was retrieved. A latest period more than 4 months old caps the data grade at Medium and is a caution in the controls; more than 8 months old is a failed control.

Valuation confidence

Confidence (High, Medium or Low) says how far the estimate can be relied on, not whether the stock is attractive. Fixed rules add or subtract points: data quality grade, years of reported history, how predictable revenue has been, cyclicality (revenue fell by more than 5% in at least two years, or a cyclical industry), business maturity, how much of the DCF comes from the terminal value, the bull-to-bear scenario spread, the number of usable peers, how closely the methods agree, and input plausibility (a measured beta that had to be bounded, a cost of capital outside 6–14%, more than a quarter of peers excluded, or an exit multiple outside its sanity band). A score of 3 or more is High, 0 to 2 is Medium and below 0 is Low; a Low data grade always means Low confidence. Each factor and its effect is listed in Advanced mode.

Four separate measures. Confidence is also reported as four measures, so one strong measure cannot hide a weak one:

• Data quality: the data grade above.
• Model reliability: starts High. It drops one step for a Low classification confidence, one step for a terminal value above 75% of the DCF, one step for one implausible input and two for two or more (a beta outside 0.6–2.0, a WACC outside 6–14%, an exit multiple above 25x, a sales-to-capital ratio outside 0.2–5), and two steps when no primary method for the business type produced a valid value.
• Valuation uncertainty: Low when the published range is under 30% of the central value, Medium up to 60%, High above; one step higher when the bull-to-bear spread exceeds 100% or the business is cyclical.
• Method agreement: High when every family of methods is within 15% of the median family value, Medium within 35%, Low otherwise or with only one family.

The overall confidence can be no higher than the weakest of data quality, model reliability and method agreement, and no higher than Medium when uncertainty is High.

Valuation assessment, outlook and controls

The valuation assessment says how the estimated fair value compares with the market price. It is not a recommendation to buy or sell and is never converted into one. It is not chosen by AI: it is the rule below, applied to the weighted fair value, the market price and the confidence grade. Quick and Advanced read the same result, so they always agree.

Upside = fair value ÷ market price − 1
Margin of safety = (fair value − market price) ÷ fair value

Upside and margin of safety are different measures: a fair value of 120 against a price of 100 is 20% upside but a 16.7% margin of safety.

Quick mode labels. Quick mode shows the direction without the strength: Potentially Undervalued (either undervalued band), Approximately Fairly Valued, Potentially Overvalued (either overvalued band) or Insufficient Evidence. Advanced mode shows the full band.

Confidence cap. At low confidence, "Significantly" is shown as "Moderately", because the evidence does not support a strong reading. Confidence never strengthens an assessment.

Insufficient Evidence. No assessment is made when any of these holds: fewer than 3 years of reported financial statements; no market price, or a price more than 5 days old; no valuation method could be completed, or the fair value is zero or negative; a Low data quality grade; the methods disagree significantly (highest minus lowest method value exceeds +100% of their median); fewer than two families of methods agree within ±35% (the publish gate above); the security is a fund, has no meaningful revenue or is a holding company without segments; or the latest annual financials are more than 18 months old. Under Insufficient Evidence the fair value and upside are not published; the individual method results remain visible in Advanced mode. Confidence reads Insufficient Data when there is no reported period, price or fair value.

Each method. The methods table applies the same bands to each method's own value, before the confidence cap, so you can see which methods point which way.

Time horizon. The horizon is shown separately from the forecast: the explicit forecast years, and the fact that a fair value is not a price target for any particular date.

Financial health

Five indicators are rated Strong, Moderate, Weak or Insufficient Data with rules that depend on the business type:

• Revenue growth. Revenue CAGR over the reported years. Strong at or above the business type's threshold (8% generally; 4% utilities and telecoms; 5% REITs, banks and insurers; 6% consumer; 15% software) with the latest year not shrinking more than 5%; Weak below 0% (5% for software) or when the latest year shrank more than 5%; otherwise Moderate. Needs 3 reported years.
• Profitability. Non-financial companies: average ROIC minus WACC. Strong at +5pp or more, Weak below −2pp or with an operating loss, otherwise Moderate. Without ROIC: EBIT margin (Strong above 15%, Weak below 5%). Banks and insurers: average ROE minus cost of equity (Strong at +3pp or more, Weak below −2pp).
• Financial strength. Net debt ÷ EBITDA and EBIT ÷ interest expense. Strong with net cash, or leverage below the business type's limit (1.0x generally; 2.5x telecoms; 4x utilities; 5x REITs) and coverage of at least 8x. Weak above 3.5x (4x telecoms, 6x utilities, 7x REITs), coverage below 3x, or net debt with no positive EBITDA. Banks: CET1 ratio (Strong ≥ 12%, Weak < 10%) when filed, otherwise equity ÷ assets as a proxy (Strong ≥ 10%, Weak < 7%). Insurers: equity ÷ assets (Strong ≥ 20%, Weak < 10%), because solvency ratios are not in standard filings data.
• Cash flow quality. Free cash flow ÷ net income on average (conversion) and the share of years with positive free cash flow. Strong when FCF was positive in at least 80% of years and conversion is at least 80%; Weak when the latest FCF is negative and FCF was negative in most years, or conversion is below 40%; otherwise Moderate. Not assessed for banks and insurers, whose operating cash flow includes lending and claims flows.
• Competitive position. Evidence from reported results only (market share, pricing and customer data are not in the connected sources): share of years ROIC (ROE for financials) exceeded the cost of capital, gross-margin stability, and gross margin and revenue growth against the peer median. Each measure scores +1, 0 or −1. Strong at +2 or more, Weak at −1 or less, otherwise Moderate. Needs at least two measures.

Investment outlook

The outlook describes the direction of the business, not the share price, and is shown next to the valuation assessment, never combined with it. Each financial-health indicator scores Strong +1, Moderate 0, Weak −1; operating-margin, revenue-growth and leverage trends add +1 when improving and −1 when deteriorating. Positive at +2 or more, Negative at −2 or less, otherwise Neutral. Needs at least three measured indicators. The market price and the valuation are not used. A company can be undervalued with a negative outlook, or overvalued with a positive one; the page says which.

Mandatory controls

Every valuation runs these checks, each of which returns Pass, Caution, Fail or Not applicable with the numbers it compared: currency consistency, fiscal period alignment, stock splits and share continuity, diluted share count, debt and cash reconciliation, the DCF arithmetic recomputed step by step, the DCF recomputed by a second, separately written implementation (also available on the server from the Controls tab), CAPM and WACC recomputed, discount rate above terminal growth by at least 1pp and between 4% and 20%, realistic growth, terminal value share (caution above 75%, fail above 90%), no P/E or EV/EBITDA in the result when earnings or EBITDA are negative, missing inputs, stale data, and no look-ahead (only filings known at the valuation date). A result is labelled Verified only when every applicable check passes. Any failure is shown as Failed checks and removes the assessment; cautions are disclosed.

Configuration. The thresholds live in one configuration file and the administrator can change them without a code change by setting NEXT_PUBLIC_SENSE_SIGNAL_RULES to a JSON object, for example {"bands":{"strongBuy":0.4,"buy":0.15}}. Invalid settings are rejected and the defaults above are used. The rules in force are version signal-rules/1 (default settings); this page always shows the values currently in force.

Quick and Advanced modes

Both modes run the same valuation engine on the same data and assumptions. Advanced mode shows every method, input, source, period, weight, scenario and sensitivity table, and lets you change any assumption. Quick mode reads the same result and explains it in plain language: current price, estimated fair value (the weighted central value), upside or downside, the valuation assessment, confidence, the time horizon, five financial health indicators, a table of the primary methods and cross-checks with their weights, three opportunities and three risks with evidence, and a final assessment that shows the valuation assessment and the investment outlook separately. Advanced mode has ten sections (valuation overview, financial health, financial statements, valuation methods, DCF analysis, comparable companies, historical valuation, scenario and sensitivity, assumptions, and data sources and methodology) plus research and reports. The DCF section is hidden for banks, insurers, REITs, mortgage REITs and pre-revenue companies unless you add the DCF as a method. ETFs open the ETF analysis, which has its own Quick and Advanced modes. Quick mode calculates nothing of its own, and switching between the modes does not re-run anything. If you change an assumption in Advanced mode, Quick mode says so and offers to return to the automatic assumptions.

How AI is used

After the numbers are retrieved and calculated, an AI model writes explanations: what drives the valuation, which assumptions matter most, and the risks and catalysts visible in the data. The model receives only the retrieved and calculated figures, is instructed never to introduce numbers that are not in them, and its output is labelled AI analysis. It does not give buy or sell recommendations. If the AI service is unavailable, the valuation is unaffected.

Reproducibility

Each saved valuation stores its model version, valuation date, data timestamps and sources, the periods used, the classification and methods, the peers, every assumption with its label, your overrides and the weights. Because only filings known at the valuation date are used, a saved valuation can be reopened and reproduced. Exports to Excel include the formulas.

Backtesting

The engine includes a point-in-time backtest. At each test date it sees only statements filed by then (a period without a filing date counts as filed 90 days after it ends), prices up to that date, the Treasury yield on that date, no analyst consensus and no current provider beta. It records the fair value, the published range and the return over the following 12 months, and reports coverage, rank correlation between model upside and the later return, quintile spread, direction hit rate, mean absolute error and bias against the price 12 months later, how often that price fell inside the published range, and how much fair values move between dates. Results on small samples are labelled as not statistically meaningful.

Limitations

• Automatic statements come from SEC filings, so non-US companies that do not file with the SEC depend on the data provider plan for coverage.
• Funds, mortgage REITs, pre-revenue companies and holding companies without segment inputs do not receive a published fair value; ETFs get the ETF analysis and pre-revenue companies a pipeline model you fill in.
• Segment data, production volumes, reserves, regulatory capital ratios and company-defined metrics such as ARR are not in standard tagged filings.
• Analyst consensus and precedent transactions need a data plan or provider that supplies them; otherwise forecasts are model-generated and labelled as such.
• Default assumptions such as the equity risk premium and terminal growth are reasonable conventions, not facts; the sensitivity tables show how much they matter.
• Valuation is not a forecast of the share price. See the risk disclosure.